# The Economic Impact of Spacecraft Down-Mass and Orbital Reentry Operations on the U.S. National Airspace System

**A Doctoral Dissertation**

**Candidate:** DOWNMASS-01
**Program:** COLLEGIUM 1st Battalion
**NORTH STAR / JPL category:** Space Infrastructure Systems
**Methodological anchors (Hall of Shoulders):** Akhil Rao (orbital-economy externalities); Joshua Angrist and Jorn-Steffen Pischke (research design and identification); Douglass North (institutions and transaction costs)
**Date:** 2026-06-15


## Abstract

The National Airspace System is a shared public trust, and the task this dissertation takes up is to measure, honestly and in full, what the return of spacecraft through that airspace asks of the aircraft and the travelers who depend on it. Commercial space launch and reentry operations are integrated into the U.S. National Airspace System (NAS) today by closing large, static volumes of airspace around the planned trajectory. Each closure forces nearby flights to delay, hold, or reroute, and each response consumes fuel, extends flight time, and raises direct operating cost. As reentry frequency rises with large-constellation deorbit and the emergence of a commercial down-mass sector, the cumulative cost of these closures is expected to scale faster than the safety benefit the closures provide. This dissertation states and is designed to test one falsifiable contribution: spacecraft reentry and down-mass operations impose a measurable disruption cost on the NAS, expressed in delay, reroute, and closure, and improved reentry prediction combined with dynamic airspace management yields a quantifiable avoided cost. The null hypothesis is that reentry operations impose no measurable NAS cost.

The study specifies an event-study and staggered difference-in-differences design. The treatment is a reentry-driven Aircraft Hazard Area (AHA) activation over a defined airspace sector and time window. The outcome is the change in realized flight-level cost (delay minutes, added distance, fuel burn, and direct operating cost) for flights exposed to that sector relative to comparable unexposed flights. Four named data layers are joined: the EU Space Surveillance and Tracking (EU SST) reentry catalog (treatment timing, location, and prediction-uncertainty bounds), FAA NAS and System Wide Information Management (SWIM) operations data (flight tracks, delay, and reroute records), the four completed IAC-26 down-mass PRISMA systematic reviews (the parameterized prior base), and the ReentryFlow reentry-to-airspace economic model (which maps a reentry trajectory and its dispersion to an affected-flight set and a cost, and simulates the dynamic-closure counterfactual). The primary estimator is Callaway and Sant'Anna, with the Goodman-Bacon decomposition and the de Chaisemartin and D'Haultfoeuille robust estimator as diagnostics for staggered, heterogeneous treatment timing.

The dissertation is presented at the design stage. The estimating equations, the variable construction, and the threats to validity are specified in full. No estimate has been executed on the assembled panel, which does not yet exist; every number that appears is labeled expected or illustrative and is not presented as an executed estimate. The contribution is the measurable-cost claim and the avoided-cost claim, each stated so that a near-zero, statistically insignificant treatment effect would falsify it. The value to NASA, JPL, and the civil-space enterprise is the same whether the measured effect is large or small: a defensible, design-based price for the externality that reentry imposes on aviation, and for the cost that better prediction avoids, is the evidence base on which any reentry-authorization or corridor-pricing institution must rest.


## Table of Contents

- Abstract
- Chapter 1: Introduction
- Chapter 2: Theoretical Framework
- Chapter 3: Literature Review
- Chapter 4: Data and Measurement
- Chapter 5: Research Design and Identification
- Chapter 6: Analysis Plan and Expected Results
- Chapter 7: Discussion
- Chapter 8: Conclusion
- References
- Appendix A: Variable and Data Dictionary
- Appendix B: Derivations
- Appendix C: Instrument and Query Details
- Appendix D: Supplementary Tables (Specified-but-Unpopulated Templates)

**List of Tables.** Table 3.1 (Theme A synthesis); Table 3.2 (Theme B synthesis); Table 3.3 (Theme C synthesis); Table 3.4 (Theme D synthesis); Table 4.1 (Variable operationalization); Table A.1 (Variable and data dictionary); Tables D.1-D.4 (Specified-but-unpopulated result-table templates).

**List of Figures.** None. The event-study profile of Section 6.7 and the cost-surface representation of Section 6.5 are specified at the design stage and are reported as templates rather than rendered figures, consistent with the design-stage posture.


# Chapter 1: Introduction

## 1.1 The answer this chapter defends

The airspace above us is a commons held in common, kept safe by careful stewardship and shared by every operator and traveler who passes through it; the purpose of this work is to give those who steward it a defensible measure of what spacecraft reentry asks of that shared resource. This dissertation is built to establish one finding and to make that finding falsifiable. The finding is that spacecraft reentry and down-mass operations impose a measurable, positive cost on the United States National Airspace System (NAS), expressed in delay, reroute, and closure, and that improved reentry prediction combined with dynamic airspace management yields a quantifiable avoided cost. The finding matters because the policy now in force, closing a large and static volume of airspace around each predicted reentry trajectory, was calibrated to an era of rare events, and reentry cadence is no longer rare. Large constellations are beginning to deorbit at scale and a commercial down-mass sector is forming, so a rule that withholds a worst-case volume per event becomes a structural drag on the NAS rather than an occasional contingency. The action that the finding enables is concrete: a defensible price for the externality that reentry imposes on aviation, and a target avoided cost against which a prediction-and-modeling investment can be judged worth funding.

I state the answer first, before the background, because the chapter is organized as an argument rather than as a survey. Everything that follows in this chapter, the problem in full, the institutional and historical context, the research questions, the significance to NASA and the Jet Propulsion Laboratory (JPL) and the named stakeholders, the scope, the definitions, and the roadmap, is in service of that single claim and of the null hypothesis that would defeat it. The claim is presented honestly at the design stage. No estimate in this dissertation has been executed on the full dataset. The estimating equations, the variable construction, the data sources, and the threats to validity are specified in full, and where a number appears it is labeled as an expected range drawn from prior literature or as an illustrative value used to demonstrate a procedure. The contribution is the design and the falsifiable pair of claims it operationalizes, not a set of executed coefficients.

The chapter proceeds by framing the problem in the standard form that governs the whole dissertation. The current state is integration by large static closures. The desired state is a prediction-informed dynamic closure rule priced against a measured aviation cost. The gap is that no study estimates, with a credible causal design, the marginal NAS cost of a reentry event or the cost that better prediction would avoid. Leaving the gap open carries a consequence: as cadence rises, conservative static segregation scales aviation cost faster than it scales safety benefit, and the regulator has no measured price with which to weigh a narrower closure's residual risk against its cost saving. The rest of the chapter develops each element of that frame and then converts it into testable hypotheses.

## 1.2 The problem in full

### 1.2.1 A one-decade shift from episodic to high-frequency commercial space

Within a single decade the commercial space sector moved from an episodic, government-dominated enterprise to a high-frequency commercial one. The change is most visible in launch economics. Reusable launch vehicles reduced the cost of reaching low Earth orbit by roughly a factor of twenty, from approximately 54,500 United States dollars per kilogram in the Space Shuttle era to approximately 2,720 United States dollars per kilogram with Falcon 9, expressed in 2017 dollars [\[25\]](#ref-25), [\[26\]](#ref-26). That collapse in unit cost is not merely a price story; it is a frequency story. When the marginal cost of placing mass in orbit falls by an order of magnitude, the number of launches, the number of objects on orbit, and, with a lag set by orbital decay, the number of objects returning through the atmosphere all rise together. The aviation-facing consequence of cheaper launch is therefore not confined to the ascent phase. It propagates downstream, to the descent phase, where the returning object must traverse occupied airspace.

The descent-phase consequence is already documented as a real, not hypothetical, burden on aviation. Hook and colleagues record that the number of successful rocket launches more than doubled between 2015 and 2023, from 87 to 212, and that of the 212 launches in 2023, 128 rocket bodies were abandoned in orbit and left to reenter uncontrollably [\[38\]](#ref-38). They further record a concrete instance of the airspace consequence: a reentering Long March 5B rocket body in November 2022 caused the closure of airspace over Europe, delaying 645 flights, with a plausible economic impact measured in millions of euros [\[38\]](#ref-38). Wright and colleagues frame airspace closures caused by reentering space objects explicitly as a forward-looking and growing burden tied to constellation deorbit rates rather than as a rare contingency [\[1\]](#ref-1). Pardini, examining the risk on the ground and in the airspace posed by uncontrolled reentries, asks directly whether the growth observed in recent years should be considered worrying and treats the upward trend as the central object of concern [\[64\]](#ref-64). These sources converge on the empirical foundation for the first element of the problem frame: the closures are real, they touch real flights, and their frequency is increasing.

### 1.2.2 The down-mass sector and why the return leg is integral

The binding constraint on the next stage of growth has shifted downstream of propulsion. The cost of reaching orbit, once the dominant barrier, is no longer the variable that paces the emergence of the activities this dissertation is concerned with. For the down-mass sector, defined here as the commercial return to Earth of objects mined, refined, or manufactured in space, the value of the returned product depends on delivering it through occupied airspace to a point near its insertion into a terrestrial supply chain. The return leg is therefore integral to the activity rather than incidental to it. A space-manufactured pharmaceutical, a refined material, or a recovered resource has no realized commercial value until it clears the airspace between the upper atmosphere and its terrestrial buyer. The literature on space resource utilization makes the customer-location dependence explicit. Vergaaij, McInnes, and Ceriotti compare material sources and customer locations for commercial space resource utilization and show that the economics of a space-resource activity depend on where the material must be delivered, which for a terrestrial customer means delivery through the atmosphere and the airspace beneath it [\[20\]](#ref-20). Work on orbital facilities and in-space operations similarly treats the efficient movement of mass to and from orbit as the operative economic variable rather than the act of reaching orbit itself [\[87\]](#ref-87).

The down-mass case sharpens the airspace problem because reentry is not launch played in reverse. Reentry trajectories carry greater inherent prediction uncertainty than ascent, disperse hazard over larger geographic areas, and intersect aircraft cruise altitudes differently [\[4\]](#ref-4), [\[6\]](#ref-6). A launch is a powered, instrumented, actively guided event whose trajectory is known and controlled from ignition; an uncontrolled reentry is a ballistic event whose timing and footprint are estimated, not commanded. The asymmetry means that the airspace tools developed for launch, while a sound starting point, cannot simply be applied to reentry without adaptation, and it means that the airspace cost of a reentry event is governed by a different and larger uncertainty than the airspace cost of a launch. This asymmetry is the hinge of the dissertation, because the prediction uncertainty of reentry is exactly what the second hypothesis proposes to attack.

### 1.2.3 The mechanism that converts reentry into an economic problem for aviation
The mechanism by which a reentry event becomes an aviation cost is well documented, and I state it here once, plainly, as the causal chain the whole dissertation tests. The driver is a reentry event with large prediction uncertainty. The mechanism is a conservative static Aircraft Hazard Area (AHA) sized to a worst-case dispersion envelope and fixed in advance. The observable effect is a large withheld airspace volume over a sector and time window. The operational consequence is that exposed flights delay, hold, or reroute, which produces added delay minutes, added flown distance, additional fuel burn, and additional direct operating cost; and because the NAS is a network, the cost propagates beyond the closed window as delay cascades downstream. The strategic implication is that this is an unpriced externality on aviation that scales with cadence, and measuring it is the precondition for a lower-transaction-cost authorization institution and for any corridor-pricing instrument.

The Federal Aviation Administration (FAA) manages launch and reentry integration chiefly by closing these large, predetermined hazard areas around the planned trajectory, for windows historically measured in hours [\[2\]](#ref-2), [\[4\]](#ref-4). Static segregation is conservative by design, and it has preserved a strong safety record. The cost of that conservatism scales with the size and duration of the withheld volume and with the density of the traffic that would otherwise have used it. At one event per month the cost is small. At one event per week or per day the same policy becomes a structural drag on the NAS. This follows from a simple proportionality between withheld volume, displaced traffic, and cost, evidenced in the documented closure events and their measured consequences [\[38\]](#ref-38), [\[1\]](#ref-1) and in the launch-era simulation literature that translated closure geometries into flight-level penalties [\[2\]](#ref-2), [\[3\]](#ref-3). The precise per-event cost is not yet measured causally, and that is the gap this dissertation occupies. One objection should be registered so that the claim is not overstated: operators might fully anticipate and pre-absorb closures so that no marginal cost appears at activation. The research design confronts this directly by inspecting pre-activation effects.

I keep correlation and causation distinct in stating this mechanism. The launch-era literature establishes a strong, consistently signed association between closures and flight penalties, but it does so descriptively and by simulation, not by a credible causal design. The present dissertation supplies the identification that converts the association into a causal estimate. Until that estimate is executed, the confidence attached to the magnitude of the per-event cost is moderate, grounded in convergent prior evidence on direction but not yet on a measured causal coefficient.

## 1.3 Institutional and historical context

### 1.3.1 The launch-cost collapse and the migration of the binding constraint

The institutional history that produced this problem is short and unusually well documented. The defining event is the reduction in launch cost already cited, from roughly 54,500 to roughly 2,720 United States dollars per kilogram across the transition from the Shuttle to reusable commercial vehicles [\[25\]](#ref-25), [\[26\]](#ref-26). Jones documents the recent large reduction in space launch cost and attributes it principally to reusability and to the displacement of cost-plus government procurement by competitive commercial provision [\[25\]](#ref-25). Kang and colleagues examine the cost effectiveness of reusable launch vehicles as a function of payload capacity and confirm that the favorable economics depend on flight frequency, because reuse amortizes fixed costs across many flights [\[26\]](#ref-26). The two findings together explain why cheaper launch and higher cadence are the same phenomenon rather than two separate ones, and why the airspace consequence of the launch-cost collapse is a frequency consequence.

The migration of the binding constraint is the institutional core of the problem. When launch was expensive and rare, the cost of reaching orbit dominated every other consideration, and the airspace cost of the occasional return was negligible by comparison. As launch cost fell and cadence rose, the relative weight of the downstream constraints grew. For activities whose value is realized on the ground, the return leg, and the airspace it must traverse, became the operative economic margin. This is not a claim that propulsion ceased to matter; it is a claim that the marginal economic decision moved downstream, from how to reach orbit cheaply to how to return mass through occupied airspace without imposing a cost that scales away the value of the activity. The space-economy literature frames this transition in exactly these terms, treating market formation as governed by institutional design and the pricing of externalities rather than by technology readiness alone [\[16\]](#ref-16), [\[17\]](#ref-17), [\[18\]](#ref-18).

### 1.3.2 The airspace institution and why it is the right unit of analysis

The institution that governs reentry through shared airspace is the authorization and segregation regime administered by the FAA, supported by the data systems that feed it. That regime is the right unit of analysis because the NAS is shared national infrastructure that commercial reentry must use, and because the question at issue is one of system-level capacity and cost rather than of vehicle design. The FAA has begun building the data infrastructure that a more responsive regime would require. The Space Data Integrator platform automates previously manual, time-consuming, and resource-intensive procedures for supporting commercial launch and reentry operations, and it provides situational awareness on vehicle position and mission parameters during operations [\[29\]](#ref-29). The existence and direction of that investment is itself evidence that the institution recognizes the problem and is moving, however incrementally, from static toward more dynamic handling.

The historical pattern in adjacent domains gives the institutional framing its weight. North's account of institutions distinguishes the rules of the game from the players and holds that institutions exist to lower the transaction costs of impersonal exchange [\[24\]](#ref-24). Applied to airspace, static case-by-case AHA closure is a high-transaction-cost rule: every event is adjudicated conservatively and the cost falls on aviation users who are not party to the authorization decision. A repeatable, prediction-informed, dynamic-closure rule would be a lower-transaction-cost institution, but moving to it requires a measured price for the disruption that the current rule imposes, because without that price the regulator cannot weigh a narrower closure's residual risk against its cost saving. This is the institutional reason the measurement at the center of this dissertation is load-bearing rather than academic. The historical context, in short, is an institution built for a low-frequency world now confronting a high-frequency one, equipped with improving data systems but lacking the priced externality that would let it change the rule defensibly.

## 1.4 The research questions, broken out explicitly

The dissertation is organized around one research question with two estimable parameters. I state the question, then decompose it, then state precisely what each parameter is and what value of it would speak for or against the contribution.

**Research question.** Do spacecraft reentry and down-mass operations impose a measurable disruption cost on the United States National Airspace System, and does improved reentry prediction combined with dynamic airspace management yield a quantifiable avoided cost?

The question has two clauses joined by an "and," and each clause corresponds to one estimable parameter. The first clause asks whether the cost exists and is measurable; the second asks whether better prediction plus dynamic management reduces it by a measurable amount. The design is built so that each clause can be answered independently, because it is possible in principle for the first to be confirmed and the second refuted, or for both to fail.

**Parameter one, the disruption parameter.** This is the per-event NAS cost. Formally it is the aggregated average treatment effect on the treated (ATT) of a reentry-driven AHA closure on the cost outcomes borne by exposed flights. It answers the first clause of the research question. If it is positive and statistically distinguishable from zero, reentry imposes a measurable cost; if it is near zero and precisely estimated, it does not.

**Parameter two, the avoided-cost parameter.** This is the difference in the per-event cost between a static-closure regime and a prediction-informed dynamic-closure regime that narrows the hazard volume in time and space as the reentry prediction tightens. It answers the second clause of the research question. If it is positive and distinguishable from zero, better prediction plus dynamic closure pays; if it is near zero, it does not, and a move from static to dynamic closure buys nothing on cost.

The two parameters are distinct quantities, and the dissertation reports them separately. The disruption parameter is estimated from the realized record of reentry closures and their flight-level consequences. The avoided-cost parameter is estimated by comparing each realized static closure against a counterfactual dynamic closure generated by the ReentryFlow model and, separately, by exploiting natural variation in closure tightness across events that differ in prediction uncertainty, using prediction uncertainty as a continuous treatment intensity. The mechanism behind the second parameter is the same chain as the first, acted upon at a different point: prediction uncertainty falls sharply as the object nears reentry, so a dynamic closure sized to the late-available uncertainty withholds a smaller volume and displaces fewer flights [\[21\]](#ref-21), [\[22\]](#ref-22). The expected sign of both parameters is positive, but the design does not assume the sign; it is built to detect a near-zero effect and to treat such a finding as a falsification rather than a failure.

## 1.5 Significance for NASA, JPL, and the named stakeholders

### 1.5.1 Why this is a Space Infrastructure Systems question

This work sits in the Space Infrastructure Systems category because the NAS is shared national infrastructure that commercial reentry must use, and because the question is one of system-level capacity and cost rather than vehicle design. I make that infrastructure framing in plain economic language and do not import an architecture-modeling vocabulary, because the contribution is an econometric measurement of an externality and an avoided cost, not the design of a system or capability. The NAS is the shared resource; the reentry operator is the party that consumes part of it without bearing the resulting aviation cost; the measurement supplies the price of that consumption. That is the entire architectural claim, and it is stated in ordinary terms because the dissertation is about pricing a congestible common resource, not about specifying its engineering.

### 1.5.2 The NASA Langley and Analytical Mechanics Associates line of work

The significance to NASA is direct and concrete. NASA Langley Research Center and Analytical Mechanics Associates maintain an active line of work on airspace modeling for launch and reentry authorization, and the joint IAC-26 study from which this dissertation draws its parameterized evidence base argues that government can act as an economic enabler of the down-mass sector by adapting launch-era airspace capabilities to reentry rather than by building new infrastructure. The architectural direction of the field supports the enabling premise. Hilton and colleagues argue that space operations must be integrated within the existing air traffic management network through advanced communication, navigation, and surveillance capabilities, and that such integration is a precondition for space transport to be technically and commercially viable rather than merely a safety refinement [\[7\]](#ref-7). Thangavel and colleagues develop a multi-domain traffic management architecture that brings conventional air traffic management, higher-altitude operations, and space traffic management together with interoperable separation assurance, and that names reentry as a first-class object on equal standing with launch [\[8\]](#ref-8). The dissertation's significance to this line of work is that a credible estimate of the per-event NAS cost and of the avoided cost of better prediction is the evidence base for the enabling role. It tells a regulator how much airspace disruption a given reentry cadence will cause, and it tells a prediction-and-modeling investment such as ReentryFlow what avoided cost it must clear to be worth funding.

### 1.5.3 Significance to JPL and the civil-space enterprise

For JPL and the broader civil-space enterprise, the same estimate sets a defensible price on the externality that reentry imposes on aviation, which is the quantity that any future corridor-authorization or fee regime must reference. The economic literature on space as a congestible common resource frames the need for such a price clearly. Rao and colleagues analyze the orbital-use externality and show that an unpriced common resource is over-consumed and that a properly designed use fee can raise the value of the industry by internalizing the externality [\[23\]](#ref-23). Weinzierl argues that the transition from public to commercial space activity is paced by institutional design and the correct pricing of externalities, not by engineering milestones alone [\[16\]](#ref-16). The airspace case is the terrestrial analogue of the orbital case: the reentry operator imposes a delay-and-reroute cost on aviation users, and the absence of a price on that cost is why a measurement of it is a precondition for any corrective instrument. The contribution to the civil-space enterprise is therefore the supply of that price. It is valuable to a regulator weighing cadence against burden, to a modeling investment seeking a target return, and to any future institution that would charge for reentry corridor access.

The significance holds whether the measured effect is large or small. A defensible, design-based number for the cost that reentry imposes on aviation, and for the cost that better prediction avoids, is the evidence base on which any reentry-authorization or corridor-pricing institution must rest, and a precisely estimated small effect is as decision-relevant as a large one, because it would tell the enterprise that the airspace externality is not the binding constraint and that investment should go elsewhere.
## 1.6 Scope, delimitations, and definitions

### 1.6.1 Scope and delimitations

This dissertation measures an airspace disruption cost and an avoided cost through a design-stage econometric study. Several boundaries follow from that scope. They are stated explicitly so that the contribution is not read for more than it claims.

The study is design-stage. No estimate is executed on the full dataset. The EU Space Surveillance and Tracking (EU SST) reentry catalog and the FAA System Wide Information Management (SWIM) operations data are accessed under program credentials but are not yet joined into the analysis panel. Every numerical value in the dissertation is therefore an expected range drawn from prior literature or an illustrative value used to demonstrate a procedure, and is labeled as such. The result tables in the analysis-plan chapter are specified but, by design, unpopulated.

The study measures cost, not safety. The withheld-volume policy exists to manage a ground-and-air collision risk, and the dissertation does not propose to relax that risk tolerance. The avoided-cost parameter is the cost saving available at equal safety: the saving from sizing the closure to the prediction actually available rather than to a worst-case envelope, not a saving bought by accepting more risk. The residual-risk boundary is bounded by validated prediction accuracy and by the existing ground-and-air risk limits, and the dissertation treats those limits as fixed constraints rather than as variables to be traded.

The study is local to the observed cadence and event mix. The design-based tradition is explicit that an estimate is local to the events and margins observed. The dissertation honors that caution by reporting the cost as a function of cadence and prediction uncertainty rather than as a single scalar, so that extrapolation to a future high-cadence down-mass regime is conditional and stated rather than assumed.

The down-mass sector is treated as the motivating frontier, not as the current empirical base. The realized record from which the disruption parameter is estimated is dominated by uncontrolled reentries of rocket bodies and satellites, because a commercial down-mass sector at scale does not yet exist. The dissertation is careful not to overstate the reentry-specific empirical depth available; reentry-specific quantification leans on a small set of sources [\[1\]](#ref-1), [\[38\]](#ref-38), [\[64\]](#ref-64), and the down-mass framing is the reason the measurement matters going forward, not a claim that down-mass events already populate the data.

### 1.6.2 Definitions of key terms

The following terms are used throughout the dissertation with fixed meanings.

**Down-mass.** The commercial return to Earth of objects mined, refined, or manufactured in space. Down-mass is distinguished from the uncontrolled reentry of spent rocket bodies and defunct satellites by intent and value: a down-mass object is returned deliberately because its terrestrial delivery is the point of the activity. The current empirical record is dominated by uncontrolled reentries, but the economic motivation for the measurement is the emerging down-mass sector.

**Aircraft Hazard Area (AHA).** The volume of airspace closed to aircraft around a predicted launch or reentry trajectory for a defined window, to manage the collision risk from the vehicle or its potential debris. A static AHA is sized to a worst-case dispersion envelope and fixed in advance; a dynamic AHA is sized to the prediction uncertainty available at the closure-decision time and may narrow as the prediction tightens.

**Exposure.** The condition of a flight whose filed or flown route and time intersect an active AHA polygon and window. Exposure is the treatment in the research design and is constructed by intersecting each flight's route and time against the AHA polygons and windows derived from the reentry catalog through the ReentryFlow exposure mapping.

**Disruption parameter.** The per-event NAS cost, defined as the aggregated average treatment effect on the treated of a reentry-driven AHA closure on the cost outcomes of exposed flights. The cost outcomes are delay minutes, added flown distance, modeled fuel burn, and direct operating cost.

**Avoided-cost parameter.** The difference in the per-event cost between a static-closure regime and a prediction-informed dynamic-closure regime. It is estimated as the realized static-closure cost minus the simulated dynamic-closure cost, per event and then averaged, and separately by using prediction uncertainty as a continuous treatment intensity.

**Treatment intensity.** The prediction-uncertainty bound at the closure-decision time, taken from the reentry catalog, which scales the size of the closure. It is used as a continuous treatment intensity in the estimation of the avoided-cost parameter.

These definitions are fixed and are not redefined in later chapters; later chapters operationalize them into measured variables but do not change their meaning.

## 1.7 The single falsifiable contribution, stated as H0 and H1

The contribution of this dissertation is a single, falsifiable, two-part claim, stated as competing hypotheses. The hypotheses are stated here in their fixed form and are not reworded elsewhere.

**H0, the null.** Spacecraft reentry and down-mass operations impose no measurable cost on the NAS. The average treatment effect of a reentry-driven AHA activation on exposed-flight cost is statistically indistinguishable from zero.

**H1, the alternative.** Spacecraft reentry and down-mass operations impose a measurable, positive disruption cost on the NAS in delay, reroute, and closure, and improved reentry prediction combined with dynamic airspace management yields a quantifiable avoided cost that is positive and statistically distinguishable from zero.

H1 decomposes into the two estimable parameters defined in Section 1.4. The first is the disruption parameter, the average treatment effect of a reentry-driven closure on the cost borne by exposed flights, denoted the per-event NAS cost. The second is the avoided-cost parameter, the difference in that cost between a static-closure regime and a prediction-informed dynamic-closure regime. A finding that the disruption parameter is near zero and statistically insignificant falsifies the first part of H1. A finding that the avoided-cost parameter is near zero falsifies the second part and would imply that better prediction does not pay. An underpowered null is inconclusive and is never treated as confirmation of H0; the analysis plan reports the minimum detectable effect at the available sample size and treats a null that the data cannot distinguish from an economically meaningful effect as inconclusive rather than as support for H0.

These hypotheses stand at the head of an argument that the rest of the dissertation develops in five steps. Reentry closures actually withhold airspace and touch real flights [\[1\]](#ref-1), [\[2\]](#ref-2), [\[3\]](#ref-3), [\[38\]](#ref-38). The cost they impose is non-trivial and grows with cadence and traffic density [\[4\]](#ref-4), [\[1\]](#ref-1), [\[11\]](#ref-11), [\[12\]](#ref-12). Prediction uncertainty drives closure size, and tighter prediction combined with dynamic closure shrinks the withheld volume [\[9\]](#ref-9), [\[10\]](#ref-10), [\[21\]](#ref-21), [\[22\]](#ref-22). Dynamic, prediction-informed closure dominates static segregation on cost at equal safety [\[7\]](#ref-7), [\[8\]](#ref-8). The narrowing of the closure is bounded by validated prediction accuracy and by existing ground-and-air risk limits, so it is not bought by relaxing them. The chapters that follow develop each of these in turn, so that the hypotheses are understood as the conclusion of a sustained argument rather than as an isolated pair of claims.

## 1.8 Roadmap of the dissertation

The dissertation is organized into eight chapters and a backmatter, and the logic moves from problem to framework to evidence to design to plan to interpretation to conclusion.

Chapter 2 develops the theoretical framework. It treats each of the three methodological anchors in turn: Rao's orbital-economy externalities, which supply the structure that makes the reentry closure a capacity shock whose cost the imposing operator does not bear; the congestion-externality and delay-economics literature, which supplies the cost framework and the network-propagation account of how a localized disruption spreads [\[11\]](#ref-11), [\[12\]](#ref-12); and North's institutions-and-transaction-costs framework, which supplies the institutional reason the measurement is load-bearing [\[24\]](#ref-24). It closes by assembling these into the conceptual model the empirical work tests: closure size is a function of prediction uncertainty, exposed-flight cost is a function of closure size and traffic density, and the avoided cost is the cost difference between the static and dynamic regimes.
Chapter 3 is the literature review and is the longest chapter. It treats four themes: the airspace-disruption baseline from the launch era; the architectural direction from static to dynamic closure and from launch tools to reentry; delay as a measured cost and its propagation through the network; and the space-economy framing of market formation. Each theme closes with a synthesis and an explicit statement of the gap it leaves. The chapter states plainly that no prior study supplies a causal estimate of a reentry event's effect on realized flight cost, which is the intersection this dissertation occupies.

Chapter 4 specifies the data and the measurement. It treats each of the four named data layers in turn, the EU SST reentry catalog, the FAA NAS and SWIM operations data, the four IAC-26 down-mass PRISMA systematic reviews, and the ReentryFlow model, and then operationalizes every variable in the notation. It treats the reentry-prediction surface as the measurement backing for the treatment intensity, explaining why prediction uncertainty is large and improvable. It is candid about the measurement error in the treatment, because footprint polygons are model outputs whose uncertainty bounds must be propagated rather than treated as exact.

Chapter 5 is the research design. It justifies the staggered difference-in-differences event study of Callaway and Sant'Anna as the primary estimator [\[14\]](#ref-14), explains the two-way fixed-effects bias under staggered and heterogeneous timing that motivates that choice [\[13\]](#ref-13), [\[15\]](#ref-15), writes out the specification, argues the identification assumptions formally, treats the threats to validity with their mitigations, lays out the robustness and placebo battery, reports the power and minimum-detectable-effect analysis, and commits to a pre-registration and computational plan.

Chapter 6 is the analysis plan. It states the fixed five-step estimation procedure and the pre-committed decision rule on H0 and H1, so that neither outcome can be retrofitted after the fact. It states the expected signs with their mechanism reasoning and a calibrated confidence statement on each, designs the illustrative static-versus-dynamic closure simulation with every quantity left as a symbol, handles continuous treatment and spatial spillover, interprets the event-study profile, and presents the specified-but-unpopulated result tables.

Chapter 7 is the discussion. It develops the implications under both outcomes, returns the contribution to each anchor, draws out the policy and mission implications for NASA, JPL, and the civil-space enterprise, engages the rival explanations fully, and states the external-validity posture honestly, reporting the cost as a function of cadence and prediction uncertainty rather than as a single transferable number.

Chapter 8 concludes. It restates the single falsifiable contribution and what stands even if the hypothesis is not confirmed, states the limitations honestly, lays out a concrete future-research program for executing the design on the full dataset and extending it to higher-cadence down-mass regimes, and closes.

The backmatter compiles the full reference list from the project corpus in one consistent style with clickable identifiers, and provides the variable and data dictionary, the derivations of the group-time average treatment effect aggregation and the avoided-cost differencing identity, the instrument and query details, and the supplementary result-table templates.

The thread that runs through all eight chapters is the one stated at the head of this chapter. Reentry imposes a measurable, positive, per-event cost on the NAS, and better prediction plus dynamic closure avoids a quantifiable share of it, unless the data show otherwise, in which case the design is built to say so. The value to NASA, JPL, and the civil-space enterprise is a defensible, design-based price for an externality that is currently unpriced, and that value is realized whether the measured effect is large or small.


# Chapter 2: Theoretical Framework

## 2.1 Chapter thesis: three frameworks converge on one load-bearing measurement

This chapter advances a single claim and then earns it. Three independent theoretical traditions, each developed for a problem that is neither air traffic nor reentry, converge on the same conclusion when they are turned toward the question this dissertation asks: the marginal cost that a spacecraft reentry imposes on the U.S. National Airspace System is the load-bearing quantity in the entire reentry-airspace policy problem, and until it is measured with a credible design, neither the regulator nor the prediction-and-modeling investor can act on a defensible number. What makes that quantity consequential is not its size, which remains unknown at the design stage, but its absence. A cost that is real, growing, and unpriced is a cost that policy cannot see, and a policy that cannot see a cost cannot weigh it. The orbital-economy framework of Rao and colleagues tells us why the cost exists at all and what kind of object it is: an externality, a real burden imposed by one party on another who is not party to the decision that creates it. The congestion-externality and delay-economics tradition of Mayer and Sinai tells us how that externality is structured inside a network, why a localized closure radiates cost beyond the cells it directly touches, and why the absence of a price is the textbook condition under which a system over-consumes the scarce resource. The institutional framework of Douglass North tells us what the measured price is *for*: it is the input that allows the authorization rule to migrate from a high-transaction-cost case-by-case adjudication to a lower-cost, repeatable, prediction-informed institution, and without that input the rule cannot move.

These are not three decorations on a single argument. Each licenses a different link in the chain that runs from the physical reentry event to the policy decision. Rao supplies the reason the cost is an externality and therefore a candidate for pricing. Mayer and Sinai supply the reason the externality propagates and the reason a market left to itself will not internalize it. North supplies the reason measuring the externality is not an academic flourish but the binding precondition for the institutional change that captures the value. The chapter develops each anchor in its own substantive section, states its primary sources, and specifies exactly how it transfers to the reentry-airspace problem, making the logic of each transfer explicit and naming the strongest objection to it rather than suppressing it. A fourth section imports the policy-uncertainty and Pigouvian-pricing literature to supply the pricing instruments that the first three sections imply but do not themselves construct. The chapter closes by assembling all of this into the conceptual model that the empirical chapters test: closure size is a function of prediction uncertainty; exposed-flight cost is a function of closure size and traffic density; and the avoided cost is the difference in that cost between a static-closure regime and a prediction-informed dynamic-closure regime.

The problem this chapter addresses, in the current-state, desired-state, gap, consequence frame that governs the dissertation, is theoretical rather than empirical. The current state of the relevant theory is that each of these three frameworks is mature in its home domain and silent on the reentry-airspace intersection: Rao prices the on-orbit commons but not the airspace one; Mayer and Sinai analyze congestion delay but never as a consequence of a space event; North gives the institutional logic of transaction-cost reduction but was never applied to an Aircraft Hazard Area. The desired state is a single conceptual model in which the reentry closure is named as an externality, located inside a congestible network, and tied to the institutional change a measured price would permit. The gap is that no prior work performs this synthesis, so the empirical design that follows has no off-the-shelf theoretical scaffold and must build one. Leaving the gap unfilled has a clear consequence: an empirical estimate of the per-event NAS cost would arrive without a theory that says what the number means or what decision it should drive, which is precisely the failure mode that makes descriptive airspace-impact studies decision-inert. This chapter supplies the scaffold so that the measured number, when it is produced, is interpretable as a price on an externality and actionable as an input to an institutional choice.

## 2.2 Rao and the orbital-economy externality: the reentry closure as an unpriced capacity shock

### 2.2.1 The framework and its primary sources

The first anchor is the orbital-economy externality framework, whose canonical recent statement is the integrated-assessment work of Rao, Burgess, and Kaffine [\[23\]](#ref-23). The framework treats orbital space as a congestible common-pool resource. Each satellite placed in orbit raises the collision and debris risk borne by every other operator, and because the launching party does not pay for the risk it imposes, the private cost of an additional satellite is below its social cost. The standard externality wedge follows: the market over-supplies orbital occupancy relative to the social optimum, and a corrective instrument, in Rao and colleagues' analysis an orbital-use fee set to the marginal external cost, restores efficiency and, on their integrated-assessment estimates, substantially raises the present value of the industry by averting a congestion-driven collapse of the resource. The companion literature reinforces the structure from an explicitly economic-modeling direction. Bongers and Torres model orbital debris and the market for satellites as a dynamic externality problem in which the debris stock is a negative-productivity capital that accumulates because no operator internalizes its contribution to it [\[57\]](#ref-57). Weinzierl frames the entire transition from public to commercial space activity as governed first by the correct pricing of externalities and the design of institutions, and only secondarily by engineering progress [\[16\]](#ref-16). These three sources share a single analytical move: identify a shared physical resource, identify the party whose private decision degrades it, observe that the degradation cost falls on third parties who are not compensated, and conclude that a measured price on that cost is the precondition for efficient use.

### 2.2.2 The transfer to the reentry-airspace problem

The transfer is as follows. A spacecraft reentry that triggers an AHA closure is an unpriced negative externality of exactly the structure Rao prices for the on-orbit commons, with the NAS sector-time window in the role of the congestible common resource and the reentry operator in the role of the party whose private decision withholds it from other users.

The basis for this reading is direct. A reentry-driven AHA withholds a finite volume of airspace over a sector-time window from the aircraft that would otherwise have used it, and those aircraft are not party to the reentry authorization and receive no compensation for the delay, reroute, and fuel cost they bear. The launch-and-reentry airspace literature documents that real closures touch real flight populations: international carriers on the order of 8 to 10 percent of affected flights and general aviation roughly a third at a major coastal site under fast-time simulation [\[3\]](#ref-3), with the disruption translatable into delay, distance, and fuel-burn penalties on the affected population [\[2\]](#ref-2), and with reentry-driven closures specifically identified as a growing burden tied to constellation deorbit rates [\[1\]](#ref-1).

The principle that licenses treating this cost as an externality is Rao's: when a party's private use of a shared, congestible resource imposes a cost on other users that the party does not bear, that cost is an externality, and the social optimum requires it to be priced [\[23\]](#ref-23). The airspace sector-time window is shared, because many operators contend for it; it is congestible, because its capacity is finite and a closure exhausts it for the window; and the reentry operator's decision to reenter through it imposes a cost on aviation users that the operator does not pay. The three conditions that make the orbital case an externality therefore hold in the airspace case. Two further bodies of work reinforce this. Bongers and Torres demonstrate that the externality logic is not specific to collision risk but applies to any shared resource whose state one party degrades without compensation [\[57\]](#ref-57), and Weinzierl establishes that externality pricing is the governing economic variable for the space sector's commercialization rather than a peripheral one [\[16\]](#ref-16). The congestion-externality structure has, moreover, been formally identified inside aviation itself by Mayer and Sinai, treated in the next section, which shows that the airspace resource exhibits exactly the congestion property this argument requires [\[11\]](#ref-11).

The argument holds for the *structure* of the cost, not yet for its *magnitude*. Rao's framework establishes that the reentry closure is an externality and therefore a candidate for pricing; it does not by itself tell us how large the per-event cost is. The magnitude is precisely what the empirical design is built to estimate, and at the design stage it is unmeasured. Confidence in the structural claim is high because the three defining conditions are individually documented; confidence in any particular magnitude is withheld until the estimate is executed.

The most serious objection is that the orbital externality and the airspace externality differ in a way that breaks the analogy: the orbital commons is degraded *cumulatively and irreversibly*, since debris persists for decades, whereas an airspace closure is *transient*, since the window reopens. This objection is partly conceded and partly answered. It is conceded that the airspace resource is not cumulatively degraded in the way the orbital stock is, so the dynamic-stock externality of Bongers and Torres does not transfer in full [\[57\]](#ref-57). It is answered by observing that transience does not remove the externality; it changes its time profile. A transient closure still imposes an uncompensated cost on third parties during its window, and the network-propagation evidence treated in Section 2.3 shows that the cost outlives the window, so the externality is not as instantaneous as the objection supposes. The static (flow) externality that Mayer and Sinai price is the correct analogue, not the dynamic (stock) externality, and the design adopts the flow version.

### 2.2.3 The mechanism, named

The transfer is not a bare analogy, because the dissertation's causal mechanism names every link. The driver is a reentry event carrying large prediction uncertainty. The mechanism is a conservative static AHA sized to a worst-case dispersion envelope. The observable effect is a large withheld airspace volume over a sector-time window. The operational consequence is that exposed flights delay, hold, or reroute, generating added delay minutes, added distance, fuel burn, and direct operating cost, with network propagation extending the cost beyond the closed window. The strategic implication is an unpriced externality on aviation that scales with reentry cadence, and measuring it is the precondition for a lower-transaction-cost authorization institution and any corridor-pricing instrument. Rao's contribution is to identify the final link of that chain, the strategic implication, as an externality, and thereby to tell us what kind of number the measured per-event cost is: not a curiosity, but a price.
## 2.3 The congestion-externality and delay-economics framework: structure, propagation, and the precondition for pricing

### 2.3.1 The framework and its primary sources

The second anchor supplies the cost framework and the internal structure of the externality. Its foundational statement is Mayer and Sinai, who show that air traffic delay is in substantial part a congestion externality rather than pure waste [\[11\]](#ref-11), [\[45\]](#ref-45). When one operator schedules into a congested hub or consumes scarce airspace capacity, it imposes delay on other operators that it does not bear; a single operator that internalizes the congestion it causes (a dominant carrier at its own hub) delays itself less than the atomistic-competition benchmark would predict. This is the empirical signature Mayer and Sinai exploit to demonstrate that the externality is real and quantitatively important. The result establishes the textbook condition under which a system under-provides the avoidance of disruption: when the party that consumes the scarce capacity does not face the full social cost of doing so, capacity is over-consumed and disruption is under-avoided.

The framework's second pillar is the network-propagation account of delay, which establishes that the externality is not local. Rebollo and Balakrishnan characterize and predict air traffic delays statistically and document how a disruption at one node raises delay at downstream nodes through aircraft, crew, and slot linkages [\[12\]](#ref-12). Fleurquin, Ramasco, and Eguiluz model systemic delay propagation across the U.S. airport network and show that local perturbations can cascade into network-wide congestion through a reinforcing mechanism [\[43\]](#ref-43), [\[40\]](#ref-40). Zhang and colleagues simulate how propagation depends on network configuration, demonstrating that the same initial delay produces different total cost under different topologies [\[41\]](#ref-41). Wu and colleagues model propagation jointly across the airport and airspace network rather than airports alone, which matters here because a reentry closure is an airspace event, not an airport event [\[42\]](#ref-42). Earlier and complementary work establishes the empirical lineage: Beatty and colleagues provided an early evaluation of delay propagation through an airline schedule [\[74\]](#ref-74), Arikan, Deshpande, and Sohoni built stochastic models of airline-network reliability from empirical data [\[77\]](#ref-77), and Gopalakrishnan and Balakrishnan synthesize the control-and-optimization view of air traffic networks in which a localized capacity shock is a network-flow perturbation with system-level consequences [\[39\]](#ref-39).

### 2.3.2 The transfer to the reentry-airspace problem

The transfer is as follows. A reentry-driven AHA closure is a capacity shock with the congestion-externality structure of Mayer and Sinai, and its cost propagates through the NAS network so that the within-window cost understates the true per-event cost.

The closure withholds capacity from a sector-time window, which is the same scarce-capacity-consumption event Mayer and Sinai analyze, except that the consuming party is a space operator external to aviation rather than a competing carrier [\[11\]](#ref-11). The propagation rests on documented cascade dynamics: a localized airspace perturbation raises downstream delay through the network [\[12\]](#ref-12), [\[43\]](#ref-43), [\[42\]](#ref-42), and the magnitude of the cascade depends on network configuration and on whether the perturbation hits the airspace layer rather than only an airport [\[41\]](#ref-41), [\[39\]](#ref-39).

Mayer and Sinai's framework licenses treating any uncompensated consumption of scarce airspace capacity as a congestion externality [\[11\]](#ref-11), and the propagation literature licenses treating any localized airspace capacity shock as having network-wide consequences that exceed its local footprint [\[12\]](#ref-12), [\[43\]](#ref-43). Both apply to the reentry closure without modification because the closure is, in the network's terms, indistinguishable from any other source of withheld capacity once it is active. The network does not know that the cause is a reentry rather than a storm; it responds to the withheld volume. The strongest support for this transfer is the convergence of independent propagation studies on the same qualitative result across different methods: statistical characterization in Rebollo and Balakrishnan, network-physics modeling in Fleurquin and colleagues, configuration-dependent simulation in Zhang and colleagues, joint airport-airspace modeling in Wu and colleagues. When studies using different methods all conclude that local disruption cascades, the conclusion is robust to any single method's assumptions.

The propagation magnitude is configuration-dependent and not a constant [\[41\]](#ref-41), so the degree to which within-window cost understates total cost varies by event, sector, and traffic state. The design therefore treats network propagation as a bounding consideration on the construct rather than a fixed multiplier, and reports the within-window effect as a lower bound on the per-event cost, with the propagation evidence used to characterize, not to point-estimate, the omitted downstream component. Confidence that propagation exists is high; confidence in any specific propagation multiplier is low and is not asserted.

One objection holds that reentry closures may be small enough or rare enough that propagation is negligible, so the within-window cost is the whole cost. This is conceded as possible at low cadence and contested at the cadence the dissertation is concerned with. At one event per month the propagation may be lost in network noise; at one event per week or per day, the same closure recurs often enough that the cumulative propagated cost becomes material, which is why the design estimates the cost as a function of cadence rather than as a single scalar. The objection, if correct, narrows the claim to high-cadence regimes rather than refuting it.

### 2.3.3 Why this framework makes the measurement a precondition

The decisive contribution of this anchor is not the cost accounting; it is the demonstration that the market will not internalize the externality on its own. Mayer and Sinai's result is that uninternalized congestion is the equilibrium when the consuming party does not face the full social cost [\[11\]](#ref-11). Transferred to reentry, this means the space operator, absent an instrument, will continue to impose airspace cost it does not bear, and the cost will grow with cadence. The economic literature is unambiguous that the corrective instrument requires a measured marginal external cost as its input. The propagation literature sharpens this by showing that the cost to be measured is larger than the visible local cost [\[43\]](#ref-43), [\[12\]](#ref-12). Together they establish the precondition claim that the chapter thesis asserts: a measurement is not optional, because without it the externality remains both unpriced and underestimated.

## 2.4 North and the institutional logic: why the measured price lets the rule change

### 2.4.1 The framework and its primary sources

The third anchor is Douglass North's institutional economics, whose canonical statement is *Institutions, Institutional Change and Economic Performance* [\[24\]](#ref-24). North distinguishes institutions, the rules of the game, from organizations, the players who operate within them, and holds that institutions exist to reduce the transaction costs of impersonal exchange. Institutional change is driven by changes in relative prices and in the cost of operating a given set of rules; a rule persists when the cost of changing it exceeds the gain, and changes when a measured shift in relative cost makes the new rule worth its switching cost. North's framework is therefore not merely descriptive of institutions but predictive of when they move: the binding requirement for institutional change is a perceived change in the relative cost of the existing rule versus its alternative, and that perception must be grounded in measurement for a deliberate (as opposed to evolutionary) change to be made. Weinzierl applies this logic to the space sector, arguing that the pace of commercialization is set by institutional design and externality pricing rather than by engineering [\[16\]](#ref-16), and Adilov and Alexander locate the creation of new space markets in deliberate institutional design, with authorization frameworks and corridor access among the operative levers [\[18\]](#ref-18).

### 2.4.2 The transfer to the reentry-airspace problem

The transfer is as follows. The reentry authorization regime is an institution in North's sense, the static case-by-case AHA closure is a high-transaction-cost rule, the prediction-informed dynamic-closure regime is a lower-transaction-cost rule, and the move from the first to the second requires a measured price for the disruption the current rule imposes.

Static segregation adjudicates each reentry conservatively and individually, which is the operational signature of a high-transaction-cost rule: every event is a fresh, costly negotiation between safety conservatism and airspace demand, and the cost falls on aviation users who are not party to the adjudication [\[2\]](#ref-2), [\[1\]](#ref-1). A dynamic, prediction-informed rule would replace the per-event adjudication with a repeatable procedure that narrows the closure as the prediction tightens, lowering the recurring cost of operating the rule.

North's framework supports treating an authorization regime that governs impersonal exchange, here the exchange of airspace access between space operators and aviation users, as an institution, and it holds that deliberate movement from a costly rule to a cheaper one requires a measured change in relative cost to justify the switching cost [\[24\]](#ref-24). This applies to the reentry-airspace regime because that regime has every feature North's institutions have: it governs interaction among parties who do not personally bargain, it imposes transaction costs that vary by rule, and it can be redesigned to lower those costs. Weinzierl and Adilov and Alexander show that this institutional-change logic is the governing dynamic for space-sector commercialization specifically, not merely a general claim about institutions [\[16\]](#ref-16), [\[18\]](#ref-18); their work establishes that authorization predictability and corridor access function as first-order economic variables in this sector, which is the sector-specific evidence North's general logic needs to be load-bearing here.

North's framework holds that a measured price is *necessary* for deliberate institutional change but does not assert that it is *sufficient*: political economy, distributional conflict, and bureaucratic inertia can block a change that a cost measurement would otherwise justify. The dissertation's contribution is therefore the necessary input, not a guarantee of the institutional outcome. Confidence that the measurement is necessary is high; confidence that supplying it will by itself produce the rule change is deliberately not claimed.

One objection holds that the rule could change without a measured price, by analogy or political mandate, so the measurement is not strictly necessary. This is conceded as logically possible and contested as practically unsafe. A regulator that narrows a safety closure without a measured cost to weigh against the closure's residual risk has changed the rule on an unquantified basis, which is the failure the safety-conservative status quo exists to prevent. North's framework is consistent with rules changing for non-measured reasons, but the *defensible* change, the one a safety regulator can adopt and sustain, requires the cost on one side of the ledger to be measured so it can be set against the risk on the other. The objection narrows the claim from "measurement is logically necessary" to "measurement is necessary for a defensible, durable change," which is the claim the dissertation actually needs.

### 2.4.3 The mechanism by which the price enables the institution to change

The institutional mechanism completes the causal chain. The driver is the measured per-event externality cost. The mechanism is that the measurement supplies the missing term in the regulator's cost-risk comparison: a narrower, prediction-informed closure trades a quantified reduction in airspace cost against a bounded increase in residual risk, and the trade can only be adjudicated if the cost side is measured. The observable effect is that the regulator can, for the first time, evaluate a dynamic-closure policy on its merits rather than defaulting to conservatism for lack of a number. The operational consequence is a migration from per-event adjudication to a repeatable rule. The strategic implication is a lower-transaction-cost authorization institution that scales with cadence without scaling aviation cost proportionally. North's contribution is to identify this final link: the measured price is valuable not as a fact about the world but as the input that lets the rule change, and the rule change is where the economic value is captured.

## 2.5 Policy-uncertainty and Pigouvian-pricing analogues: the instruments the first three sections imply

### 2.5.1 The framework and its primary sources
The first three anchors establish that the reentry closure is an externality (Rao), that it propagates and will not self-internalize (Mayer and Sinai), and that measuring it is the precondition for institutional change (North). They imply, but do not construct, the pricing instrument that internalizes the externality and the value of reducing the uncertainty that drives the closure. This section imports two complementary literatures to supply what is implied.

The first is the policy-uncertainty literature. Handley and Limao show, theoretically and with firm-level evidence, that uncertainty in a policy regime is itself a cost: when the rules under which a firm will operate are unpredictable, the firm under-invests and under-commits even if the expected level of the policy is favorable, because option value makes waiting attractive [\[19\]](#ref-19). Uncertainty raises the value of delay and depresses irreversible investment. Transferred to reentry, the predictability of the authorization regime is a first-order economic variable in its own right, independent of the average level of closure cost: a down-mass venture facing unpredictable closure outcomes will under-invest in the return-leg capability even if the average closure cost is modest, because the variance itself is costly. This connects the institutional argument of Section 2.4 to investment behavior. The value of North's lower-transaction-cost rule is not only the lower average cost but the lower uncertainty, and Handley and Limao quantify why that uncertainty reduction matters.

The second is the Pigouvian-pricing and congestion-pricing literature, which supplies the instrument that prices the externality once it is measured. Small and Gomez-Ibanez document the transition of road-congestion pricing from theory to policy, including the practical design problems (where to set the charge, how to handle heterogeneity, how to make the instrument politically durable) that any externality charge faces [\[85\]](#ref-85). Parry and Small derive the optimal corrective tax in a setting with multiple externalities (congestion, accidents, pollution) and show how the optimal charge is assembled from the separately estimated marginal external costs [\[86\]](#ref-86). The transferable content is the method: the optimal corrective charge equals the marginal external cost, the marginal external cost must be measured for each externality component, and the instrument's design must contend with heterogeneity across users (which, for reentry, is the sharp difference in impact between general aviation and international carriers documented in the launch-era evidence) [\[3\]](#ref-3).

### 2.5.2 The transfer, briefly stated

The transfer is as follows. The measured per-event NAS cost is the marginal external cost that a corrective instrument, a corridor fee or an equivalent charge, would be set to, and the avoided-cost parameter is the value of the uncertainty reduction that drives the closure size. The congestion-pricing literature sets the optimal charge equal to the measured marginal external cost [\[86\]](#ref-86), [\[85\]](#ref-85), and the policy-uncertainty literature establishes that reducing regime uncertainty has investment value beyond the average-cost reduction [\[19\]](#ref-19).

Pigouvian theory sets a corrective charge to the marginal external cost; the measured disruption parameter *is* that marginal external cost, so it is the natural argument of the instrument. The policy-uncertainty result, in turn, makes the variance reduction, not only the mean reduction, part of the value the dynamic rule delivers. The road-pricing transition from theory to policy provides the worked precedent that an externality charge grounded in a measured marginal cost is implementable and durable [\[85\]](#ref-85), and the multi-externality optimal-tax derivation provides the assembly rule for combining components [\[86\]](#ref-86).

These are analogues, not identities. Airspace is not a road, and the heterogeneity, safety-criticality, and international-jurisdiction features of airspace make the instrument design harder than congestion pricing. The transfer carries the *form* of the instrument, charge equals measured marginal external cost, and the *recognition* that uncertainty reduction has independent value, not a claim that road-pricing parameters carry over. Confidence in the form is high; confidence in any transferred parameter value is not claimed.

One objection holds that safety externalities should be managed by regulation, not pricing, so the Pigouvian analogue is misplaced. This is partly conceded: the dissertation does not propose to price *safety*, which remains regulated. It prices the *airspace-disruption* externality conditional on a safety constraint, exactly as congestion pricing operates alongside, not instead of, safety regulation of roads. The instrument internalizes the disruption cost within the envelope that safety regulation defines.

## 2.6 The conceptual model the empirical work tests

### 2.6.1 Assembling the anchors into a testable structure

The three anchors and the pricing analogues assemble into a single conceptual model with three linked relationships, each of which becomes an estimable object in the empirical chapters. The model is stated here in words and notation consistent with the dissertation's fixed bible; the estimating equations themselves are developed in the research-design chapter, not here.

**Relationship 1: closure size is a function of prediction uncertainty.** The first relationship is physical and is the lever the avoided-cost claim pulls. The size of the withheld airspace volume is determined by the dispersion envelope that the AHA must cover, and that envelope is driven by the uncertainty in the reentry prediction at the time the closure decision is made. Reentry prediction uncertainty is large and improvable: it is driven by atmospheric density uncertainty, ballistic coefficient, and reentry angle, and it falls sharply as the object approaches reentry. Atmospheric density uncertainty during disturbed conditions can reach 20 to 30 percent and propagate to along-track position errors approaching 100 kilometers per day, which is the magnitude that a worst-case envelope must accommodate [\[21\]](#ref-21), and the short-term reentry-prediction literature documents that these uncertainties tighten as the object nears reentry [\[22\]](#ref-22). The relationship is therefore: a closure decision made hours in advance, sized to a worst-case envelope, withholds a larger volume than a decision made minutes before reentry, sized to the late-available uncertainty. This relationship supplies the treatment intensity in the empirical design: the prediction-uncertainty bound at closure-decision time scales the closure size.

One epistemic commitment must be stated explicitly here. The relationship described above rests on two well-supported premises: that prediction uncertainty is large at long lead times and narrows near reentry, and that a closure sized to a smaller uncertainty envelope withholds a smaller airspace volume. Both premises are documented in the prediction literature [\[21\]](#ref-21), [\[22\]](#ref-22), [\[9\]](#ref-9), [\[33\]](#ref-33). What the literature does not contain, and what an exhaustive search of the reentry-science corpus failed to locate, is a source that directly measured the realized change in AHA footprint area as a function of prediction lead time on operational reentry events. Weitz, Gruber, and Rozen derive the accuracy requirements that would permit dynamic closure [\[9\]](#ref-9); Gruber, Weitz, and Rozen extend those requirements to the reentry case [\[33\]](#ref-33); and Kaltenhaeuser and colleagues begin to test a coordination concept that reduces the launch closure footprint in a European concept demonstration [\[10\]](#ref-10). These are accuracy-requirement derivations and a concept-scale test, not measurements of realized footprint shrinkage on a distribution of uncontrolled reentry events. The functional dependence of AHA footprint on prediction uncertainty is therefore treated in this dissertation as a physically grounded design assumption, not as an empirically confirmed law. It is derived from first principles and consistent with the requirements literature; its magnitude for a given event distribution is an estimand that the execution phase must measure, not a fixed input. If the execution phase finds that footprint shrinkage as a function of prediction tightening is smaller than the derivations suggest, the avoided-cost parameter will be correspondingly smaller, and the framing here commits the research to that finding rather than suppressing it.

**Relationship 2: exposed-flight cost is a function of closure size and traffic density.** The second relationship is economic and supplies the disruption parameter. Given a withheld volume over a sector-time window, the cost imposed is the product of the affected-flight population and the per-flight penalty, mediated by network propagation. The affected-flight population rises with closure size (a larger volume intersects more routes) and with baseline traffic density (a denser sector loses more flights to the same volume), consistent with the launch-era evidence that closure impact varies with site, geometry, and traffic state [\[2\]](#ref-2), [\[3\]](#ref-3), [\[4\]](#ref-4). The per-flight penalty is the delay, added distance, fuel burn, and direct operating cost of the displaced trajectory, and the network-propagation framework establishes that the total cost exceeds the within-window sum because delay cascades downstream [\[12\]](#ref-12), [\[43\]](#ref-43), [\[42\]](#ref-42). The congestion-externality framework establishes that this cost is borne by parties external to the reentry decision and is therefore an externality [\[11\]](#ref-11). This relationship is the object of the staggered difference-in-differences estimate: the average treatment effect of a reentry-driven AHA activation on exposed-flight cost outcomes, denoted the disruption parameter.

**Relationship 3: avoided cost is the difference between static and dynamic regimes.** The third relationship combines the first two and supplies the avoided-cost parameter. A static-closure regime sizes the volume to a worst-case envelope fixed early, when uncertainty is large (Relationship 1), producing a large affected population and a large cost (Relationship 2). A prediction-informed dynamic-closure regime sizes the volume to the late-available, tighter uncertainty (Relationship 1), producing a smaller affected population and a smaller cost (Relationship 2). The avoided cost is the difference: realized static-closure cost minus simulated dynamic-closure cost, per event, then averaged. North's framework establishes that this avoided cost is the relative-cost change that justifies the institutional migration to the dynamic rule [\[24\]](#ref-24), and the Pigouvian analogue establishes that the disruption parameter is the marginal external cost a corrective instrument would price [\[86\]](#ref-86).

### 2.6.2 The model as a directed chain and its empirical counterparts

The three relationships form a directed chain that mirrors the dissertation's named causal mechanism. Prediction uncertainty determines closure size (Relationship 1); closure size and traffic density determine exposed-flight cost (Relationship 2); the difference in that cost across the static and dynamic regimes is the avoided cost (Relationship 3). Each link has an empirical counterpart. Relationship 1 is operationalized through the EU SST reentry catalog, which supplies the prediction-uncertainty bound that defines treatment intensity, and through the reentry-prediction surface that establishes the uncertainty is improvable [\[21\]](#ref-21), [\[22\]](#ref-22). Relationship 2 is operationalized through the join of the EU SST treatment layer to the FAA NAS and SWIM outcome layer via the ReentryFlow exposure mapping, estimated with the Callaway and Sant'Anna staggered difference-in-differences design. Relationship 3 is operationalized through the ReentryFlow dynamic-closure counterfactual, which recomputes each event's cost under a prediction-informed policy so the avoided cost can be formed as a per-event difference.

The model's internal consistency is what makes the empirical claim falsifiable rather than foregone. If Relationship 2 returns a near-zero, precisely estimated disruption parameter, the externality identified by Rao is empirically immaterial in the current regime, and the first part of the contribution is falsified despite the structural argument. If Relationship 3 returns a near-zero avoided cost, then tightening prediction does not shrink the closure cost, the lever in Relationship 1 does not translate into Relationship 2, and the second part is falsified. The conceptual model thus does double duty: it explains what the measured numbers mean (a price on an externality, and the value of an institutional change), and it specifies the precise findings that would refute each part of the contribution.

### 2.6.3 What the conceptual model commits the empirical work to

The model commits the empirical chapters to three things, each carrying a confidence statement calibrated to design-stage evidence. First, it commits to estimating the disruption parameter as a causal treatment effect, not a simulated impact, which is the identification that the descriptive launch-era literature lacks; confidence that this is the right target is high because the externality structure (Section 2.2) and the propagation structure (Section 2.3) both require a causal, network-aware estimate rather than a per-flight accounting. Second, it commits to estimating the cost as a function of cadence and prediction uncertainty rather than as a single scalar, because the propagation magnitude is configuration-dependent (Section 2.3) and the institutional value is cadence-dependent (Section 2.4); confidence that the scalar would be misleading is high, so the functional form is a requirement rather than a refinement. Third, it commits to validating the ReentryFlow instrument before its counterfactual is trusted, because Relationship 3 depends entirely on the instrument's exposure mapping being accurate; confidence in the avoided-cost parameter is therefore explicitly conditional on that validation, and is the lowest-confidence object in the model at the design stage.

The argument that runs through the dissertation is visible in this model. Reentry closures withhold real airspace from real flights (Section 2.2). The resulting externality propagates and grows with cadence (Section 2.3). Prediction uncertainty drives closure size and is improvable (Relationship 1). The dynamic rule is a lower-transaction-cost institution than static segregation (Section 2.4). And the narrowing is bounded by validated prediction accuracy and a maintained safety constraint (Section 2.5). This chapter has supplied the theoretical foundation for each of these claims; the empirical chapters that follow put them to the test.

## 2.7 Chapter summary

This chapter built the theoretical scaffold the empirical work requires by developing three anchors in depth and assembling them into a testable conceptual model. Rao's orbital-economy framework establishes that the reentry-driven AHA closure is an unpriced negative externality, a capacity shock whose cost the imposing operator does not bear, transferred from the on-orbit commons to the congestible airspace common with the structural conditions individually documented [\[23\]](#ref-23), [\[57\]](#ref-57), [\[16\]](#ref-16). The congestion-externality and delay-economics framework of Mayer and Sinai and the network-propagation literature establish that this externality is structured inside a network, that it propagates so the within-window cost understates the true cost, and that the market will not internalize it without an instrument, which makes a measurement a precondition for any correction [\[11\]](#ref-11), [\[12\]](#ref-12), [\[43\]](#ref-43), [\[42\]](#ref-42), [\[41\]](#ref-41), [\[39\]](#ref-39). North's institutional framework establishes that the measured price is the load-bearing input that lets the authorization rule migrate from high-transaction-cost case-by-case adjudication to a lower-cost, repeatable, prediction-informed institution [\[24\]](#ref-24), [\[18\]](#ref-18). The policy-uncertainty and Pigouvian-pricing analogues supply the instrument the first three anchors imply: a corrective charge set to the measured marginal external cost, and an explicit value for the uncertainty reduction the dynamic rule delivers [\[19\]](#ref-19), [\[85\]](#ref-85), [\[86\]](#ref-86).

The conceptual model assembles these into three linked relationships: closure size as a function of prediction uncertainty [\[21\]](#ref-21), [\[22\]](#ref-22); exposed-flight cost as a function of closure size and traffic density [\[2\]](#ref-2), [\[3\]](#ref-3), [\[4\]](#ref-4), [\[1\]](#ref-1); and avoided cost as the difference between the static and the prediction-informed dynamic regime. Each relationship has a named empirical counterpart and each carries a confidence statement calibrated to the design stage, with the avoided-cost parameter explicitly the most conditional object because it depends on the ReentryFlow instrument's validation. The model is constructed so that a near-zero disruption parameter or a near-zero avoided cost would falsify the respective part of the contribution, which is the property that distinguishes a genuine theoretical scaffold from a rationalization of a foregone conclusion. The chapters that follow operationalize this model: the literature review situates it against the three bodies of work it synthesizes, the data and measurement chapter constructs its variables, and the research-design and analysis-plan chapters specify the estimator that turns the model's relationships into testable parameters.


# Chapter 3: Literature Review
## 3.1 Chapter thesis and the shape of the gap

Three established literatures bear directly on the question this dissertation asks, and not one of them closes the intersection where the question actually lives. The airspace-disruption literature measures how many flights a launch or reentry closure touches and what the modeled time and fuel penalty is, but it does so descriptively, by simulation against historical traffic, and it never estimates the causal effect of a reentry event on realized flight cost against a credible counterfactual. The air-transportation economics literature holds a mature account of the cost of delay and of the congestion externality that one airspace user imposes on others, but it does not treat a spacecraft reentry as a source of delay at all. The space-economy literature treats market formation as governed by institutional design and the pricing of externalities, and it names authorization predictability and corridor access as first-order economic variables, but it does not measure the airspace cost those variables are meant to manage. The chapter thesis is therefore precise and falsifiable in its own right. The marginal National Airspace System cost of a spacecraft reentry event, and the cost that improved prediction combined with dynamic closure would avoid, sit at the unoccupied intersection of these three bodies of work, and the design proposed in this dissertation supplies the causal identification the descriptive literature lacks rather than extending an existing causal result.

This is the longest chapter of the dissertation because the case for the gap must be built source by source, not asserted. A reader is entitled to be skeptical that a gap this consequential could survive in a literature as active as airspace integration has become since 2015. That skepticism is healthy, and the chapter answers it the only honest way, by reviewing the work substantively. For each major source the chapter states what it found, by what method, with what limitation, and how that limitation relates to the gap. The review does not catalogue activity; it demonstrates convergence. Every quantified study points the same direction on the sign and rough magnitude of the disruption, and every study leaves the same hole on identification. That convergence is what licenses the propositions stated at the close of the chapter and carried into the research design.

The problem this chapter addresses follows the four-part shape the dissertation uses throughout. The current state of the literature is a descriptive and simulation-based airspace-impact corpus that establishes order-of-magnitude effects for launch closures and, more recently, for a handful of high-profile uncontrolled reentries. The desired state is a credible causal estimate of the per-event NAS cost and of the avoided cost of a prediction-informed dynamic closure regime, expressed so that a near-zero effect would falsify the claim. The gap is that no study estimates either parameter with a defensible identification strategy. The consequence of leaving it open is serious. As reentry cadence rises with large-constellation deorbit and an emerging commercial down-mass sector, regulators and program planners must reason about a structural and growing aviation burden using numbers that were never designed to bear a causal interpretation, and that therefore cannot price the externality or justify the modeling investment that would reduce it.

The chapter proceeds through four thematic reviews, each closing with a synthesis table and an explicit statement of the gap that theme leaves. Theme A treats the airspace-disruption baseline drawn chiefly from the launch era and the small but growing reentry-specific record. Theme B treats the architectural direction of the field, the documented transition from static segregation toward dynamic, prediction-informed closure, and the maturity of its components. Theme C treats delay as a measured economic cost and the network propagation that makes the within-window cost an understatement of the total. Theme D treats the space-economy framing in which authorization and corridor access become economic variables. Section 3.7 integrates the four into a single statement of the intersection gap and states the propositions that follow into the design. Throughout, where the literature offers only correlation, the text says so and downgrades confidence accordingly. The dissertation's entire methodological purpose is to convert a body of suggestive correlation into an identified causal estimate, and that purpose is poorly served by overstating what the prior work already establishes.

## 3.2 Theme A: the airspace-disruption baseline

The empirical anchor for the disruption claim is the launch-era record, and the foundational source is Srivastava and colleagues, who assessed the impact of space launch and reentry operations on the NAS using historical traffic patterns [\[2\]](#ref-2). Their methodological contribution is the translation procedure that converts a closure geometry into delay, distance, and fuel-burn estimates for the affected flight population. Given a hazard area polygon and an active window, they intersect historical flight tracks with the withheld volume, reroute the affected flights around it, and tally the added distance and time. This procedure is the conceptual ancestor of the ReentryFlow exposure mapping that this dissertation uses as its bridge from the reentry catalog to the flight layer, and the present design treats Srivastava and colleagues as the source of the method rather than as a source of a causal estimate. Its limitation, shared with the entire descriptive tradition, is that the counterfactual is the analyst's reroute assumption rather than an observed comparison group. The study reports what the displaced flights would cost under a modeled rerouting, not what comparable unexposed flights actually did. That distinction is the whole methodological burden the dissertation carries, and it is why this work supports the expected sign and magnitude but cannot stand as the identified effect.

Tinoco and colleagues extended the simulation approach and produced the most frequently cited quantitative figures in the field [\[3\]](#ref-3). In a four-dimensional closure analysis at Cape Canaveral they report that international carriers accounted for 9.5 percent of flights impacted by vertical launches and 8.3 percent by horizontal spaceplane landings, while general aviation represented 33 percent of impacted flights. The same research group's companion simulation, reported by Tinoco, Eudy, and Cannon, analyzes four-dimensional airspace closures due to commercial space operations and confirms that the majority of impacted flights were domestic U.S. carriers while international carriers absorbed a disproportionate share relative to their traffic count [\[32\]](#ref-32). These two figures, the roughly 8 to 10 percent international-carrier share and the roughly one-third general-aviation share, recur as the prior ranges the dissertation carries into its expected-findings discussion. The method is fast-time simulation over historical or representative traffic, and its strength is that it resolves the impact distribution by aircraft class, which is the heterogeneity the dissertation's covariate set retains. Its limitation is again the absence of a credible control. The affected-flight percentages are counts of intersected flights under a closure scenario, not treatment effects estimated against unexposed comparators, and the cost attributed to them is modeled rather than realized.

Young projected the effects of future launch and reentry operations on the NAS and occupies a distinctive place in the corpus because it is one of the few sources that couples launch and reentry demand explicitly rather than treating reentry as an afterthought to launch [\[4\]](#ref-4). Young's value to this dissertation is the forward-looking framing. The study reasons about how impact scales with operational frequency, which is the cadence dependence the design makes central by reporting cost as a function of cadence rather than as a single scalar. The limitation is that the projection rests on assumed growth trajectories and assumed closure geometries; it is a scenario analysis, and its outputs are conditional on inputs that the present design would instead estimate. Robson, Bolic, and Cook examined air traffic management strategies for, and impacts of, space launches from the European side [\[5\]](#ref-5), finding that closure impact varies by aircraft type and flight profile and that the coordination load on air navigation service providers rises with cadence. The European vantage matters because it surfaces the institutional cost of coordination that the U.S.-centric literature tends to fold into the closure itself. The limitation is that the study is again descriptive and strategy-oriented, mapping the management options rather than estimating the cost of any one of them.

The reentry-specific record, as distinct from the launch-centric one, is thinner and more recent, and Wright and colleagues are its center of gravity [\[1\]](#ref-1). They quantified airspace closures caused by reentering space objects and, for the dissertation's framing, they cast those closures not as a rare contingency but as a forward-looking and growing burden tied to constellation deorbit rates. This is the source that converts the problem from an episodic curiosity into a structural one, and it supplies the cadence-growth premise that motivates the entire study. The companion line of work by Hook, Wright, Byers, and Boley on uncontrolled reentries and aviation safety sharpens the empirical anchor with a concrete event [\[38\]](#ref-38). The November 2022 reentry of a Long March 5B rocket body caused the closure of airspace over Europe, delaying 645 flights with a plausible economic impact in the millions of euros. That single documented event is the closest the prior literature comes to a measured per-event cost, and it is instructive precisely because of what it is and is not. It is an existence proof that a reentry closure imposes a large, countable aviation cost. It is not an identified estimate. The 645 delayed flights are a count under one closure, the euro figure is a plausibility assessment rather than a treatment effect, and there is no comparison group of unexposed flights against which the closure's marginal effect is isolated from the day's other disruptions. The dissertation builds directly on this anchor and supplies the missing element.

Three further reentry-specific sources fill out the theme and each contributes a distinct piece. Pardini's analysis of the risk on the ground and in the airspace posed by uncontrolled reentries asks whether the growth observed in recent years should be considered worrying, and concludes from the trend in reentering mass and frequency that the airspace dimension of the problem is rising in step with the ground-risk dimension that the debris-survival literature has long studied [\[64\]](#ref-64). The Acta Astronautica treatment of the need to assess and mitigate the risk from uncontrolled reentries reinforces the same direction from the risk-assessment side, framing the hazard area as the operational instrument through which the on-orbit population converts into a terrestrial and airspace cost [\[63\]](#ref-63). Stefanescu, Constantinescu, and Pleter, in a study explicitly aimed at minimizing air traffic disruption from uncontrolled space debris reentries, establish two numbers that the dissertation uses carefully [\[37\]](#ref-37). They estimate the probability of collision during such events to be on the order of ten to the minus seven, classified as extremely remote yet still requiring mitigation action, and they observe that any given location remains at risk for no more than approximately one minute. The minute-scale temporal exposure against the hours-scale static closure is the precise wedge that motivates the avoided-cost hypothesis, because it shows that the conservatism of static segregation is not a fixed physical necessity but a function of how the spatial problem is converted into a temporal one. The limitation of Stefanescu and colleagues, candidly stated in their own framing, is that the dynamic allocation they propose is a forward proposal whose efficiency gain is asserted from the geometry rather than measured against realized operations, which is again the hole the dissertation's avoided-cost parameter is built to fill.

Two more sources round out the disruption baseline with method detail the design depends on. Bojorquez and Chen developed a risk-level analysis for the hazard area during commercial space launch [\[30\]](#ref-30), dividing the hazard area into multiple sections and dynamically evaluating each for a risk level that combines the uncertain debris trajectory model with launch failure probabilities. Their contribution is the demonstration that a hazard area is not an indivisible block but a graded field whose risk varies across its extent, which is the technical premise that makes a narrower, prediction-informed closure conceptually coherent rather than merely a hope. Assessing the factors that affect the safety of space launch and reentry operations in the NAS, Tao and colleagues catalogue the determinants of closure size and duration [\[34\]](#ref-34), and although the study is oriented toward safety rather than cost, its enumeration of the drivers, trajectory uncertainty, debris dispersion, traffic density, is the same driver set that the dissertation's mechanism chain names. Rabu's treatment of the handling of external risks, including launch and reentry events, in the aviation and maritime sectors [\[6\]](#ref-6) situates the airspace problem within a broader risk-governance frame and confirms that the institutional response to reentry has been conservative segregation, which is the current-state premise of the problem frame.

Theme A establishes that reentry and launch closures actually withhold airspace, touch real flights, and impose a non-trivial cost whose sign is unambiguously positive. The evidence is the convergent quantification across Srivastava [\[2\]](#ref-2), Tinoco [\[3\]](#ref-3), Tinoco-Eudy-Cannon [\[32\]](#ref-32), Young [\[4\]](#ref-4), Robson [\[5\]](#ref-5), Wright [\[1\]](#ref-1), Hook-Wright-Byers-Boley [\[38\]](#ref-38), Stefanescu [\[37\]](#ref-37), and the reentry-risk assessments of Pardini [\[64\]](#ref-64) and the Acta need-to-assess treatment [\[63\]](#ref-63). When descriptive studies using different sites, geometries, traffic samples, and simulation engines all return the same sign and overlapping magnitude ranges, the existence and approximate scale of the effect are well supported even though no single study identifies it causally. The documented 645-flight European closure, an observed event rather than a simulation, anchors the magnitude in reality. One caution is essential. Every one of these magnitudes is a count or a modeled cost, not an identified treatment effect, and the affected-flight percentages are sensitive to site, geometry, and timing. A critic might add that the apparent cost could be an artifact of the reroute assumptions baked into the simulations; the dissertation answers, in later chapters, by replacing the modeled reroute with a realized-cost comparison against unexposed flights, which is the element this theme shows the prior work lacks. Confidence that the disruption sign is positive is high; confidence in any specific magnitude is moderate and is held deliberately at the level of a prior range rather than a point estimate.

**Table 3.1. Theme A synthesis: the airspace-disruption baseline.**

| Source | Finding | Method | Limitation | Relation to gap |
|--------|---------|--------|------------|-----------------|
| Srivastava et al. [\[2\]](#ref-2) | Closure geometry maps to delay, distance, fuel for affected flights | Historical-traffic intersection and reroute | Counterfactual is the analyst's reroute, not an observed control | Supplies the exposure-mapping method, not an identified effect |
| Tinoco et al. [\[3\]](#ref-3) | Intl carriers 8.3-9.5% of impacted flights; GA ~33% | Fast-time 4D simulation, Cape Canaveral | Counts under a scenario, not treatment effects | Source of the prior affected-flight shares by class |
| Tinoco, Eudy, Cannon [\[32\]](#ref-32) | Majority domestic; intl share disproportionate to count | 4D closure simulation | Modeled cost, no control group | Confirms class heterogeneity the design retains |
| Young [\[4\]](#ref-4) | Impact scales with operational frequency | Forward scenario projection | Conditional on assumed growth and geometry | Motivates cost-as-function-of-cadence framing |
| Robson, Bolic, Cook [\[5\]](#ref-5) | Impact varies by aircraft type; ANSP coordination load rises with cadence | Descriptive strategy analysis (Europe) | Maps options, estimates none | Surfaces coordination cost folded into closures |
| Wright et al. [\[1\]](#ref-1) | Reentry closures are a growing, structural aviation burden | Quantification tied to deorbit rates | Descriptive, no causal design | Supplies the cadence-growth premise |
| Hook, Wright, Byers, Boley [\[38\]](#ref-38) | LM-5B 2022: 645 flights delayed, millions of euros plausible | Single-event documentation | Count under one closure, no comparator | The observed magnitude anchor; missing identification |
| Stefanescu et al. [\[37\]](#ref-37) | P(collision) ~1e-7; location at risk ~1 minute | Risk and temporal-exposure analysis | Dynamic gain asserted from geometry | Motivates the avoided-cost wedge (minutes vs hours) |
| Bojorquez, Chen [\[30\]](#ref-30) | Hazard area is a graded risk field, not a block | Sectioned dynamic risk mapping | Launch-focused, safety not cost | Makes a narrower closure conceptually coherent |
| Pardini [\[64\]](#ref-64); Acta [\[63\]](#ref-63); Rabu [\[6\]](#ref-6) | Reentry airspace risk rising; response is conservative segregation | Risk assessment and governance framing | Descriptive | Establishes the current-state premise |

The gap this theme leaves is the one the dissertation names as its first contribution. The descriptive and simulation literature establishes, convergently and credibly, that reentry closures withhold real airspace and impose a positive, non-trivial, class-heterogeneous cost. It does not estimate that cost as a causal treatment effect against an observed counterfactual, and the reentry-specific quantification in particular leans on a small number of sources and one documented event. The design supplies the identification this body of work lacks; it does not extend an existing causal result, because there is none to extend.

## 3.3 Theme B: the architectural direction from static to dynamic closure

The second theme establishes the claim on which the dissertation's avoided-cost hypothesis depends, namely that the transition from static segregation to prediction-informed dynamic closure is not hypothetical but a documented and partly mature direction of the field, so that the avoided-cost parameter measures the value of completing an adaptation the field has already begun rather than the value of inventing a capability. The architectural anchor is Hilton and colleagues, who argue that space operations must be integrated within the existing air traffic management network through advanced communication, navigation, and surveillance (CNS) capabilities, and that such integration is a precondition for space transport to be technically and commercially viable rather than merely a safety refinement [\[7\]](#ref-7). The method is a synthesis of space traffic management requirements against ATM capabilities, and the finding that matters for the dissertation is the explicit reframing of integration as an economic enabler. The limitation is that Hilton and colleagues set the direction at the level of architecture and requirements without quantifying the cost saving that integration would deliver, which is the quantity the avoided-cost parameter targets.

Thangavel and colleagues advance the architecture substantially with a multi-domain traffic management concept that brings conventional air traffic management, higher-altitude operations, and space traffic management together with interoperable separation assurance, and that names reentry as a first-class object on equal standing with launch [\[8\]](#ref-8). This is the source that most directly authorizes the dissertation's treatment of reentry as a managed traffic event rather than a contingency to be segregated around, and its placement of reentry as an equal-standing object is what justifies treating a reentry-driven Aircraft Hazard Area (AHA) activation as a well-defined treatment. The limitation is that the multi-domain architecture is a design concept whose separation-assurance performance, and therefore whose closure-footprint reduction, is specified rather than measured. The European integration literature reinforces the direction. Radtke and colleagues, in their study of integrating air traffic management and space traffic management across the evolving European aerospace, document the concrete institutional and technical challenges of interfacing ATM and STM and frame higher-airspace integration as the near-term frontier [\[27\]](#ref-27). Tullmann and colleagues, in their white paper on the implementation of a European space traffic management system, supply the systems-level blueprint and confirm that the infrastructure required for dynamic handling, tracking, surveillance, and coordination centers, is the same infrastructure the launch-era tools have begun to provide [\[78\]](#ref-78). Both are white-paper and concept-level contributions, which is their honest limitation; they establish feasibility and direction, not measured efficiency.

The trajectory-accuracy literature is where the dynamic-closure direction acquires quantitative content, and it is the most load-bearing part of Theme B for the dissertation's mechanism. Weitz, Gruber, and Rozen derive the predicted-trajectory accuracy requirements that let automation decide which flights can clear a hazard area before it is activated [\[9\]](#ref-9), which is the technical definition of a dynamic closure: a closure sized to the prediction available at decision time rather than to a worst-case envelope fixed hours in advance. Their companion work, reported by Gruber, Weitz, and Rozen, determines the predicted-trajectory accuracy requirements to reduce the aviation impact of space launch and reentry operations specifically, extending the accuracy-requirement framework to the reentry case the dissertation studies [\[33\]](#ref-33). The finding common to both is that closure size is a decreasing function of prediction accuracy, and that there exists an accuracy threshold above which the dynamic closure withholds materially less airspace than the static one. This is the lever the avoided-cost parameter pulls, and it is the strongest grounds in the corpus for expecting that parameter to be positive. The limitation, which the dissertation respects, is that these are accuracy-requirement derivations for launch and reentry that establish the relationship between prediction accuracy and closure size without measuring the realized cost saving on operational traffic. They tell us the lever exists and how it works, not how much it is worth in delay minutes and operating cost across a real event distribution.

Two further sources establish that the dynamic regime has been instantiated, not merely specified. Kaltenhaeuser and colleagues test a launch coordination center concept in European airspace and begin to quantify how dynamic handling reduces the closure footprint relative to static policy [\[10\]](#ref-10), which is the closest the corpus comes to an observed dynamic-versus-static comparison, albeit for launch and at concept-demonstration scale. Lash and colleagues, in their work on enabling tactical airspace management during space launch and reentry operations through collaborative information exchange [\[28\]](#ref-28), document the operational mechanism, real-time information sharing between the space operator and the air navigation service provider, that converts a static planned closure into a dynamically managed one. The companion uncontrolled-reentry study by Kaltenhauser, Rabus, Freer, and Bogdan on uncontrolled reentry risk for aviation and the benefits of real-time information services [\[80\]](#ref-80) is the single most on-point source for the reentry case. It identifies that current hazard areas are often too large to manage effectively, that the size of the hazard area combined with the uncertainty of location is the core problem, and that real-time monitoring tools such as the U.S. Space Data Integrator and Europe's SpaceTrack already offer the information substrate for narrowing them. The limitation, stated plainly in that work, is that a gap persists specifically for uncontrolled reentries, where prediction timing and precision remain hardest, which is both a caution on how far the dynamic gain extends and a confirmation that the gain is real where prediction is tractable.

Two earlier and two newer sources frame the historical arc and the current frontier. Torres and colleagues developed a dynamic air traffic management approach to operationally responsive space as early as 2007 [\[79\]](#ref-79), with a dynamic flight path tool that uses real-time situational information to minimize the impact of space flight on operations, demonstrating across sample scenarios that the impact of space launch operations on air traffic can be significantly reduced by dynamic rather than static handling. That the core idea is nearly two decades old strengthens the dissertation's framing that the dynamic regime is a maturing adaptation rather than a speculative one; its limitation is that the demonstration is scenario-based and pre-dates the constellation-deorbit cadence that makes the problem urgent. At the current frontier, the Acta Astronautica treatments of space traffic management for large constellations [\[67\]](#ref-67) and of space capacity management and its interaction with space traffic management [\[68\]](#ref-68) establish that the on-orbit population growth driving reentry cadence is itself now a managed-capacity problem, which closes the loop between the upstream driver and the downstream airspace cost the dissertation measures.

Theme B establishes, with an explicit causal mechanism, that prediction uncertainty drives closure size and that tighter prediction combined with dynamic closure shrinks the withheld volume, so the intervention acts on the mechanism rather than merely correlating with reduced cost. The chain runs as follows. A reentry event carries large prediction uncertainty; a conservative static AHA is sized to a worst-case dispersion envelope; the observable effect is a large withheld airspace volume; exposed flights respond by delaying, holding, or rerouting; and the result is an unpriced externality that scales with cadence. The dynamic lever acts on the same chain, because prediction uncertainty falls sharply as the object nears reentry, so a closure sized to the late-available uncertainty withholds a smaller volume and displaces fewer flights. The evidence for reading this as a mechanism rather than a bare correlation is twofold: the accuracy-requirement derivations of Weitz [\[9\]](#ref-9) and Gruber [\[33\]](#ref-33), which establish the functional dependence of closure size on prediction accuracy from first principles, and the operational demonstrations of Kaltenhaeuser [\[10\]](#ref-10), Lash [\[28\]](#ref-28), and Torres [\[79\]](#ref-79), which show the lever working in practice. A relationship derived from requirements and confirmed in demonstration is mechanistic, not coincidental. One caution holds throughout. The realized cost saving has not been measured on a real reentry event distribution, and the uncontrolled-reentry case carries a residual prediction-timing limitation [\[80\]](#ref-80) that bounds how far the gain extends. Confidence that the dynamic lever reduces closure size is high; confidence in the magnitude of the realized cost saving is low to moderate at the design stage, which is why the avoided-cost parameter is labeled expected and is built to be falsifiable.

**Table 3.2. Theme B synthesis: from static segregation to dynamic closure.**

| Source | Finding | Method | Limitation | Relation to gap |
|--------|---------|--------|------------|-----------------|
| Hilton et al. [\[7\]](#ref-7) | Integration via CNS is an economic precondition, not just safety | STM-vs-ATM requirements synthesis | No quantified cost saving | Sets the enabler framing the avoided cost prices |
| Thangavel et al. [\[8\]](#ref-8) | Multi-domain TM with reentry as a first-class object | Integrated architecture concept | Separation performance specified, not measured | Authorizes reentry-AHA as a defined treatment |
| Radtke et al. [\[27\]](#ref-27); Tullmann et al. [\[78\]](#ref-78) | ATM-STM integration feasible; same infrastructure as launch tools | Concept and white paper (Europe) | Feasibility and direction, not measured efficiency | Establishes the dynamic regime is instantiable |
| Weitz, Gruber, Rozen [\[9\]](#ref-9) | Closure size decreases with prediction accuracy; threshold exists | Trajectory-accuracy requirement derivation | Launch-focused; realized saving not measured | Defines the avoided-cost lever mechanistically |
| Gruber, Weitz, Rozen [\[33\]](#ref-33) | Accuracy requirements extended to reentry | Requirement derivation (launch + reentry) | Relationship, not realized cost | Extends the lever to the reentry case |
| Kaltenhaeuser et al. [\[10\]](#ref-10) | Dynamic handling reduces footprint vs static | Launch coordination center concept test | Launch, concept-scale | Closest observed dynamic-vs-static comparison |
| Lash et al. [\[28\]](#ref-28) | Real-time information exchange enables dynamic management | Operational concept | Mechanism shown, gain not quantified | Names the static-to-dynamic conversion mechanism |
| Kaltenhauser et al. [\[80\]](#ref-80) | Hazard areas too large; SDI/SpaceTrack enable narrowing; gap persists for uncontrolled | Reentry information-service analysis | Uncontrolled-reentry prediction still hard | Most on-point; bounds the dynamic gain |
| Torres et al. [\[79\]](#ref-79) | Dynamic flight-path tool significantly reduces impact | Scenario demonstration (2007) | Scenario-based, pre-cadence | Shows the idea is mature, not speculative |
| STM-constellations [\[67\]](#ref-67); capacity mgmt [\[68\]](#ref-68) | On-orbit growth is now a managed-capacity problem | Acta STM analyses | Upstream of the airspace cost | Closes the driver-to-cost loop |
The gap this theme leaves is specific, and it is the avoided-cost contribution. The literature establishes the dynamic-closure direction, derives the relationship between prediction accuracy and closure size, and demonstrates dynamic handling in concept and at small scale, predominantly for launch. It does not measure the avoided cost of a prediction-informed dynamic closure on a realized reentry-event distribution. That quantity has no direct empirical precedent and is supported only by component evidence and analogue. The dissertation states it as an expected and illustrative parameter and builds the design so that a near-zero avoided cost would falsify the second part of the contribution.

## 3.4 Theme C: delay as a measured cost and its propagation

The third theme supplies the cost framework and, for the design's validity, the evidence that a closure's cost outlives its active window because delay propagates through the network. The cost of delay is well measured in the air-transportation economics literature, and the propagation result is what tells the dissertation that the within-window treatment effect understates the true per-event cost.

The cost framework begins with the recognition that air traffic delay is in part a congestion externality. Mayer and Sinai showed, in their analysis of network effects, congestion externalities, and air traffic delays [\[11\]](#ref-11), that when one user consumes scarce airspace or runway capacity it imposes delay on others that the consuming user does not bear, the textbook condition under which a market under-provides the avoidance of disruption. Their National Bureau of Economic Research working-paper version develops the same result on all domestic flights by major U.S. carriers from 1988 to 2000 [\[45\]](#ref-45), finding delay increasing in hubbing activity and the externality component substantial. This is the structural template the dissertation transfers to reentry: a reentry closure is a capacity shock imposed on the NAS by an operator who does not bear the resulting aviation cost, the same externality form Mayer and Sinai identify among air carriers. Rebollo and Balakrishnan characterized and predicted air traffic delays statistically [\[12\]](#ref-12), documenting how a localized disruption propagates through the network and how delay at one node raises delay downstream. Theirs is the source that most directly licenses the dissertation's use of a localized reentry closure as a primary delay whose downstream consequences are estimable. The earlier foundational evidence agrees: Beatty and colleagues, in a preliminary evaluation of flight delay propagation through an airline schedule [\[74\]](#ref-74), established the schedule-mediated mechanism by which a single late flight delays its successors, the rotation-carried propagation the network studies later formalized. The propagation literature that builds on these foundations is mature and convergent. Fleurquin, Ramasco, and Eguiluz established, in their study of systemic delay propagation in the U.S. airport network [\[43\]](#ref-43), that primary delays affecting some flights propagate, magnify, and can involve a significant part of the network, and they introduced a model that reproduces the observed propagation patterns and identifies a non-negligible risk of systemic instability even under normal operations. Their companion characterization of delay propagation in the U.S. air-transportation network [\[40\]](#ref-40) resolves the topological structure and the role of aircraft rotation in carrying delay from one flight to the next. The method in both is complex-network analysis on operational performance data, and the finding that matters here is that delay is not contained to the disrupted flight or the disrupted window; it cascades. These studies characterize propagation in general rather than attributing it to a reentry closure specifically, but that is the right division of labor: the dissertation supplies the reentry-specific treatment, and the propagation literature supplies the bound on how far its cost extends beyond the closed window.

The propagation result is reinforced from several methodological directions, which strengthens the case for using it as a bound. Zhang, Wu, Zhang, and Witlox modeled flight delay propagation under different network configurations using an epidemic-spreading analogy [\[41\]](#ref-41), finding that original airport traffic, airport connection structure, and turnaround service levels govern the probability of propagation, and that structural changes such as emerging secondary hubs alter propagation patterns. Wu and colleagues modeled flight delay propagation in the airport and airspace network jointly [\[42\]](#ref-42), extending the analysis from airports to the airspace links the dissertation's closures actually sever. Kafle and Zou developed an analytical-econometric approach to modeling flight delay propagation [\[75\]](#ref-75), methodologically the closest to the dissertation's own econometric posture, and it supplies a template for treating propagation as an estimable quantity rather than only a simulated one. Li and colleagues, in their review of flight delay propagation modeling covering data, methods, and future opportunities [\[44\]](#ref-44), synthesize the field and confirm both the robustness of the propagation finding and the absence of any treatment of space operations as a propagation source. Giannikas and colleagues contributed a data-driven method to assess the causes and impact of delay propagation [\[76\]](#ref-76), which matters because it moves the field toward causal attribution of propagation, the same direction the dissertation moves the airspace-impact field. The shared limitation across this cluster is that none treats a spacecraft reentry as the primary disruption; the cluster is the cost-and-propagation framework into which the dissertation inserts reentry as a new and unstudied source.

Two operations-research sources bound the economic stakes and the infrastructure that carries them. Arikan, Deshpande, and Sohoni built stochastic models of airline networks using empirical data [\[77\]](#ref-77) and report the order of magnitude at issue: a U.S. congressional committee estimated the total cost to the U.S. economy from flight delays at as much as 41 billion dollars in 2007. That figure is not a reentry cost and is not used as one; it is the scale against which a per-event reentry cost multiplied by a rising cadence must be judged for materiality, and it is the reason the dissertation treats the cost's materiality as well supported even before any reentry-specific estimate exists. Gopalakrishnan and Balakrishnan, in their review of the control and optimization of air traffic networks [\[39\]](#ref-39), confirm that operational inefficiencies including delay produce both economic and environmental impacts and that Markov jump linear system models capture the salient dynamics of delay networks, which supplies the dynamical-systems backing for treating the post-activation decay of the event-study profile as a structured, model-consistent object rather than noise.

Theme C establishes that a reentry closure's true cost exceeds the cost measured within its active window, because the displaced delay propagates downstream. The evidence is the convergent propagation findings of Fleurquin and colleagues [\[43\]](#ref-43), [\[40\]](#ref-40), Zhang and colleagues [\[41\]](#ref-41), Wu and colleagues [\[42\]](#ref-42), Kafle and Zou [\[75\]](#ref-75), Li and colleagues [\[44\]](#ref-44), and Giannikas and colleagues [\[76\]](#ref-76), across complex-network, epidemic-analogy, econometric, and data-driven methods. A finding this stable across this many methods is a property of the network, not of any one modeling choice, and so it applies to a reentry-induced primary delay as it does to a weather-induced or traffic-induced one. The mechanism is explicit: a closure forces delay on exposed flights; those flights carry the delay into their subsequent rotations and connections; the delay magnifies along the network topology; the total cost is the within-window cost plus the propagated cost. This is a causal mechanism, not a bare correlation, because the propagation studies identify the carrying mechanism, aircraft rotation and connection structure, directly. One caution holds: the propagation literature does not attribute any propagation to a reentry closure specifically, so the dissertation uses it to bound and interpret the event-study decay profile rather than to estimate the propagated cost directly. Confidence that within-window cost understates the total is high; confidence in the magnitude of the understatement is moderate and is operationalized through the event-study profile rather than asserted.

**Table 3.3. Theme C synthesis: delay cost and network propagation.**

| Source | Finding | Method | Limitation | Relation to gap |
|--------|---------|--------|------------|-----------------|
| Fleurquin et al. [\[43\]](#ref-43) | Primary delays propagate, magnify, risk systemic instability | Complex-network model on performance data | Not attributed to reentry | Bounds cost beyond the closed window |
| Fleurquin et al. [\[40\]](#ref-40) | Aircraft rotation carries delay flight-to-flight | Network topology analysis | General, not space-specific | Identifies the carrying mechanism |
| Zhang et al. [\[41\]](#ref-41) | Traffic, connectivity, turnaround govern propagation | Epidemic-spreading simulation | Configuration-dependent | Confirms structural drivers of decay shape |
| Wu et al. [\[42\]](#ref-42) | Propagation extends across airspace links, not just airports | Joint airport-airspace model | General | Covers the links closures sever |
| Kafle, Zou [\[75\]](#ref-75) | Propagation is econometrically estimable | Analytical-econometric model | Not space-specific | Template for the dissertation's posture |
| Li et al. [\[44\]](#ref-44) | Propagation robust; no space-ops treatment | Field review | Identifies the absence | Confirms reentry is unstudied as a source |
| Giannikas et al. [\[76\]](#ref-76) | Causal attribution of propagation is feasible | Data-driven causal method | Not space-specific | Parallels the dissertation's causal turn |
| Arikan et al. [\[77\]](#ref-77) | Delay cost to U.S. economy ~41bn USD (2007) | Stochastic network model | Total, not per-event reentry | Sets the materiality yardstick |
| Gopalakrishnan, Balakrishnan [\[39\]](#ref-39) | Delay networks are model-tractable dynamical systems | Markov jump linear systems review | General | Backs the event-study decay interpretation |

The gap this theme leaves is narrower than in Themes A and B but real. The cost-of-delay and propagation literature is mature, convergent, and methodologically sophisticated, and it supplies both the cost framework and the propagation bound the dissertation needs. It contains no treatment of spacecraft reentry as a source of delay or of propagated delay. The dissertation inserts reentry into this framework as a new primary-disruption source and uses the propagation result to interpret why the within-window estimate is a lower bound on the true per-event cost.

## 3.5 Theme D: the space-economy framing of market formation

The fourth theme is the shortest because its role is to establish, not to measure: it shows that the leading space-economy literature already treats authorization predictability and corridor access as first-order economic variables, which is what makes a measured airspace cost a load-bearing economic quantity rather than an operational curiosity. Peeters, in his analysis of the paradigm shift of NewSpace and the new business models driving growth of the space economy [\[17\]](#ref-17), locates the sector's expansion in business-model and institutional design rather than in technology readiness alone, and he names the regulatory and access conditions under which a new space activity becomes commercially viable. The method is a sectoral and business-model synthesis, and the finding that matters for the dissertation is the placement of authorization and access predictability among the operative levers of market formation. The limitation, which is precisely the dissertation's opening, is that Peeters identifies these as economic variables without measuring the airspace cost that the access condition is meant to manage; the variable is named but not priced.

This theme connects directly to the dissertation's theoretical anchors developed in Chapter 2, and the connection is stated here only to close the literature's logical circuit, not to re-argue the framework. Weinzierl's account of space as the final economic frontier [\[16\]](#ref-16) establishes at sector scale that the transition from public to commercial space activity is paced by institutional design and the correct pricing of externalities rather than by engineering milestones, the general principle of which the airspace-cost measurement is a specific instance. Adilov and Alexander, on engineering the new space economy through market creation and institutional design [\[18\]](#ref-18), reinforce that authorization frameworks and corridor access are among the levers that determine whether a market forms. The space-economy literature thus arrives at the threshold of the dissertation's question, identifying corridor access as economic, without crossing it to measure the access cost. That threshold is the fourth face of the same intersection gap the other three themes approach from their own directions.

Theme D is brief, and its conclusion is that the airspace cost the dissertation measures is an economic variable of the first order in the space-economy framework, not a peripheral operational detail. The evidence is Peeters [\[17\]](#ref-17), with the framework supplied by Weinzierl [\[16\]](#ref-16) and Adilov and Alexander [\[18\]](#ref-18). When the leading accounts of market formation name authorization and corridor-access predictability as operative levers, a credible measurement of the cost those levers manage is directly relevant to the central question of how the market forms. This literature is framing rather than evidence on the magnitude; it establishes that the measurement matters, not what the measurement is. Confidence that the airspace cost is an economically load-bearing variable is high; the literature offers no estimate of its size, which is the dissertation's task.

**Table 3.4. Theme D synthesis: market formation and the access variable.**

| Source | Finding | Method | Limitation | Relation to gap |
|--------|---------|--------|------------|-----------------|
| Peeters [\[17\]](#ref-17) | Authorization/access predictability is a market-formation lever | Business-model and sector synthesis | Names the variable, does not price it | Establishes the measurement is economically load-bearing |
| Weinzierl [\[16\]](#ref-16) | Sector transition paced by institutions and externality pricing | Economic-perspective synthesis | General principle, not airspace-specific | Supplies the sector-scale rationale |
| Adilov, Alexander [\[18\]](#ref-18) | Corridor access among the institutional-design levers | Market-creation analysis | Framing, no magnitude | Fourth face of the intersection gap |

The gap this theme leaves is the economic-relevance face of the intersection. The space-economy literature establishes that authorization and corridor access are first-order economic variables and that externality pricing paces sector formation, but it does not measure the airspace externality that corridor access is meant to manage. The dissertation supplies that measurement, the input the access variable requires to move from a named lever to a priced one.

## 3.6 Cross-theme integration and the documented absence of a causal estimate

Before stating the gap formally, one point runs across all four themes and a careful reader will already have inferred it: the absence of a causal estimate of reentry NAS cost is not an oversight in the literature but a genuine and confirmable void, and the dissertation's honesty depends on saying so plainly rather than implying a result is being extended. The airspace-impact corpus is rich in simulation, Srivastava [\[2\]](#ref-2), Tinoco [\[3\]](#ref-3), Tinoco-Eudy-Cannon [\[32\]](#ref-32), Young [\[4\]](#ref-4), Robson [\[5\]](#ref-5), the dynamic-tool demonstrations [\[79\]](#ref-79), and the risk analyses [\[30\]](#ref-30), [\[34\]](#ref-34), and it is increasingly rich in reentry-specific quantification [\[1\]](#ref-1), [\[37\]](#ref-37), [\[38\]](#ref-38), [\[63\]](#ref-63), [\[64\]](#ref-64), [\[80\]](#ref-80). It contains no prior causal estimate of a reentry event's effect on realized flight cost using a credible identification strategy. The delay-propagation corpus [\[40\]](#ref-40), [\[41\]](#ref-41), [\[42\]](#ref-42), [\[43\]](#ref-43), [\[44\]](#ref-44), [\[75\]](#ref-75), [\[76\]](#ref-76) is methodologically ready for causal attribution but has not been pointed at reentry. The space-economy corpus [\[16\]](#ref-16), [\[17\]](#ref-17), [\[18\]](#ref-18) names the access variable but does not price it. The convergence of these absences from four independent directions is itself the strongest evidence that the intersection is unoccupied; if any of the four literatures had closed it, at least one of the others would cite the result, and none does.

This matters for the dissertation's contribution in a specific way, and the four themes together make it precise. Theme A establishes, through convergent quantification and one observed 645-flight event, that reentry closures withhold real airspace and impose a positive cost [\[1\]](#ref-1), [\[2\]](#ref-2), [\[3\]](#ref-3), [\[38\]](#ref-38). Themes A and C establish that the cost is non-trivial, grows with cadence and traffic density, and propagates beyond the closed window against a national delay-cost backdrop measured in the tens of billions [\[4\]](#ref-4), [\[5\]](#ref-5), [\[39\]](#ref-39), [\[43\]](#ref-43), [\[44\]](#ref-44), [\[77\]](#ref-77). Theme B establishes that prediction uncertainty drives closure size and that tighter prediction plus dynamic closure shrinks the withheld volume, derived from accuracy requirements and demonstrated in practice [\[9\]](#ref-9), [\[10\]](#ref-10), [\[28\]](#ref-28), [\[33\]](#ref-33), [\[80\]](#ref-80), and that dynamic, prediction-informed closure dominates static segregation on cost at equal safety in the concept and demonstration record [\[7\]](#ref-7), [\[8\]](#ref-8), [\[27\]](#ref-27), [\[78\]](#ref-78), [\[79\]](#ref-79). The risk-assessment literature bounds the narrowing by validated prediction accuracy and ground and air risk limits [\[30\]](#ref-30), [\[37\]](#ref-37), [\[63\]](#ref-63), [\[64\]](#ref-64). Each of these five points is supported by literature; what no literature supplies, and what the dissertation contributes, is the identified estimate that turns the first two from convergent simulation into a causal number and the third and fourth from demonstrated direction into a measured avoided cost.

## 3.7 The gap and the propositions that follow

The gap, stated with the precision the dissertation requires, is the intersection of the four themes. No study estimates, with a defensible causal design, the marginal NAS cost imposed by a spacecraft reentry event, and no study estimates the cost that improved reentry prediction combined with dynamic airspace management would avoid. Theme A establishes the disruption descriptively but not causally and leans, for reentry specifically, on a thin record and a single observed event. Theme B establishes the dynamic-closure lever mechanistically and demonstrates it for launch but does not measure its realized avoided cost on a reentry-event distribution. Theme C supplies the cost framework and the propagation bound but has never been applied to reentry as a disruption source. Theme D establishes that the airspace cost is an economically load-bearing variable but does not price it. The four gaps are four faces of one void, and the void is the dissertation's contribution: a credible, design-based estimate of the per-event reentry NAS cost and of the avoided cost of prediction-informed dynamic closure, stated so that a near-zero, precisely estimated effect would falsify it.

This is the right point to be explicit about epistemic posture, because the strength of the gap claim and the honesty of the contribution depend on calibrating confidence to the evidence grade. The claim that the gap exists is held at very high confidence: it rests not on the absence of a single result but on the mutually corroborating silence of four active literatures, and the design-based tradition the dissertation follows treats such a convergent absence as a well-identified opportunity rather than a hopeful conjecture. The claim that the disruption parameter will be positive is held at high confidence on sign and moderate confidence on magnitude, because every quantified mechanism in Theme A points the same way while no study identifies the effect causally. The claim that the avoided-cost parameter will be positive is held at moderate confidence on sign and low confidence on magnitude, because it rests on the derived prediction-to-closure relationship of Theme B and on analogue rather than on any measured reentry avoided cost; the corpus is candid that this parameter has no direct empirical precedent, and the dissertation labels it expected and illustrative throughout. What would raise these confidences is what the design is built to produce: a populated panel joining the reentry catalog to the flight layer through a validated exposure model, an executed staggered difference-in-differences estimate of the disruption parameter robust to the modern corrections, and a static-versus-dynamic differencing of the avoided cost. What would lower them, and would constitute a genuine and publishable falsification rather than a failure, is a precisely estimated near-zero disruption effect or a near-zero avoided cost.

From the four-theme synthesis the following propositions carry into the research design, each tied to the literature that grounds it and each stated so the design can confront it.

Proposition one, the disruption proposition. A reentry-driven AHA activation imposes a positive, measurable cost on exposed flights, expressed in delay minutes, added distance, fuel burn, and direct operating cost, with the cost concentrated in a minority of exposed flights and the impact share differing by aircraft class along the lines the simulation literature reports, international carriers on the order of 8 to 10 percent of affected flights and general aviation near one third [\[3\]](#ref-3), [\[32\]](#ref-32). This is the dissertation's first estimable parameter, the disruption parameter, and the literature grounds its expected sign while leaving its causal identification to the design.
Proposition two, the propagation proposition. The cost measured within the closure's active window is a lower bound on the true per-event cost, because the displaced delay propagates downstream through aircraft rotation and network connectivity [\[40\]](#ref-40), [\[43\]](#ref-43), [\[44\]](#ref-44). The design operationalizes this through the event-study profile, in which a post-activation effect that persists beyond the closure window indicates that the within-window estimate understates the total and raises the estimated per-event cost accordingly.

Proposition three, the avoided-cost proposition. Because closure size is a decreasing function of prediction accuracy and prediction uncertainty falls sharply as the object nears reentry, a prediction-informed dynamic closure withholds a smaller volume and displaces fewer flights than a static closure sized to a worst-case envelope fixed hours in advance, yielding a positive avoided cost [\[9\]](#ref-9), [\[33\]](#ref-33), [\[37\]](#ref-37), [\[80\]](#ref-80). This is the second estimable parameter, the avoided-cost parameter. The literature grounds the mechanism but leaves the realized magnitude, which has no empirical precedent, to the design as an expected and falsifiable quantity.

Proposition four, the materiality proposition. The per-event cost scales with cadence and traffic density, so that as reentry frequency rises with large-constellation deorbit and an emerging down-mass sector, conservative static segregation scales aviation cost faster than safety benefit, against a national delay-cost backdrop large enough to make even a modest per-event cost materially consequential at scale [\[1\]](#ref-1), [\[4\]](#ref-4), [\[5\]](#ref-5), [\[67\]](#ref-67), [\[68\]](#ref-68), [\[77\]](#ref-77). The design addresses this by reporting cost as a function of cadence and prediction uncertainty rather than as a single scalar, so that extrapolation to a future high-cadence regime is explicit and conditional rather than assumed.

Proposition five, the economic-relevance proposition. The measured airspace cost is a first-order input to the institutions that govern reentry through shared airspace, because authorization predictability and corridor access are economic variables that a price on the airspace externality makes operable [\[16\]](#ref-16), [\[17\]](#ref-17), [\[18\]](#ref-18). This proposition connects the empirical estimate to the policy contribution developed in the discussion: a defensible per-event cost and avoided cost are the evidence base on which any reentry-authorization or corridor-pricing institution must rest.

These five propositions are the bridge from the literature to the design. They are not predictions the dissertation hopes to confirm; they are the structured claims the design is built to test, each grounded in a body of prior work that establishes its plausibility and each left, by that same body of work, without the causal identification the dissertation supplies. The chapter closes where it opened, on its thesis: three literatures bear on this problem, they converge on the sign and rough scale of the effect and on the direction of its remedy, and they leave unoccupied the intersection where a credible causal estimate of the reentry NAS cost and the avoided cost of better prediction would sit. That intersection is the dissertation's contribution, and the research design that follows is the instrument built to occupy it.


# Chapter 4: Data and Measurement

## 4.1 Chapter thesis and the measurement problem

Four heterogeneous data layers join, through one deterministic instrument, into a single estimable object: a panel of airspace-sector-by-time-window cells in which a reentry-driven Aircraft Hazard Area activation is a clean binary treatment and the realized economic cost of disruption is a measured outcome. The EU Space Surveillance and Tracking reentry catalog supplies the treatment, its timing, and its intensity. The FAA National Airspace System and System Wide Information Management operations data supply the outcomes and the exposure. The four completed IAC-26 down-mass PRISMA systematic reviews supply the prior ranges that bound the plausibility of every parameter. The ReentryFlow model is the bridge that converts a reentry trajectory and its dispersion into an affected-flight set, and the simulator that generates the counterfactual dynamic-closure regime against which the avoided-cost parameter is formed. The chapter's obligation is to show that each layer is real, accessible, and well enough understood for the join to carry a causal interpretation, and to be candid about exactly where each layer introduces measurement error that the design must propagate rather than set aside.

The problem this chapter addresses follows the four-part shape the dissertation uses throughout. The current state is that the two operational data sources at the core of the design, EU SST and the FAA SWIM feeds, exist, are accessed, and are individually well documented, but they have never been joined into a reentry-exposure panel; that panel does not yet exist. The desired state is a fully operationalized measurement system in which every variable in the notation of the shared bible has a construct, an operational definition, a named source, and a scale, and in which the treatment and outcome are measured with stated and bounded error. The gap is that the prior airspace-impact literature reviewed in Chapter 3 builds its quantities from simulation against historical traffic and never measures a realized cost against an observed counterfactual, so it never confronts the measurement-error structure a causal design must take seriously. The consequence of leaving that gap open is that any estimate of the reentry-to-aviation externality would inherit the unexamined assumptions of a simulation rather than rest on measured operational records, which is precisely the weakness that makes the prior numbers unsuitable as a price for the externality. This chapter closes the gap at the level of measurement, ahead of the identification argument in Chapter 5 and the analysis plan in Chapter 6.

The chapter is organized to mirror the join. Sections 4.2 through 4.5 treat each named dataset in depth, in the order in which it enters the panel: the EU SST catalog that defines treatment, the FAA operations data that build outcomes and exposure, the PRISMA reviews that set priors, and the ReentryFlow instrument that bridges and simulates. Each section follows the same five-part template, covering provenance, access, coverage, unit of analysis, and known biases, so that a reader can audit the data the way an examiner would. Section 4.6 develops the reentry-prediction surface that underlies the treatment-intensity variable, because the avoided-cost claim stands or falls on the empirical fact that prediction uncertainty is large at the static-closure decision time and improvable as the object nears reentry. Section 4.7 presents the full variable-operationalization table. Section 4.8 treats data quality, validation against known values, and the ethics and access posture. Where a quantity is a model output rather than a measured value, the text says so and the design carries the model's own error forward; this is the discipline that distinguishes a measurement from an assumption.

A standing caution governs the whole chapter and is stated once here so it need not be repeated at every turn. This dissertation is presented at the design stage. No estimate is executed on the full dataset, and the panel described below is a construction plan, not a built artifact. Where a number appears, it is an expected range drawn from the PRISMA reviews or the cited literature, or an illustrative value used to demonstrate a procedure, and it is labeled as such. The measurement system is specified in full so that it could be executed and audited; it is not claimed to have been executed.

## 4.2 The EU SST reentry catalog: the treatment layer

The claim of this section is that the EU SST reentry catalog is a fit-for-purpose source for defining the treatment, its timing, and its intensity, provided that two known biases, a size-threshold selection toward larger and better-characterized objects, and the model-output nature of its uncertainty bounds, are acknowledged and propagated rather than assumed away.

**Provenance.** The EU Space Surveillance and Tracking service is the operational space-surveillance capability of the European Union, delivering collision-avoidance, fragmentation, and reentry products to registered users from a consortium of national operators and sensors. For this dissertation the relevant product line is the reentry service, which issues predictions and post-event records for uncontrolled and controlled reentries of tracked space objects. The provenance matters because the catalog is not a research dataset assembled after the fact; it is an operational feed produced in near-real-time by the same kind of orbit-determination and atmospheric-drag modeling pipeline whose uncertainty characteristics the reentry-prediction literature studies directly [\[22\]](#ref-22), [\[53\]](#ref-53), [\[55\]](#ref-55). That operational lineage is a strength for the design, because the prediction available in the catalog at a given lead time is, to a close approximation, the prediction a regulator would actually have had when sizing a closure, which is exactly the quantity the treatment-intensity variable is meant to capture.

**Access.** Access is through the EU SST portal and its reentry-bulletin feed, for which the candidate program holds vault credentials. The operational and access posture is consistent with the parallel U.S. capability, the FAA Space Data Integrator, which Mutuel and Murray document as the platform that automates the previously manual, time-consuming, and resource-intensive FAA procedures for tracking launch and reentry vehicle position and mission parameters and pushing them to the air-traffic system [\[29\]](#ref-29). The Space Data Integrator is described here because it establishes that an operational reentry-to-airspace data channel of the type the design depends on is a real and fielded capability, not a research aspiration; the European reentry-information work of Kaltenhaeuser and colleagues makes the same point from the air-navigation-service-provider side, framing real-time reentry information services as the operational complement to the surveillance feed [\[80\]](#ref-80). The design uses the EU SST catalog as its primary reentry source and treats the existence of the U.S. Space Data Integrator as evidence that an equivalent or complementary U.S. feed could be substituted or merged in execution.

**Coverage.** The catalog supplies, per reentry event, the predicted reentry epoch and its uncertainty window, the predicted ground-track and dispersion footprint, the object identity and physical characteristics where known, and the realized reentry time and location after the event. Coverage is global for tracked objects above the catalog's size threshold. Two coverage facts are load-bearing for the design. First, the catalog records both the prediction and the realized outcome, which is what makes it possible to characterize prediction error empirically rather than assume it; the realized reentry time and location are the ground truth against which the along-track and cross-track errors are computed. Second, coverage extends across the rising population of large-constellation deorbits, which is the very growth that converts the airspace problem from episodic to structural; the tracking of Starlink reentries through the rising phase of solar cycle 25 by Oliveira and colleagues, who derive altitudes and velocities for 523 such satellites from two-line-element data, illustrates both the scale of the reentering population the catalog must follow and the geomagnetic-activity sensitivity of the predictions [\[56\]](#ref-56).

**Unit of analysis.** The unit at this layer is the reentry event. Each event with a predicted footprint intersecting U.S. airspace, and each window over which an AHA was or would be active, is a candidate treatment occurrence. The catalog therefore defines the treatment in both senses the design needs: it identifies which events are treatments, and it timestamps the window during which the treatment is active, which is the information the staggered difference-in-differences estimator requires to assign each cell its first-treatment cohort.

**Known biases.** Two biases are acknowledged and propagated. The first is a size-threshold selection bias. Small or untracked objects fall below the catalog's detection and cataloging threshold and are not represented, which biases the sample toward larger, better-characterized reentries. This bias is partly benign for the design and partly a limit on its reach. It is benign in that the larger objects are the ones that drive the largest closures and therefore the largest aviation costs, so the events the catalog captures best are the events that matter most for the disruption parameter. It is a limit in that the estimated cost is local to the larger-object regime and does not speak to the aggregate burden of the long tail of small reentries; the design states this rather than extrapolating across it. The second bias is that the uncertainty bounds are themselves model outputs and carry their own error. The predicted uncertainty window is produced by a drag-and-orbit-determination model whose accuracy is the subject of the reentry-prediction literature, and which the IADC reentry-campaign comparisons show can differ from realized outcomes by margins that vary with atmospheric conditions and object characteristics [\[53\]](#ref-53), [\[54\]](#ref-54). The design does not treat the catalog's uncertainty bound as exact; it records it, treats it as a measured-with-error treatment intensity, and tests sensitivity to its calibration in the robustness battery specified in Chapter 5.

This section establishes that the EU SST catalog defines a treatment that is real, timed, and intensity-graded. The evidence is the catalog's operational provenance, its dual recording of prediction and realized outcome, and its coverage of the deorbit population that motivates the study [\[56\]](#ref-56), [\[80\]](#ref-80). A treatment defined from an operational feed that records both what was predicted and what actually happened is exactly the input a design-based estimator needs, because it supplies the cohort timing and the realized ground truth in one source. One caution holds: the catalog is biased toward larger objects and its uncertainty bounds are model outputs with their own error. A critic might press further, that a treatment built on model-output footprints is measured with error severe enough to attenuate the estimate; the design answers by propagating the catalog's own uncertainty bound into the exposure measure and reporting the attenuation direction explicitly, which converts the criticism into a stated and bounded limitation rather than an unexamined flaw. Confidence that the catalog supports a well-defined treatment is high; confidence in the exactness of any single footprint is moderate and is held at that level deliberately.

## 4.3 FAA NAS and SWIM operations data: the outcome and exposure layer

The claim of this section is that the FAA SWIM feeds and the associated NAS operations archives are a comprehensive and appropriate source for the realized-cost outcomes and the exposure indicator, subject to two known biases, the under-representation of visual-flight-rules general aviation and the imperfect attribution of reroute cause, both of which the design confronts with explicit controls rather than ignores.

**Provenance.** System Wide Information Management is the FAA's service-oriented data-distribution architecture, the backbone through which NAS operational data are published to authorized consumers. Its provenance as the agency's own operational record is the source of its authority for this design: the delay, reroute, and airspace-activation records are not reconstructed by a researcher from public schedules but are the FAA's own accounting of what the system did. The Space Data Integrator platform documented by Mutuel and Murray sits inside this same FAA data ecosystem and is the channel through which launch and reentry mission parameters reach the air-traffic system, which is why the EU SST treatment layer and the FAA outcome layer can in principle be joined on a common operational timeline rather than on a researcher's after-the-fact alignment [\[29\]](#ref-29).

**Access.** Access is through the SWIM subscription channels and the FAA operations data archives. The access path is the institutional one used by air-traffic research generally, and the design's computational plan, specified in Chapter 5, treats the ingestion and normalization of these feeds as a defined pipeline stage rather than an incidental step.

**Coverage.** The feeds and archives provide flight-level operational records: filed and flown trajectories, departure and arrival times against schedule, ground and airborne delay, reroute advisories and traffic-management initiatives, and the activation records of special-use and hazard airspace. Coverage is comprehensive for instrument-flight-rules traffic in U.S.-controlled airspace, which is the population that carries the overwhelming majority of commercial passenger and cargo movements and therefore the majority of the economic cost at stake. The activation records of special-use and hazard airspace are particularly important because they let the design observe, rather than infer, when an AHA was active, which is the operational complement to the EU SST prediction of when one would be active.
**Unit of analysis.** The unit at this layer is the flight-segment-by-time-window observation, which aggregates into the airspace-sector-by-time-window cell that serves as the design's analytic unit. From these records the four cost outcomes are constructed. Realized delay minutes are measured relative to schedule and to an undisrupted baseline. Added flown distance is measured relative to the great-circle or filed route. Modeled fuel burn is derived from the added distance and the aircraft type. Direct operating cost is derived from standard per-block-hour and per-nautical-mile cost factors by aircraft class. Exposure, the treatment indicator at the flight level, is constructed by intersecting each flight's filed route and time with the AHA polygons and windows derived from the EU SST layer through ReentryFlow. The reroute-construction method that makes this exposure measurable from operational data descends from Srivastava's procedure for generating realistic reroutes to assess the air-traffic impact of blocked airspaces, which intersects flight tracks with a closed volume and computes the displaced routing and its penalty [\[31\]](#ref-31), and from the airspace-planning and collaborative-decision-making model of Sherali, Staats, and Trani, which formalizes the selection of surrogate trajectories around space-launch special-use airspace subject to safety, workload, and equity constraints [\[35\]](#ref-35). These sources are cited as the provenance of the exposure-and-reroute construction method, not as sources of an estimate.

**Known biases.** Two biases are acknowledged and addressed. The first is that general aviation operating under visual flight rules is under-represented in the instrument-flight-rules-centric records. This bias matters because the launch-era evidence reviewed in Chapter 3 shows that general aviation bears a disproportionate share of closure impact, on the order of one third of impacted flights in the Cape Canaveral simulations of Tinoco and colleagues [\[3\]](#ref-3). The under-representation therefore biases the measured outcome toward the instrument-flight-rules population and likely understates the total general-aviation cost. The design states this directionally, reports its estimates as applying to the instrument-flight-rules population, and treats the general-aviation share as a separately flagged extrapolation bounded by the prior simulation ranges rather than as a measured quantity. The second bias is that reroute advisories attribute cause imperfectly: a reroute record does not always state cleanly whether the cause was a space closure, weather, or a concurrent traffic-management initiative. This is the central confounding threat the identification strategy must defeat, and the design confronts it on three fronts that Chapter 5 develops and this section only names: explicit control for concurrent weather severity, exploitation of the independence of reentry timing from weather, and placebo tests on sectors adjacent to but outside the closure footprint.

This section establishes that the FAA operations data support measured outcomes and an observed exposure indicator. The evidence is the comprehensive instrument-flight-rules coverage, the agency-internal provenance of the delay and activation records, and the existence of a documented reroute-construction method [\[29\]](#ref-29), [\[31\]](#ref-31), [\[35\]](#ref-35). Realized delay, distance, and activation records drawn from the operator's own accounting are the realized cost that the prior simulation literature only modeled, and that element upgrades the design from description to measurement. One caution holds: general aviation is under-represented and reroute cause is attributed imperfectly. A critic might object that imperfect cause attribution means the measured outcome cannot be assigned cleanly to the reentry closure. The design answers that the identification does not rely on the record's own cause label but on the comparison of exposed and unexposed cells under a credible parallel-trends assumption, with weather and concurrent initiatives controlled, so that the design establishes cause rather than reading it off the advisory. Confidence that the data support measured outcomes is high; confidence that the general-aviation cost is fully captured is low and is flagged as such.

## 4.4 The four IAC-26 down-mass PRISMA reviews: the prior-range layer

The claim of this section is that the four completed IAC-26 down-mass systematic reviews are a legitimate and disciplined source of prior ranges for every model parameter, and that their status as secondary syntheses, several of whose anchor figures are simulation- or analogy-derived rather than measured on reentry, is a recorded limitation that bounds how the priors may be used rather than a reason to discard them.

**Provenance.** The four reviews are project artifacts produced under the PRISMA 2020 systematic-review protocol for the joint IAC-26 down-mass paper, and they are read and cited directly as the parameterized evidence base. They are not external literature in the ordinary sense; they are the dissertation's own structured synthesis of the field, which is why they function as a source of priors rather than as a source of an independent finding. P1, the business-case review, supplies the launch-cost trajectory, the constraint-tier structure, and the return-leg economics that make airspace access an economic variable. P2, the airspace-integration review, supplies the disruption baseline, the affected-flight percentages, the static-to-dynamic management transition, and the transferability assessment of launch-era tools to reentry. P3, the reentry-modeling review, supplies the trajectory, footprint, and demise tool lineage and the dominant uncertainty drivers. P4, the governance review, supplies the authorization-architecture options and the digital-licensing processing-time reductions.

**Access.** Access is to the completed reviews at the IAC-26 down-mass paper repository. Collectively the four reviews carry 502 included references with full PRISMA 2020 documentation, which means that any prior range drawn from them is traceable to its underlying primary sources rather than asserted, a property the design exploits when it reports the provenance of each prior.

**Coverage and unit of analysis.** The unit at this layer is the review-extracted parameter. The coverage is the union of the four facets: business case, airspace integration, reentry modeling, and governance. The parameters the design draws from this layer are the affected-flight fractions, the per-flight cost penalties, the prediction-uncertainty magnitudes, and the dynamic-closure efficiency gains. Each of these enters the dissertation as a prior range that bounds the plausibility of the corresponding estimate, not as the estimate itself. The affected-flight fractions, for instance, set the expectation that the disruption cost will concentrate in a minority of exposed flights with a class-differentiated impact share; the dynamic-closure efficiency gains set the order of magnitude the avoided-cost parameter can plausibly reach.

**Known biases and limitation.** The defining limitation is that these are secondary syntheses, and several anchor figures are simulation-derived or analogy-derived rather than measured on reentry. The clearest case is the governance review's reported 67 to 84 percent processing-time reductions from digital-licensing analogues. That figure is an order-of-magnitude prior for what process and modeling improvements can achieve, drawn from licensing-process analogues, and it is applied in this dissertation to closure footprint rather than to paperwork; it is therefore an analogy, not a measured reentry avoided cost, and the design labels it as such everywhere it appears. The same care attaches to the affected-flight fractions, which originate in the launch-era simulations of Tinoco and colleagues and others [\[3\]](#ref-3) and are reentry-relevant by transfer rather than by direct reentry measurement. Recording these provenances in the certainty assessment is what licenses the use of the priors: a prior whose simulation or analogy origin is documented can legitimately bound a plausibility range, whereas the same number presented as a measured reentry result would be a misrepresentation.

This section establishes that the PRISMA reviews supply disciplined prior ranges, and it holds one caution carefully. The evidence is the PRISMA 2020 protocol, the 502 traceable included references, and the four-facet coverage that spans every parameter family the model needs. A systematic review conducted under a registered protocol with documented inclusion criteria is the standard instrument for setting evidence-based priors, and its traceability lets each prior be audited to its primary source. The central caution is that the reviews are secondary and several anchor figures are simulation- or analogy-derived; they bound plausibility, they do not measure the reentry quantity. A critic might object that priors derived from launch-era simulation and licensing analogues could systematically mislead a reentry estimate. The design answers that the priors are used only to bound and contextualize, never to substitute for the realized-data estimate, and that the divergence between any executed estimate and its prior would itself be a reported finding. Confidence that the reviews supply legitimate priors is high; confidence that any single anchor figure transfers exactly to reentry is low, which is the precise reason the figures are used as ranges and not as point inputs.

## 4.5 The ReentryFlow model: the bridge and the counterfactual simulator

The claim of this section is that ReentryFlow is the deterministic instrument that makes the panel possible, performing two distinct roles, exposure mapping and dynamic-closure simulation, and that because both roles place the model on the causal path, the model is itself an object of validation whose precision and recall against realized FAA exposure must be reported before any of its counterfactual outputs are trusted.

**Provenance.** ReentryFlow is the reentry-to-airspace economic model maintained in the project's GitHub repository, with the concept documented at the MITRE project directory. Its provenance as a project-maintained instrument, rather than a third-party black box, is an advantage for the design because the model's internals, its trajectory-and-dispersion ingestion, its airspace-intersection logic, and its cost-factor application, are inspectable and auditable, which is a precondition for validating it as a measurement instrument.

**Access.** Access is to the model code and its documentation in the project repository and the MITRE project directory. The model is run as part of the panel-assembly pipeline rather than queried as an external service.

**Function and the two roles.** ReentryFlow ingests a reentry trajectory and dispersion footprint, intersects it with an airspace traffic scenario, and returns the affected-flight set and an associated cost in delay, distance, fuel, and direct operating cost. Its first role is the deterministic mapping from a treatment, a reentry event and its footprint, to a predicted exposed-flight set and a predicted cost; this is the bridge between the EU SST event layer and the FAA flight layer, the operation that turns a footprint polygon and a traffic record into the exposure indicator that defines treatment at the flight level. Its second role is as the simulator that generates the counterfactual dynamic-closure regime: for each real event, ReentryFlow recomputes the closure under a prediction-informed dynamic policy and returns the implied cost, so that the avoided-cost parameter can be formed as the difference between the realized static-closure cost and the simulated dynamic-closure cost. The model thus occupies both ends of the second hypothesis, and its intersection-and-reroute logic belongs to the same family as the realistic-reroute generation of Srivastava [\[31\]](#ref-31) and the surrogate-trajectory selection of Sherali, Staats, and Trani [\[35\]](#ref-35), with the probabilistic-footprint estimation of Falsone and Prandini supplying the methodological pattern for characterizing the occupied airspace region under reentry uncertainty as a chance-constrained footprint [\[36\]](#ref-36).

**Validation status as an object.** Because the model is on the causal path in both roles, it cannot be treated as a transparent conduit; it is an object of validation. The validation is specified concretely: ReentryFlow's predicted exposed-flight sets are compared against the realized FAA exposure records for past events, and precision and recall are computed and reported before any counterfactual output is trusted. Precision is the fraction of flights the model predicts as exposed that were in fact exposed; recall is the fraction of actually exposed flights the model captures. This is the validation-against-known-values step for the bridge instrument, and it is non-negotiable: if the model's exposure prediction does not reproduce realized exposure within a stated tolerance, its dynamic-closure counterfactual, which has no realized counterpart to check against, cannot be trusted either. The dynamic-closure counterfactual is the more delicate of the two roles precisely because it is unobserved; the discipline that protects it is to earn trust in the model on the observable exposure-mapping role first, then extend that trust to the counterfactual role only to the degree the validation supports.

This section establishes that ReentryFlow is a usable bridge and simulator conditional on validation. The evidence is the model's inspectable, project-maintained provenance and the established lineage of its intersection-and-reroute logic [\[31\]](#ref-31), [\[35\]](#ref-35), [\[36\]](#ref-36). An instrument on the causal path is only as trustworthy as its agreement with realized values, and an instrument whose internals are auditable can be validated in a way a black box cannot. One caution holds: the model's counterfactual outputs are unobserved and inherit whatever error the validation does not catch. A critic might object that the entire avoided-cost parameter rests on a simulated counterfactual with no realized comparator. The design answers by anchoring the model's credibility in the observable exposure-mapping validation and reporting the avoided-cost parameter as conditional on that validated precision and recall, downgrading confidence wherever the validation is weak. Confidence in the exposure-mapping role, once precision and recall are reported, is moderate to high; confidence in the dynamic-closure counterfactual is moderate at best and is explicitly tied to the validation result.

## 4.6 The reentry-prediction surface: measurement backing for treatment intensity

The claim of this section is that the treatment-intensity variable, the prediction-uncertainty bound at closure-decision time, rests on a well-established physical fact: reentry prediction uncertainty is large at the lead times when static closures are sized and falls sharply as the object nears reentry, so the closure size that the uncertainty drives is itself a lever, and measuring it is the empirical foundation of the avoided-cost claim.

The mechanism is physical and is stated as the chain the bible fixes. The driver is a reentry event whose predicted epoch and location carry large uncertainty at the time a static closure must be decided. The proximate cause of that uncertainty is the difficulty of predicting atmospheric drag, which depends on the object's ballistic coefficient, its reentry angle, and the state of the thermosphere through which it decays. The observable effect is a large along-track and cross-track dispersion in the predicted footprint. The operational consequence is a large withheld airspace volume sized to that dispersion, and therefore a large displaced-flight population and cost. The strategic implication is that any reduction in the prediction uncertainty available at the closure-decision time translates, through the same chain, into a smaller withheld volume and a smaller cost, which is the avoided-cost lever the second hypothesis tests.

Each link in this chain is supported by the reentry-prediction literature, and that support is interpreted rather than merely listed. The dominant role of atmospheric-density uncertainty is established by Geul, Mooij, and Noomen, who analyze the uncertainties and modeling in short-term reentry predictions and decompose the error into its drivers, showing that density uncertainty is the principal contributor and that the prediction tightens as the object approaches reentry [\[22\]](#ref-22). The magnitude of the density problem under disturbed conditions is documented in the thermospheric literature. Sutton, Forbes, and Nerem show, from CHAMP accelerometer data during the severe 2003 geomagnetic storms, that thermospheric density can be enhanced by 200 to 300 percent during maximum geomagnetic activity relative to quiet-time baselines [\[82\]](#ref-82), and the long-term thermospheric-density studies of Keating, Tolson, and Bradford establish that the baseline itself drifts over decades [\[83\]](#ref-83). These density swings propagate directly into along-track position error, which is why a closure sized hours in advance under disturbed conditions must be far larger than one sized minutes before reentry. The ballistic-coefficient contribution is quantified by Gondelach, Armellin, and Lidtke, who estimate ballistic coefficients for reentry prediction of rocket bodies in eccentric orbits from two-line-element data across 101 upper stages and characterize how the prediction accuracy improves at shorter lead times [\[55\]](#ref-55), and the atmospheric-density-model dependence is characterized by Pardini and Anselmo, whose performance evaluation of density models for satellite reentry predictions during high solar activity finds reentry-prediction errors on the order of ten percent computed approximately one month, one week, and one day before final orbital decay, with the error structure depending on the chosen density model [\[54\]](#ref-54).

The tightening of the prediction near reentry, the property the avoided-cost lever depends on, is documented concretely in the IADC reentry-campaign analyses. Seong and colleagues, analyzing the reentry of the CZ-5B rocket body in the May 2021 IADC campaign, report a ballistic-coefficient-estimation method that achieved a difference of 73 seconds from the realized reentry time, an accuracy attainable only in the final approach to reentry [\[53\]](#ref-53). The solar-cycle and geomagnetic dependence of decay, and therefore of prediction difficulty, is established by Oliveira and colleagues for 523 Starlink reentries [\[56\]](#ref-56), by Ashruf and colleagues, who identify a threshold at roughly 67 to 75 percent of the solar-cycle peak above which orbital-decay rates increase sharply [\[84\]](#ref-84), and by the decay-time-estimation work for low-Earth-orbit spacecraft [\[66\]](#ref-66). The footprint-dispersion side of the uncertainty, the cross-track and downrange spread of the debris that the closure must envelope, is characterized by the enhanced covariance-propagation work of Chen, Cao, and Ren, which shows how initial-velocity increments and area-to-mass-ratio diversity amplify the anisotropic spread of the debris cloud and the asymmetry of the hazard-zone boundaries [\[81\]](#ref-81), and by the re-entry-survival and ground-risk-assessment literature that connects the surviving debris population to the area that must be withheld [\[65\]](#ref-65). The risk on the ground and in the airspace from uncontrolled reentries, and whether its recent growth should be considered worrying, is assessed directly by Pardini, who ties the rising reentering mass and frequency to a rising airspace dimension of the hazard [\[64\]](#ref-64).

This section establishes, with a clearly stated confidence calibration, that prediction uncertainty is large at the static-closure decision time, is driven principally by atmospheric-density uncertainty modulated by ballistic coefficient and reentry angle, and falls sharply as the object nears reentry. The evidence is the convergent decomposition and quantification across Geul, Mooij, and Noomen [\[22\]](#ref-22), Pardini and Anselmo [\[54\]](#ref-54), Gondelach, Armellin, and Lidtke [\[55\]](#ref-55), Seong and colleagues [\[53\]](#ref-53), the thermospheric-density record [\[82\]](#ref-82), [\[83\]](#ref-83), the solar-cycle dependence [\[56\]](#ref-56), [\[84\]](#ref-84), and the dispersion-propagation work [\[81\]](#ref-81), [\[65\]](#ref-65), [\[66\]](#ref-66). When independent studies using different objects, sensors, and density models converge on the same error structure and the same tightening-near-reentry behavior, the physical fact on which the treatment-intensity variable rests is well established. The 73-second realized-versus-predicted agreement achievable in the final approach [\[53\]](#ref-53) is an observed accuracy that demonstrates the tightening is real and not merely modeled. One caution holds: the magnitude of the uncertainty at any specific lead time is itself model-dependent and condition-dependent, varying with the density model, the solar and geomagnetic state, and the object's properties. A critic might add that if the uncertainty does not in fact fall enough between the static and dynamic decision times, the avoided-cost lever is small. The design answers by measuring the uncertainty at both decision times empirically from the EU SST catalog's recorded predictions and realized outcomes, letting the data, not the assumption, set the magnitude of the lever. Confidence that prediction uncertainty drives closure size and tightens near reentry is high; confidence in the exact fraction by which it tightens for a given event is moderate and is treated as an estimand rather than an assumption.

A further epistemic limit requires explicit statement. The literature reviewed in this section, and an exhaustive search of the AMOS conference corpus and Acta Astronautica holdings that together cover the principal reentry-science and space-surveillance venues from 2006 to 2026, did not return a source that directly measured the realized reduction in AHA footprint area as a function of prediction lead time on a distribution of operational uncontrolled reentry events. The closest sources are the accuracy-requirement derivations of Weitz, Gruber, and Rozen [\[9\]](#ref-9) and Gruber, Weitz, and Rozen [\[33\]](#ref-33), which establish the functional relationship between prediction accuracy and closure size from first principles, and the concept-demonstration result of Kaltenhaeuser and colleagues [\[10\]](#ref-10), which shows the relationship operating at launch scale in European airspace. Neither constitutes a measurement of realized AHA footprint reduction on reentry events. The relationship between prediction uncertainty and closure size is therefore classified in this dissertation as a physically grounded design assumption, derived from prediction mechanics and accuracy-requirement analysis, consistent with all known evidence, but not yet validated against a distribution of realized reentry closures. The execution phase carries the obligation to supply that validation, and the avoided-cost parameter is reported throughout as an expected and conditional quantity rather than a confirmed one. Should the execution-phase measurement find the footprint-shrinkage relationship weaker than the derivations anticipate, the avoided-cost parameter will be revised downward accordingly; this framing commits the research to that outcome rather than protecting against it.

## 4.7 Variable operationalization
This section operationalizes every variable in the notation of the shared bible: for each, it states the construct, the operational definition, the source dataset, and the scale. By convention, the dependent variables are measured from the FAA operations layer, the treatment and its intensity from the EU SST layer through ReentryFlow, the priors that bound the variables from the PRISMA reviews, and the covariates from whichever layer carries them natively. Measurement-error properties are stated for the variables that carry them, because a causal design that hides its measurement error is not honest about its identification.

The four dependent variables are the cost outcomes for cell \(i\) in window \(t\), jointly denoted \(Y_{i,t}\). Delay minutes are the construct of realized schedule disruption, operationalized as the difference between realized and scheduled gate or fix times for the flights in the cell relative to an undisrupted baseline, sourced from the FAA SWIM and NAS operations records, measured on a continuous nonnegative minutes scale. Added distance is the construct of routing inefficiency imposed by the closure, operationalized as flown distance minus the great-circle or filed-route distance, sourced from the FAA flown-trajectory records, measured in nautical miles. Fuel burn is the construct of the physical resource cost of the added routing, operationalized as a function of the added distance and the aircraft type using standard burn factors, sourced from the FAA records for distance and aircraft type with published burn coefficients, measured in kilograms or pounds of fuel. Direct operating cost is the construct of the monetized cost of the disruption, operationalized from the delay minutes and added distance using standard per-block-hour and per-nautical-mile cost factors by aircraft class, sourced from the FAA records with published cost factors, measured in U.S. dollars. The expected sign on all four, developed as an expectation and not a result in Chapter 6, is positive under H1.

The treatment is AHA exposure, the construct of being subject to a reentry-driven closure. It is operationalized as a binary indicator that the cell intersects an active AHA polygon and window, sourced from the EU SST footprint and window mapped onto FAA routes through ReentryFlow, measured as a zero-or-one indicator at the cell level. The treatment cohort \(G_i = g\) is the construct of first-treatment timing, operationalized as the window in which cell \(i\) first receives a reentry closure, sourced from the EU SST event timestamps, measured as a discrete window index; the never-treated cells have \(G_i\) undefined and serve as clean controls. The treatment is measured with error because the footprint polygons are model outputs. The EU SST uncertainty bounds are propagated into the exposure indicator rather than treated as exact, and the attenuation that classical measurement error in a binary treatment induces toward the null is acknowledged, so that a positive estimate is read as a likely lower bound.

The treatment intensity is the prediction-uncertainty bound at closure-decision time, the construct of how large the closure must be. It is operationalized as the along-track and cross-track dispersion magnitude in the predicted footprint at the time the closure decision is made, sourced from the EU SST catalog's prediction record and backed physically by the reentry-prediction surface of Section 4.6 [\[22\]](#ref-22), [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55), measured on a continuous distance scale (kilometers of dispersion) and used as a continuous treatment intensity. This variable lets the design estimate the avoided-cost parameter directly from natural variation in closure tightness across events with differing prediction uncertainty, in addition to estimating it through the ReentryFlow static-versus-dynamic counterfactual.

The covariates are operationalized as follows. Baseline sector traffic density is the construct of how much traffic the closure displaces, operationalized as flights per unit time in the sector in an undisrupted baseline, sourced from FAA records, measured as a continuous count rate. Time-of-day, day-of-week, and season are temporal-demand constructs, operationalized as categorical or cyclical encodings of the window timestamp, sourced from the window definition, measured categorically. Concurrent weather severity is the construct of the principal confounder, operationalized as a weather-severity index over the sector and window, sourced from meteorological records joined on the operational timeline, measured on a continuous or ordinal severity scale; its inclusion is what lets the design separate the reentry effect from the weather effect, exploiting the fact that reentry timing is independent of weather. Concurrent non-space traffic-management initiatives are the construct of other operational disruptions, operationalized as an indicator or count of active traffic-management initiatives in the cell, sourced from the FAA traffic-management-initiative records, measured as an indicator or count. Aircraft-class mix is the construct of the impact-distribution heterogeneity that the launch-era evidence makes central, operationalized as the share of flights by class (international carrier, domestic carrier, general aviation) in the cell, sourced from FAA records, measured as a set of proportions; it is retained precisely because Tinoco and colleagues show the impact distribution differs sharply by class [\[3\]](#ref-3). Great-circle distance of the affected routes is the construct of route length that scales the absolute cost, operationalized as the great-circle distance of the affected flights' origin-destination pairs, sourced from FAA records, measured in nautical miles.

Two derived parameters complete the operationalization. The disruption parameter is the construct of the per-event NAS cost, operationalized as the aggregated average treatment effect on the treated of AHA exposure on the cost outcomes, estimated by the Callaway and Sant'Anna staggered difference-in-differences estimator [\[14\]](#ref-14), and reported on the same scales as the cost outcomes. The avoided-cost parameter is the construct of the value of prediction-informed dynamic closure, operationalized as the realized static-closure cost minus the simulated dynamic-closure cost per event, then averaged, and estimated separately using prediction uncertainty as continuous intensity; it is sourced from the FAA realized cost and the ReentryFlow dynamic-closure counterfactual, and reported in U.S. dollars per event and per planning horizon. Both derived parameters are design-stage estimands; no value is filled in.

The operationalization rests on one point, held with one caution. Every variable in the bible has a construct, a definition, a source, and a scale, and the variables carrying measurement error declare it; the variable-by-variable mapping above, anchored in the named sources [\[3\]](#ref-3), [\[14\]](#ref-14), [\[22\]](#ref-22), [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55), is the evidence. A measurement system whose every variable is operationalized to its source and scale, with measurement error stated where it exists, is auditable in the way a doctoral design must be. The caution is that the treatment and its intensity are measured with error from model-output footprints, and the dependent variables omit some economic cost (passenger time value, network knock-on) addressed in Chapter 5. A critic might object that the cost proxies understate the true economic cost; the design answers that the omitted downstream cost is bounded using the network-propagation evidence rather than ignored, so the reported cost is a conservative, stated-floor proxy. Confidence that the operationalization is complete and auditable is high; confidence that the proxies capture the full economic cost is moderate and is bounded rather than asserted.

**Table 4.1. Variable operationalization (construct, operational definition, source, scale).**

| Variable (symbol) | Construct | Operational definition | Source | Scale |
|---|---|---|---|---|
| Delay minutes (\(Y_{i,t}\)) | Realized schedule disruption | Realized minus scheduled times relative to undisrupted baseline | FAA SWIM / NAS | Continuous, minutes (>=0) |
| Added distance (\(Y_{i,t}\)) | Routing inefficiency | Flown distance minus great-circle or filed distance | FAA flown trajectories | Continuous, nautical miles |
| Fuel burn (\(Y_{i,t}\)) | Physical resource cost | Function of added distance and aircraft type via burn factors | FAA records + burn coefficients | Continuous, kg or lb fuel |
| Direct operating cost (\(Y_{i,t}\)) | Monetized disruption cost | Delay and distance via per-block-hour and per-nm cost factors by class | FAA records + cost factors | Continuous, USD |
| AHA exposure (treatment) | Subject to reentry closure | Binary: cell intersects active AHA polygon and window | EU SST footprint via ReentryFlow | Binary {0,1}; measured with error |
| First-treatment cohort (\(G_i = g\)) | First-treatment timing | Window of first reentry closure for the cell | EU SST event timestamps | Discrete window index |
| Treatment intensity | Required closure size | Along-track/cross-track dispersion at closure-decision time | EU SST prediction record | Continuous, km dispersion |
| Sector traffic density | Displaced traffic volume | Flights per unit time in undisrupted baseline | FAA records | Continuous, count rate |
| Time-of-day / day-of-week / season | Temporal demand | Categorical/cyclical encodings of window timestamp | Window definition | Categorical |
| Weather severity | Principal confounder | Weather-severity index over sector and window | Meteorological records | Continuous/ordinal |
| Concurrent TMIs | Other operational disruption | Indicator/count of active traffic-management initiatives | FAA TMI records | Indicator/count |
| Aircraft-class mix | Impact heterogeneity by class | Share of flights by class in the cell | FAA records | Proportions |
| Great-circle distance | Route length scaling cost | Great-circle distance of affected origin-destination pairs | FAA records | Continuous, nautical miles |
| Disruption parameter | Per-event NAS cost | Aggregated ATT of AHA exposure on cost outcomes | C&S estimator over panel | Cost-outcome scales |
| Avoided-cost parameter | Value of dynamic closure | Static-closure cost minus simulated dynamic-closure cost | FAA realized + ReentryFlow counterfactual | USD per event / horizon |

## 4.8 Data quality, validation against known values, and ethics and access

The claim of this section is that the measurement system carries an explicit data-quality and validation plan, that it validates its instruments against known values before trusting their outputs, and that its ethics and access posture is sound because the data are operational aggregates rather than personal information and are accessed through authorized institutional channels.

**Data quality.** The quality plan operates at three points in the pipeline. At ingestion, each source is normalized to the common operational timeline that the join requires, and records that cannot be aligned in time, the precondition for joining the EU SST treatment to the FAA outcome, are flagged rather than silently dropped, because a systematic alignment failure would itself bias the panel. At construction, the exposure indicator is the highest-risk derived quantity because it sits on the causal path and is built from a model-output footprint; its construction is therefore validated separately (below) rather than assumed correct. At estimation, the cell-level outcomes are checked for the spatial and temporal autocorrelation that the inference plan must accommodate, since cells in the same sector and adjacent windows are correlated and naive inference would understate the standard errors; the design's response, sector-level clustering and, where event counts are low, the wild-cluster bootstrap, is specified in Chapter 5 and named here so the quality plan and the inference plan are visibly consistent.

**Validation against known values.** The validation plan has two anchors. The first is the ReentryFlow exposure validation already specified in Section 4.5: the model's predicted exposed-flight sets are compared against the realized FAA exposure records for past events, and precision and recall are computed and reported before any counterfactual output is trusted. This is validation of the bridge instrument against the FAA ground truth, and it is the gatekeeper for the entire avoided-cost analysis. The second anchor is the validation of the EU SST prediction layer against its own recorded realized outcomes: because the catalog records both the predicted and the realized reentry epoch and location, the prediction error can be computed directly and compared against the independent reentry-prediction literature, which provides the known-value benchmarks. The IADC-campaign agreement of 73 seconds reported by Seong and colleagues [\[53\]](#ref-53), the roughly ten-percent reentry-prediction errors at one-month, one-week, and one-day lead times reported by Pardini and Anselmo [\[54\]](#ref-54), and the ballistic-coefficient-estimation accuracy of Gondelach, Armellin, and Lidtke across 101 upper stages [\[55\]](#ref-55) are the external known values against which the catalog's prediction error is sanity-checked; a catalog prediction error grossly inconsistent with this literature would indicate a data-quality problem in the prediction layer that must be resolved before the treatment-intensity variable is trusted. The probabilistic-footprint method of Falsone and Prandini [\[36\]](#ref-36) and the covariance-propagation work of Chen, Cao, and Ren [\[81\]](#ref-81) provide the corresponding known-value benchmarks for the dispersion footprint that drives the closure size. Validating each instrument against an independent benchmark before using its output is the discipline that keeps a model-output quantity from masquerading as a measured one.

**Ethics and access.** The ethics posture is favorable and is stated plainly rather than waved through. The FAA operations data are operational aggregates of flight movements, delays, reroutes, and airspace activations; they are records of aircraft and system behavior, not of identifiable individuals, so the human-subjects concerns that attach to personal data do not arise. Access is through authorized institutional channels, the SWIM subscription channels and FAA operations archives on the aviation side, and the EU SST portal under held vault credentials on the reentry side, so the data are used within the terms under which they are provided rather than scraped or repurposed. The Space Data Integrator documented by Mutuel and Murray is the operational platform on the FAA side through which launch and reentry data are integrated into the air-traffic system, which confirms that the data channels the design relies on are sanctioned operational infrastructure rather than informal sources [\[29\]](#ref-29). The PRISMA reviews are the dissertation's own project artifacts, cited as data sources under the project's authorship. ReentryFlow is a project-maintained model accessed through the project repository. The one access-related risk the design records is operational sensitivity: some airspace and traffic-management records carry handling expectations, and the design commits to working within those expectations and to reporting results at the level of aggregate cost rather than in any form that would expose sensitive operational detail.

This section closes the chapter by establishing that the measurement system is quality-controlled, validated against known values, and ethically and procedurally sound. The evidence is the three-point quality plan, the two-anchor validation plan benchmarked against the independent reentry-prediction literature [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55), [\[36\]](#ref-36), [\[81\]](#ref-81), and the operational-aggregate, authorized-channel access posture [\[29\]](#ref-29). A design that validates each instrument against an external known value before trusting it, and that uses operational aggregates through sanctioned channels, meets the data-quality and ethics standards a doctoral measurement chapter must meet. One caution holds: the panel does not yet exist; the EU SST and SWIM layers are accessed but not joined, so the quality and validation plan is a specification to be executed, not a record of execution. A critic might object that an unexecuted plan cannot demonstrate data quality; the design answers that the chapter's task at the design stage is to specify a measurement system precise and auditable enough to be executed and checked, and that specifying the validation benchmarks in advance, against the published literature, is exactly what protects the eventual execution from confirmation bias. Confidence that the measurement system is well specified and auditable is high; confidence that it has been executed is, by the honest design-stage posture of the whole dissertation, deliberately withheld.

## 4.9 Synthesis: the four layers as one estimable object

The chapter set out to show that four heterogeneous layers join into one estimable panel, and the sections above establish each layer's part. The EU SST catalog defines a treatment that is real, timed, and intensity-graded, biased toward larger objects and carrying model-output uncertainty that the design propagates [\[22\]](#ref-22), [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55), [\[56\]](#ref-56), [\[80\]](#ref-80). The FAA NAS and SWIM data supply measured outcomes and an observed exposure indicator with comprehensive instrument-flight-rules coverage, under-representing visual-flight-rules general aviation and attributing reroute cause imperfectly, both confronted by the identification strategy rather than ignored [\[3\]](#ref-3), [\[29\]](#ref-29), [\[31\]](#ref-31), [\[35\]](#ref-35). The four IAC-26 PRISMA reviews supply disciplined, traceable prior ranges whose simulation- and analogy-derived anchors are recorded in the certainty assessment so they bound plausibility without masquerading as reentry measurements. ReentryFlow bridges the EU SST and FAA layers and simulates the dynamic-closure counterfactual, and is validated as an object by precision and recall against realized FAA exposure before its counterfactuals are trusted [\[31\]](#ref-31), [\[35\]](#ref-35), [\[36\]](#ref-36). The reentry-prediction surface gives the treatment-intensity variable its physical foundation, establishing that prediction uncertainty is large at the static-closure decision time and falls sharply near reentry, which is the empirical fact on which the avoided-cost lever depends [\[22\]](#ref-22), [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55), [\[81\]](#ref-81), [\[82\]](#ref-82), [\[83\]](#ref-83), [\[84\]](#ref-84).

The argument runs through these layers in the form the dissertation carries throughout, and the data chapter is where two of its claims are evidenced at the level of measurement. That reentry closures withhold airspace and touch real flights rests on the FAA activation records and the documented reentry-closure events [\[29\]](#ref-29), [\[38\]](#ref-38), [\[64\]](#ref-64). That prediction uncertainty drives closure size and that tighter prediction shrinks the withheld volume rests on the reentry-prediction surface of Section 4.6 [\[22\]](#ref-22), [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55). The remaining claims, that the cost is material, that dynamic closure beats the alternatives, and that the residual risk is acceptable, are evidenced in the chapters that argue magnitude, design, and policy; the data chapter's role is to make the first two measurable rather than merely asserted. With the measurement system specified, the dissertation turns in Chapter 5 to the identification strategy that converts this panel into a causal estimate, and the threats to validity that the measurement-error properties recorded here oblige it to confront.


# Chapter 5: Research Design and Identification

## 5.1 Chapter thesis and the identification strategy

A staggered difference-in-differences event study, estimated with the Callaway and Sant'Anna (2020) group-time estimator, identifies the per-event cost that a reentry-driven Aircraft Hazard Area closure imposes on the U.S. National Airspace System under the assumption that reentry timing is conditionally independent of contemporaneous sector-specific demand shocks [\[14\]](#ref-14). Everything that follows develops, qualifies, and defends that single claim. The estimator is chosen because the variation it exploits, the staggered arrival of reentry closures across airspace sectors and dates, is exactly the variation that the design-based tradition treats as a natural experiment, and because the recent econometric literature has shown that the conventional two-way fixed-effects difference-in-differences regression is biased for precisely this kind of staggered, heterogeneous treatment [\[13\]](#ref-13), [\[15\]](#ref-15), [\[50\]](#ref-50). The chapter writes out the estimating equations, argues the identification assumptions formally, enumerates every threat to validity with a paired mitigation, specifies the robustness battery, reports a minimum-detectable-effect analysis at the available sample size, commits the design to pre-registration, and closes with the computational and software plan that makes the whole apparatus reproducible.

The problem this chapter addresses follows the shape carried through the dissertation. The current state of empirical practice is that the airspace impact of launch and reentry closures is quantified descriptively, by fast-time simulation over historical traffic, which reports how many flights a closure touches and what the modeled fuel-and-time penalty is, but does not estimate a causal treatment effect against a credible counterfactual [\[2\]](#ref-2), [\[3\]](#ref-3), [\[4\]](#ref-4), [\[32\]](#ref-32). The desired state is a defensible, design-based estimate of the marginal cost of a reentry event, expressed as a treatment effect on realized flight-level cost, that survives the modern corrections for staggered timing and reports its own residual uncertainty honestly. The gap is the absence of any such estimate in the literature: no prior study lets a credible source of variation carry the causal claim for reentry-to-aviation cost. The consequence of leaving the gap open is that the regulator who must weigh a reentry cadence against its aviation burden, and the prediction-and-modeling investment that must justify itself against an avoided cost, both proceed without a measured number, which is the same as proceeding without a price on the externality. This chapter converts the gap into an executable identification strategy.

The confidence posture here is design-stage and explicit. The claim that the chosen estimator will recover the target parameter without bias is conditional on assumptions that are argued here and tested in Chapter 6, not assumed away. Where the chapter states that a particular estimator dominates an alternative, that claim rests on the published econometric result, and the condition under which the dominance holds is stated alongside it. No empirical estimate is executed on the full dataset anywhere in this chapter; the few numerical values that appear are illustrative arithmetic or sample-size inputs to a power calculation, and they are labeled as such.

## 5.2 The estimator and why it is chosen

The primary estimator is the Callaway and Sant'Anna (2020) doubly robust group-time difference-in-differences estimator, supported by the Goodman-Bacon (2021) decomposition and the de Chaisemartin and D'Haultfoeuille (2023) robust estimator as diagnostics [\[13\]](#ref-13), [\[14\]](#ref-14), [\[15\]](#ref-15). The argument for this choice is worth stating in full, because the estimator selection is the load-bearing methodological decision of the dissertation and the rest of the design inherits it.
The Callaway and Sant'Anna estimator is the correct primary estimator for the reentry-NAS-cost question because of the structure of the treatment. Reentry-driven AHA closures arrive at staggered times across airspace sectors, the treatment effect is expected to differ across events because closure size, traffic density, and prediction uncertainty vary event to event, and the panel therefore has exactly the staggered-adoption, heterogeneous-effect structure that the recent literature studies. The established result is that, under staggered timing and heterogeneous effects, the two-way fixed-effects (TWFE) estimator does not recover a clean average treatment effect. It forms a weighted average of all possible two-group, two-period comparisons, some of which use already-treated units as controls for later-treated units, and these "forbidden comparisons" can receive negative weights, so the estimated coefficient can carry the wrong sign even when every unit-level effect is positive [\[13\]](#ref-13), [\[15\]](#ref-15). A body of methodological work derives and demonstrates this bias. Goodman-Bacon's decomposition theorem shows that the TWFE estimand is a variance-weighted average of all two-by-two DiD estimates and isolates the forbidden-comparison terms [\[13\]](#ref-13). De Chaisemartin and D'Haultfoeuille characterize the negative-weighting problem and provide an estimator that avoids it [\[15\]](#ref-15). Baker, Larcker, and Wang re-examine published staggered-DiD results and show that the corrected estimates frequently differ in magnitude and occasionally in sign from the TWFE originals, which establishes that the bias is a practical hazard rather than a theoretical curiosity [\[50\]](#ref-50). Athey and Imbens provide the design-based foundation for treating staggered adoption as the assignment mechanism rather than as a nuisance to be absorbed by fixed effects [\[69\]](#ref-69). The Callaway and Sant'Anna estimator is correct under its own maintained assumptions, principally a conditional parallel-trends assumption and a no-anticipation assumption, both argued in Section 5.4 and tested in Chapter 6. It is not assumption-free, and its credibility is exactly the credibility of those assumptions. One alternative must be entertained: a different modern estimator, the stacked estimator of Wing, Freedman, and Hollingsworth, or a synthetic difference-in-differences approach, might be preferable. These are retained as robustness checks rather than as the primary specification, for reasons given below, so that the dissertation reports a frontier-consistent primary estimate and shows that the finding does not depend on the choice among frontier estimators [\[46\]](#ref-46).

The Callaway and Sant'Anna estimator is primary rather than merely one of several acceptable options because it is transparent about the building block of the aggregation. The estimator defines a group-time average treatment effect, the effect for a cohort of cells that first receive a reentry closure in the same window, evaluated at each later window, and it estimates each such parameter using only clean controls, that is, units that are never treated in the sample or not yet treated at the comparison time. The overall average treatment effect on the treated, and the event-study profile by time since closure, are then formed as explicit, interpretable weighted aggregations of these group-time parameters. This matters for the reentry application because the number of treated events is modest and the heterogeneity across them is the substance of the question, not noise to be averaged out. An estimator that exposes the group-time building block lets the analysis report which cohorts and which horizons drive the aggregate, and lets it drop forbidden comparisons by construction rather than by hoping the data are well behaved. The de Chaisemartin and D'Haultfoeuille robust estimator is reported alongside because it is built on a different aggregation logic and a different treatment of treatment-reversal, so agreement between the two is evidence that the result is not an artifact of one estimator's particular weighting [\[15\]](#ref-15). The Goodman-Bacon decomposition is reported because it makes the TWFE bias visible: by decomposing what a naive TWFE regression would have estimated into its component comparisons, it shows directly how much of the naive estimate would have come from forbidden comparisons, which is the most legible possible demonstration that the modern estimator was necessary rather than ornamental [\[13\]](#ref-13).

Two further estimators are specified as confirmatory rather than primary, and the reasoning for their subordinate status is part of the design. The stacked difference-in-differences estimator constructs a separate clean two-by-two comparison for each treated cohort against its own not-yet-treated controls and then stacks these sub-experiments, which is intuitive and computationally light. Wing and colleagues show that the most basic stacked estimator does not in general identify any single average causal effect, because it applies different implicit weights to treatment and control trends, and that achieving a well-defined trimmed aggregate requires compositional balancing of the event window [\[46\]](#ref-46). The dissertation therefore reports a stacked estimate as a cross-check but does not rely on it for the headline number. Synthetic difference-in-differences with staggered timing is reported where the never-treated control pool is thin, because it reweights control units to match pre-treatment trends and can hold up better when parallel trends is doubtful, but it imposes its own structure and is less transparent about the group-time building block, so it too is confirmatory [\[51\]](#ref-51). The methodological reporting template follows recent applied exemplars that estimate a Callaway and Sant'Anna primary specification, supplement it with several robust estimators, and present the event-study profile as the central diagnostic, which is the standard form for a credible staggered-DiD paper [\[72\]](#ref-72), [\[73\]](#ref-73).

Confidence in the estimator choice is high. The evidence that TWFE is biased under staggered heterogeneous timing is settled and replicated [\[13\]](#ref-13), [\[15\]](#ref-15), [\[50\]](#ref-50), the Callaway and Sant'Anna estimator is a published, peer-reviewed solution with a maintained software implementation [\[14\]](#ref-14), and the design follows the same template as recent applied work [\[72\]](#ref-72), [\[73\]](#ref-73). What would lower this confidence is a demonstration that the reentry panel has so few treated cohorts that the group-time parameters are individually uninformative and only a pooled estimator is feasible; that contingency is addressed in the power analysis of Section 5.7 and in the underpowered-null rule, and it would change the precision of the estimate rather than the choice of estimator.

## 5.3 The specification written out

The estimand is built from the potential-outcomes notation fixed in the shared bible and carried verbatim here. The unit of analysis is the airspace-sector-by-time-window cell, indexed \(i\), observed over windows indexed \(t\). The outcome \(Y_{i,t}\) is one of the four cost measures, delay minutes, added flown distance, modeled fuel burn, or direct operating cost, each estimated in a separate specification so that the treatment effect is reported on each cost margin. Treatment is the binary reentry-driven AHA-exposure indicator: cell \(i\) is treated in the windows during which an AHA generated by a reentry event is active over that sector, together with the defined event window around the activation. The group variable \(G_i = g\) denotes the window in which cell \(i\) first receives a reentry closure, so that cells are sorted into cohorts by their first-treatment window. The never-treated potential outcome is \(Y_{i,t}(0)\), the outcome that cell \(i\) would have realized in window \(t\) had it never been exposed to a reentry closure.

The fundamental building block is the group-time average treatment effect on the treated:

\[
ATT(g,t) = E\big[\,Y_{i,t}(g) - Y_{i,t}(0)\ \big|\ G_i = g\,\big],
\]

the average difference, for cells whose first closure is in window \(g\), between their realized outcome and the outcome they would have had absent any closure, evaluated at window \(t\). Each \(ATT(g,t)\) is estimated by a difference-in-differences contrast that compares the change in outcome for cohort \(g\) from a pre-treatment base window to window \(t\) against the contemporaneous change for a clean control group, where the clean control group is the not-yet-treated cells (those with \(G_i > t\)) when a not-yet-treated pool is available, and the never-treated cells (those that receive no reentry closure in the sample) otherwise. The estimator is the doubly robust version of Callaway and Sant'Anna, which combines an outcome-regression model for the conditional expectation of the outcome change with an inverse-propensity weighting model for cohort membership given covariates, and is consistent if either model is correctly specified [\[14\]](#ref-14). Conditioning on covariates runs through these nuisance models rather than by entering covariates additively into a single regression, which is the design feature that allows the parallel-trends assumption to hold conditionally rather than unconditionally.

The group-time parameters are aggregated into the two objects the dissertation reports. The overall average treatment effect on the treated is a weighted average of the \(ATT(g,t)\) over all treated cohorts and all post-treatment windows, with weights proportional to cohort size, so that it answers the question "what was the average per-event cost of a reentry closure across the closures actually observed." This aggregated effect on the cost outcomes is the disruption parameter, the first of the two estimable quantities into which H1 decomposes. The event-study profile is a different aggregation: for each value of relative time \(e = t - g\), the time since closure activation, it averages the \(ATT(g, g+e)\) across cohorts, producing a coefficient at each lead and lag. The profile at negative \(e\) (the leads) is the pre-activation behavior and is the direct test of the no-anticipation and parallel-trends assumptions; the profile at non-negative \(e\) (the lags) traces how the cost rises at activation and decays afterward as displaced traffic recovers. The shape of the lag profile is itself a finding, because the network-propagation evidence implies that a closure's cost outlives its active window as delay cascades downstream, so a profile in which the post-activation effect persists beyond the closed window would indicate that the within-window cost understates the true per-event cost [\[12\]](#ref-12), [\[44\]](#ref-44).

The avoided-cost parameter, the second quantity into which H1 decomposes, is estimated by two distinct routes designed to triangulate. The first route is the static-versus-dynamic counterfactual difference. For each realized event, the static-closure cost is the realized cost recovered by the disruption estimate above; the dynamic-closure cost is the cost that the ReentryFlow model returns when it recomputes the closure under a prediction-informed dynamic policy that sizes the withheld volume to the uncertainty available shortly before reentry rather than to the worst-case envelope fixed hours in advance. The per-event avoided cost is the difference between these two, and the avoided-cost parameter is the average of that difference over events. The second route uses the prediction-uncertainty bound at closure-decision time as a continuous treatment intensity. Because closure size scales with prediction uncertainty, and because events differ in the uncertainty available when their closure was sized, the data contain natural variation in closure tightness; estimating the dose-response of cost to prediction uncertainty, holding traffic and geometry fixed, recovers the slope along which a reduction in uncertainty reduces cost, which is the same lever the dynamic regime pulls. The continuous-intensity estimate is developed in Chapter 6 with the Callaway, Goodman-Bacon, and Sant'Anna continuous-treatment extension and is mentioned here only to make clear that the avoided-cost parameter is identified by data variation and not solely by the simulator [\[47\]](#ref-47). That the two routes rest on partly different assumptions is the point: agreement between them raises confidence, and disagreement localizes the dependence on the ReentryFlow counterfactual.

## 5.4 The diagnostic estimators and what each reveals

The Goodman-Bacon decomposition and the de Chaisemartin and D'Haultfoeuille robust estimator are reported alongside the primary specification not as redundant alternatives but as instruments that each answer a question the primary estimate does not answer on its own. This section states what each one is for, because a robustness check that the reader cannot interpret is decoration rather than evidence.

The Goodman-Bacon decomposition answers a backward-looking question: had the analysis used the naive TWFE regression, how much of the resulting coefficient would have come from forbidden comparisons, and in which direction would they have pulled it. The decomposition theorem expresses the TWFE estimand as a variance-weighted average of every possible two-group, two-period difference-in-differences comparison in the panel, and it partitions those comparisons into three kinds: treated-versus-never-treated, treated-versus-not-yet-treated, and the problematic later-treated-versus-earlier-treated comparisons in which an already-treated cohort serves as a control [\[13\]](#ref-13). The third kind is the forbidden comparison, and its weight in the naive estimand, together with the sign of the effect it carries, is exactly what makes TWFE unreliable under heterogeneous timing. Reporting the decomposition for the reentry panel therefore does something the primary Callaway and Sant'Anna estimate cannot: it quantifies the size of the problem that the modern estimator was adopted to avoid. If the forbidden-comparison weight is large and the heterogeneity across cohorts is substantial, the decomposition makes visible that a TWFE estimate would have been materially misleading, which is the most legible possible justification for the entire estimator choice. If the forbidden-comparison weight is small, the decomposition reports that too, honestly, and the dissertation states that in such a case the modern and naive estimators would have agreed and the correction was a precaution rather than a rescue. Either result is informative, with the caveat that the decomposition characterizes the naive estimand's composition, not the truth of any assumption.

The de Chaisemartin and D'Haultfoeuille robust estimator answers a forward-looking question: does the per-event cost estimate survive when the aggregation logic is changed. This estimator is built on a different foundation from Callaway and Sant'Anna. It defines instantaneous and dynamic treatment effects through a different set of clean comparisons, it handles treatment that turns on and off (treatment reversal) natively, which matters because an AHA activates and then deactivates, and it is robust to heterogeneous effects without the negative-weighting pathology of TWFE [\[15\]](#ref-15). Because the AHA is a transient closure rather than an absorbing state, the treatment-reversal handling is not incidental: a sector is treated during the active window and reverts to untreated afterward, and an estimator that assumes treatment is absorbing would mishandle the post-window recovery. The de Chaisemartin and D'Haultfoeuille estimator accommodates this structure directly, so its agreement with the Callaway and Sant'Anna estimate is evidence that the result is not an artifact of how the primary estimator treats the closure's transience, and its disagreement would flag exactly that dependence. One caveat holds: both estimators share the conditional parallel-trends and no-anticipation assumptions, so their agreement is evidence against estimator-specific aggregation artifacts but not against a shared failure of the common identifying assumptions; the placebos of Section 5.6, not the cross-estimator comparison, are what test those.

The confidence attached to this diagnostic strategy is high. The two diagnostics rest on published, replicated results and maintained implementations, and the practice of reporting a primary group-time estimate with a Goodman-Bacon decomposition and a second robust estimator as cross-checks is the established template in recent applied staggered-DiD work [\[50\]](#ref-50), [\[72\]](#ref-72), [\[73\]](#ref-73). What would lower confidence is a setting in which the diagnostics could not be computed for want of clean comparisons, which is again a small-sample contingency addressed by the power analysis rather than a flaw in the diagnostic logic.

## 5.5 Identification assumptions argued formally

The causal interpretation of the disruption parameter rests on three assumptions, each of which is stated, given its mechanism, and defended with an explicit acknowledgment of what would break it.

The first and most consequential assumption is the conditional independence of reentry timing from contemporaneous sector-specific demand shocks: conditional on the covariate set, the window in which a sector first receives a reentry closure is as good as randomly assigned with respect to the outcome the sector would have had absent the closure. The mechanism that makes this plausible is physical: the reentry epoch of an uncontrolled object is set by orbital decay, which is driven by the object's ballistic coefficient, its orbital geometry, and the state of the upper atmosphere, none of which is responsive to the demand for the U.S. airspace sector that the object's footprint happens to intersect [\[21\]](#ref-21), [\[22\]](#ref-22), [\[55\]](#ref-55). A Long March 5B upper stage decays on a schedule fixed by drag physics; the fact that its dispersion footprint crosses a busy transatlantic sector on a Tuesday afternoon rather than a Sunday dawn is, to the resolution that matters for aviation demand, an accident of orbital mechanics and Earth rotation. This mechanism supports the independence assumption because an assignment driven by a process orthogonal to the outcome's determinants is, conditional on the determinants that are correlated with both (time of day, day of week, season, which are controlled), independent of the potential outcomes. The design-based tradition treats such physically-driven timing as a natural experiment and lets it carry the causal claim rather than a functional form [\[69\]](#ref-69).

The objection to this assumption must be stated and answered, because the identification stands or falls on it. The objection is that reentry timing is not in fact independent of demand: controlled reentries, and the AHA-activation decisions that follow even uncontrolled reentries, are made by human schedulers and regulators who may, consciously or not, place or time closures with reference to traffic. The response has three parts. First, for uncontrolled reentries the epoch is not chosen at all, so the objection applies only to the activation-decision margin, not the event-timing margin; the design conditions on the activation window and tests sensitivity to its definition. Second, for controlled reentries, where an operator does choose a reentry window, the design conditions on the published schedule and, as a sensitivity check, re-estimates the entire model excluding all controlled reentries, so that the headline estimate can be shown to rest on the uncontrolled events whose timing is unambiguously exogenous. Third, the placebo tests of Section 5.6, in particular the adjacent-sector placebo, provide an empirical check: if reentry closures were being timed into already-disrupted windows, the disruption would appear in sectors adjacent to the footprint as well as inside it, and a null effect in the adjacent placebo is evidence against the reverse-causation story. The assumption is therefore credibly satisfied for uncontrolled reentries unconditionally and for controlled reentries conditionally, and the dissertation reports the controlled-reentry-excluded estimate so the reader can judge how much of the result depends on the conditioning. Confidence in the assumption is moderate-to-high for the uncontrolled-only estimate and moderate for the full sample; what would raise it is a larger sample of uncontrolled events permitting the uncontrolled-only estimate to stand on its own with adequate precision, and what would lower it is evidence in the activation records that closures cluster non-randomly into high-demand windows.

The second assumption is conditional parallel trends: absent the reentry closure, the cost outcome in treated cells and in clean-control cells would have followed the same trajectory, conditional on the covariates. The mechanism is that, once time-of-day, day-of-week, season, baseline sector traffic density, concurrent weather, concurrent traffic-management initiatives, aircraft-class mix, and route distance are held fixed through the doubly robust nuisance models, the residual evolution of cost in a sector is driven by common shocks (national weather systems, network-wide delay) that affect treated and control sectors alike. The Callaway and Sant'Anna estimator identifies the group-time ATT under exactly this conditional parallel-trends assumption together with no anticipation [\[14\]](#ref-14). The assumption is testable rather than merely asserted because the event-study leads provide a direct, if imperfect, check: under parallel trends and no anticipation, the pre-activation coefficients should be statistically indistinguishable from zero, and a visible pre-trend is evidence that the assumption fails. The standard caution applies, that zero pre-trends are necessary but not sufficient for parallel trends in the post-period, so the leads are evidence rather than proof; the dissertation reports them and, following Roth, interprets the new-method event-study plots correctly, recognizing that the recent estimators construct pre-treatment coefficients asymmetrically from post-treatment ones and that the visual heuristics developed for TWFE plots do not transfer directly [\[49\]](#ref-49). The objection that a confounder trends differently in treated and control sectors in a way the covariates do not capture is answered by the weather placebo and the adjacent-sector placebo, which are designed precisely to detect a differential trend that the controls miss.

The third assumption is no anticipation: cells do not respond to a reentry closure before it activates. The mechanism by which this could fail is operational; if flight planners reroute in advance of a published closure, the pre-period outcome is already contaminated by the treatment, and the estimated effect at activation understates the total because part of the response has been pulled into the leads. This is not merely a nuisance but a substantive feature of airspace behavior, since closures are often published in advance and operators do plan around them. The design handles this in two ways that are stated here and executed in Chapter 6. First, the event-study leads are inspected directly for the signature of anticipation, a non-zero pre-activation effect; finding it relocates rather than eliminates the effect. Second, when anticipation is present, the treatment window is widened to begin at the publication time rather than the activation time, so that the pre-activation avoidance is captured inside the treated window and the aggregated effect recovers the total cost including the anticipatory component. Widening the window trades a cleaner no-anticipation assumption against a noisier treatment definition, and the dissertation reports the estimate under both the narrow and the widened window so the sensitivity is visible. Confidence that anticipation can be handled is high, because the remedy is a window redefinition that the data support; confidence that anticipation is absent under the narrow window is low, which is exactly why the widened-window estimate is reported as the preferred specification for the total-cost interpretation.

A fourth assumption is required specifically for the avoided-cost parameter when it is estimated by the continuous-intensity route, and it is more demanding than the binary-treatment assumptions, which is why the dissertation foregrounds it rather than burying it. The continuous treatment is the prediction-uncertainty bound at closure-decision time, and the dose-response interpretation, that a marginal reduction in prediction uncertainty causes a marginal reduction in cost, requires that the level of prediction uncertainty across events be conditionally independent of the potential cost outcomes. The mechanism that supports this is the physics of reentry prediction: the uncertainty available when a closure is sized is governed by the object's ballistic coefficient, the disturbance state of the upper atmosphere, the reentry angle, and how long before reentry the decision is made, and these drivers are not functions of the aviation demand in the affected sector [\[21\]](#ref-21), [\[22\]](#ref-22), [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55). An object with a poorly constrained ballistic coefficient, or one whose closure had to be sized during a geomagnetic storm that inflated thermospheric density uncertainty, carries a larger prediction envelope for reasons internal to orbital and atmospheric physics, not for reasons connected to how busy the airspace below it happens to be [\[82\]](#ref-82), [\[84\]](#ref-84). Conditional on the covariates that capture the demand side (traffic density, time-of-day, geometry), variation in prediction uncertainty is variation in an instrument-like physical quantity. The continuous-treatment difference-in-differences framework of Callaway, Goodman-Bacon, and Sant'Anna makes explicit that the dose-response comparison across treatment levels requires an assumption stronger than binary parallel trends, because parallel trends alone does not rule out selection into treatment intensity, and it supplies the alternative assumptions that eliminate that selection bias [\[47\]](#ref-47). One objection is that prediction uncertainty might correlate with event characteristics (object size, orbit class) that themselves correlate with which sectors are crossed, inducing selection; the response is to condition on those event characteristics where observed and to report the continuous-intensity estimate as corroborating rather than primary evidence for the avoided-cost parameter, with the static-versus-dynamic counterfactual route, which does not require the continuous-intensity exogeneity assumption, as the parallel estimate. This caution is firm and is the reason the avoided-cost parameter is identified by two routes: the continuous-intensity slope is causal only under the stronger no-selection-into-intensity assumption, so its agreement with the simulator-based counterfactual difference is what licenses the causal reading, and its confidence is moderate standing alone and higher in conjunction with the counterfactual route.

## 5.6 Threats to validity and their mitigations

The four classical validity threats are addressed in turn, each as a claim about a specific hazard paired with the design feature that confronts it.
### 5.6.1 Internal validity

The chief internal-validity threat is confounding by concurrent disruptions, principally weather, which also closes or constrains airspace and which dominates the delay record in absolute terms. The hazard is that a reentry closure coincides with a weather event and the design attributes the weather-driven cost to the closure. Three features confront this. First, concurrent weather severity enters the covariate set, so the doubly robust estimator conditions on it directly. Second, the identifying variation is the orbital-mechanics timing of reentry, which is independent of weather; over many events, weather coincidences average out in a way they would not if reentry timing tracked weather. Third, the weather placebo test re-runs the estimation using weather-only closures (sector-windows constrained by weather but with no reentry event) as a falsification sample: a design that correctly isolates the reentry effect should attribute no spurious reentry effect to weather-only windows. A second internal threat, the TWFE bias under staggered heterogeneous timing, is addressed by the estimator choice itself and is made visible by the Goodman-Bacon decomposition reported alongside [\[13\]](#ref-13), [\[14\]](#ref-14), [\[15\]](#ref-15). A third threat is anticipation, addressed in Section 5.4 through the lead inspection and the window widening. A fourth is reverse causation, that closures are scheduled into already-disrupted windows; the controlled-reentry exclusion and the adjacent-sector placebo address it. The mitigation for internal validity is therefore not a single device but a layered set: covariate conditioning, exogenous timing, falsification placebos, and the modern estimator, each catching a different failure mode.

### 5.6.2 External validity

The external-validity threat is that an estimate from the current U.S. reentry mix and cadence may not extrapolate to a future high-cadence down-mass regime in which reentries are frequent, controlled, and concentrated near specific recovery sites. This is the local-effect caution intrinsic to the design-based tradition: a credible causal estimate is local to the events and margins observed, and honesty requires saying so rather than presenting a local estimate as a universal constant. The design addresses this by changing the object that is reported, not by claiming external validity it cannot support. Rather than a single scalar per-event cost, the dissertation estimates the cost as a function of cadence and of prediction uncertainty, through the continuous-intensity specification and through the cohort-level heterogeneity that the group-time estimator exposes. Extrapolation to a future regime is then explicit and conditional: a reader who posits a future cadence and a future prediction-uncertainty level can read off the implied cost from the estimated function, with the caveat that the function is fitted over the observed support and that extrapolation beyond that support is the reader's assumption, not the dissertation's finding. This caution is firm. The estimated cost function is interpolative over the observed range of cadence and uncertainty, and it becomes progressively less credible the further a future scenario lies outside that range.

### 5.6.3 Construct validity

The construct-validity threat concerns the match between the measured proxies and the underlying constructs they stand for. The outcome construct is the economic cost of disruption, proxied by delay minutes, added distance, fuel burn, and direct operating cost. These proxies omit at least two cost components: the value of passenger time, which converts delay minutes into welfare loss at a rate the operating-cost factors do not capture, and the schedule-network knock-on cost, the downstream delay propagation that a localized disruption seeds across the network. The omission is acknowledged rather than ignored, and the network-propagation literature is used to bound the omitted downstream cost. The empirical characterization of how a primary delay magnifies and spreads through the U.S. air-transportation network supplies an upper-bound multiplier on the within-window cost, so the dissertation can report both the directly measured cost and a propagation-bounded total [\[12\]](#ref-12), [\[43\]](#ref-43), [\[44\]](#ref-44). The treatment construct, AHA exposure, is measured with error, because the footprint polygons that define exposure are model outputs from the EU SST prediction and from ReentryFlow rather than observed boundaries. This measurement error is not assumed away; the EU SST prediction-uncertainty bounds are propagated into the exposure measure, so that a cell near the footprint edge is assigned a probabilistic rather than a deterministic exposure, and the sensitivity of the estimate to the exposure threshold is reported. The randomized footprint-estimation method of Falsone and Prandini provides the methodological basis for treating the footprint as a probabilistic region rather than a hard boundary [\[36\]](#ref-36). The cost proxies bound the true cost from below and the propagation evidence bounds the omitted component from above, so the reported interval brackets the construct rather than claiming to measure it exactly.

### 5.6.4 Statistical-conclusion validity

The statistical-conclusion threat is that the inference understates uncertainty because the cells are spatially and temporally correlated, both because adjacent sectors share traffic and weather and because a sector's cost is serially correlated across windows. Treating cells as independent would produce standard errors that are too small and significance that is overstated. The mitigation is to cluster the inference at the airspace-sector level, which allows arbitrary within-sector correlation across windows, and, because the number of treated clusters is modest, to use the wild-cluster bootstrap for the small-sample correction rather than relying on the asymptotic cluster-robust variance, which is unreliable with few clusters. A second statistical-conclusion threat, and a serious one for this application, is low power: the number of U.S.-airspace-intersecting reentry events in the current record is small, so the design may fail to reject a false null simply because it cannot detect an effect of plausible size. This threat is addressed structurally by the minimum-detectable-effect analysis of Section 5.7 and by the standing rule, fixed in the contribution, that an underpowered null is inconclusive and never confirmation of H0. Statistical-conclusion validity therefore rests on inference that is honest about both the dependence structure and the power constraint, and on a design that pre-commits to treating an imprecise null as a non-finding rather than as evidence for the null.

## 5.7 Robustness battery and placebo tests

The robustness battery is specified in advance so that it cannot be assembled post hoc to rescue a preferred result. It comprises five families of checks, each targeting a specific assumption.

The weather placebo targets the weather-confounding threat. It re-estimates the full specification on a sample of sector-windows constrained by weather but containing no reentry event, assigning a synthetic "treatment" at the weather-constraint time. A correctly specified design returns a null reentry effect on this placebo sample, because there is no reentry to cause one; a non-null effect would indicate that the estimator is picking up weather-driven cost that it is mislabeling as reentry cost.

The adjacent-sector placebo targets both the reverse-causation threat and residual confounding. It assigns a placebo treatment to sectors that are spatially adjacent to a reentry footprint but lie outside it, during the same active window. Because these sectors are not closed, they should show no treatment effect; a measured effect would indicate either that the footprint is mis-drawn (and the adjacent sector is in fact partly affected, a measurement issue handled by the probabilistic exposure) or that some sector-level confounder correlated with proximity to reentry events is driving the result. The construction of realistic reroutes around blocked airspace, and the airspace-planning model that assigns surrogate trajectories under closure scenarios, supply the machinery for defining which adjacent sectors absorb displaced traffic and which do not, so the placebo is geometrically disciplined rather than arbitrary [\[31\]](#ref-31), [\[35\]](#ref-35).

The controlled-reentry exclusion targets the conditional-independence assumption at its weakest margin. It re-estimates the model dropping every controlled reentry, leaving only uncontrolled events whose timing is set by orbital decay and is therefore exogenous without conditioning. Agreement between the full-sample and uncontrolled-only estimates is evidence that the conditioning on controlled-reentry schedules is not doing the work; divergence localizes the dependence on that conditioning.

The anticipation-leads check targets the no-anticipation assumption. It is the inspection of the event-study leads described in Section 5.4, reported under both the narrow (activation-time) and widened (publication-time) treatment windows, with the dynamic-response counterfactual of the dynamic flight-path tool used to characterize how much pre-activation rerouting a rational operator would undertake, which calibrates how wide the window must be to capture it [\[79\]](#ref-79).

The alternative-cost-factors check targets construct validity in the outcome. It re-computes the direct operating cost and fuel burn under alternative per-block-hour and per-nautical-mile cost factors by aircraft class, and under alternative fuel-price assumptions, so that the dollar-denominated effect is shown to be robust to the cost-conversion parameters rather than an artifact of one cost schedule. Because the launch-era evidence shows that the impact distribution differs sharply between general aviation and international carriers, the cost-factor sensitivity is reported separately by aircraft class, which also tests whether the aggregate effect is driven disproportionately by one class [\[3\]](#ref-3), [\[32\]](#ref-32).

Each check is paired with a pre-stated interpretation: what result confirms the assumption and what result would force a revision. The battery is reported in full regardless of whether the headline estimate is significant, so that a reader can see the design's behavior under stress rather than only its behavior on the preferred specification.

## 5.8 Power and minimum-detectable-effect analysis

Power is a first-order concern for this design, not an afterthought, because the binding constraint is the number of U.S.-airspace-intersecting reentry events in the available record, which is modest. The claim of this section is that the design reports, in advance, the smallest treatment effect it could detect at the available sample size, and that it commits to reading an imprecise null as inconclusive. The reasoning follows standard statistical-power logic. The minimum detectable effect (MDE) at a given significance level and power is proportional to the standard error of the estimated ATT, which in a clustered design scales with the residual outcome variance, inversely with the square root of the number of treated clusters, and with the intraclass correlation that the sector-level clustering induces. With few treated sectors and serially correlated within-sector outcomes, the effective sample size is far below the raw count of cell-window observations, and the MDE is correspondingly larger than a naive calculation would suggest.

The power analysis proceeds in three steps, all design-stage and none executed on the full dataset. First, the design fixes the inference parameters: a two-sided test at the conventional five-percent level and eighty-percent power as the reporting standard, with the MDE also reported at ninety-percent power for transparency. Second, it parameterizes the variance components from the prior literature rather than from the unexecuted estimation. The residual variance of the delay-cost outcome and its intraclass correlation are seeded from the published delay-distribution and delay-propagation characterizations, which document the heavy-tailed, serially correlated structure of NAS delay [\[12\]](#ref-12), [\[40\]](#ref-40), [\[43\]](#ref-43), [\[77\]](#ref-77). These are priors on the variance, labeled as such, used only to size the MDE and not to claim a result. Third, it computes the MDE as a function of the number of treated events, producing a curve rather than a single number, so that the reader sees how the detectable effect shrinks as more events accumulate. The illustrative arithmetic of this curve is presented in Chapter 6 with every quantity left as a symbol or seeded from a labeled prior; here the design commits to the structure of the calculation and to its consequence.

The consequence is the underpowered-null rule, which is part of the dissertation's contribution and is restated here as a design commitment. If the estimated disruption effect is near zero but its confidence interval is wide enough to include economically meaningful effects, that is, wider than the MDE, the result is reported as inconclusive: the design could not have detected an effect of the size that the literature leads us to expect, so the failure to find one is uninformative about H0. H0 is retained as a substantive finding only if the disruption estimate is near zero and its confidence interval is tight enough to exclude economically meaningful effects, which requires the MDE check to pass. This asymmetry is deliberate, the safeguard against the most likely failure mode of a small-sample design, namely mistaking a wide null for evidence of no effect. The distinction is explicit: a precisely estimated null falsifies the first part of the contribution and is a publishable, decision-relevant result; an imprecise null falsifies nothing and signals that the design must wait for more events. Confidence that the power analysis correctly characterizes the design's sensitivity is high, because the calculation is standard and the variance priors are drawn from peer-reviewed delay characterizations; confidence that the available sample is adequately powered is, by the nature of the constraint, low, and the dissertation does not pretend otherwise.

## 5.9 Pre-registration commitment

The design commits to pre-registration, and the reason is exactly the reason pre-registration exists: a design with a modest sample, a two-part hypothesis, and a rich robustness battery offers many researcher degrees of freedom, and the credibility of the eventual estimate depends on those degrees of freedom being fixed before the outcome is seen. The pre-registration is therefore not a formality but the mechanism that makes the falsifiability claim honest.

The pre-registered document fixes, before any estimation on the full panel, the following elements. It fixes the primary estimator (Callaway and Sant'Anna doubly robust) and the diagnostic estimators (Goodman-Bacon decomposition, de Chaisemartin and D'Haultfoeuille robust estimator), so that the choice among frontier estimators cannot be made to favor a result [\[13\]](#ref-13), [\[14\]](#ref-14), [\[15\]](#ref-15). It fixes the four outcome variables, the binary treatment definition, the continuous treatment-intensity definition (the prediction-uncertainty bound at closure-decision time), the cohort definition (first-closure window), and the covariate set, so that no outcome or specification can be selected after the fact. It fixes the two aggregation targets (the overall ATT and the event-study profile) and the two estimation routes for the avoided-cost parameter (the static-versus-dynamic counterfactual difference and the continuous-intensity dose-response). It fixes the entire robustness battery of Section 5.6, with the pre-stated interpretation of each check, so that the placebos cannot be selectively reported. It fixes the inference procedure (sector-level clustering, wild-cluster bootstrap for the small-sample correction) and the power-reporting standard. And it fixes the decision rule on H0 and H1, including the underpowered-null rule, so that neither the rejection nor the retention of the null can be retrofitted to the realized estimate. The pre-registration distinguishes the confirmatory analyses, which are the pre-registered specifications above, from any exploratory analyses, which are permitted but labeled as exploratory and not used to adjudicate the hypotheses. Pre-registration constrains the analysis, not the data: it cannot fix a power deficit or a failed assumption, and the dissertation's commitment is to report the pre-registered analysis as run, including its failures, rather than to substitute a better-behaved post hoc specification.

## 5.10 Computational and software plan

The computational plan makes the design reproducible and names the tools, the pipeline, and the validation gates, so that the estimation is an executable procedure rather than a description.
The estimation uses maintained, peer-reviewed implementations of the frontier estimators. The group-time difference-in-differences is estimated with the Callaway and Sant'Anna estimator as implemented in its standard statistical package, which provides the doubly robust group-time ATT, the cohort and event-study aggregations, and the simultaneous-inference confidence bands for the event-study profile [\[14\]](#ref-14). The de Chaisemartin and D'Haultfoeuille robust estimator and the Goodman-Bacon decomposition run from their own maintained packages, so that the diagnostics are computed by their authors' implementations rather than by a re-implementation that might differ subtly [\[13\]](#ref-13), [\[15\]](#ref-15). The stacked and synthetic-difference-in-differences confirmatory estimates use their respective published implementations [\[46\]](#ref-46), [\[51\]](#ref-51). Inference uses a wild-cluster bootstrap implementation appropriate to few-cluster settings. The continuous-treatment-intensity estimate, developed in Chapter 6, uses the continuous-treatment difference-in-differences extension and its accompanying selection-bias diagnostics [\[47\]](#ref-47). Naming the implementations serves reproducibility: a reader with the same panel and the same package versions should recover the same estimates, and the pre-registration records the package versions, so that the computational environment is itself fixed.

The data pipeline is a five-stage process whose stages map onto the estimation procedure of Chapter 6 and whose intermediate products are validated before they pass downstream. The first stage assembles the panel by joining the EU SST reentry-event layer to the FAA NAS and SWIM flight layer through the ReentryFlow exposure mapping, producing the sector-by-window cells with treatment indicators, treatment intensities, outcomes, and covariates. The second stage is a validation gate, not an estimation step: ReentryFlow's predicted exposed-flight sets are compared against realized FAA exposure for past events, and precision and recall are reported. The counterfactual outputs of ReentryFlow are not trusted, and the avoided-cost estimation does not proceed, until this validation gate is passed, because an exposure-mapping instrument that mislabels which flights a closure touches would corrupt both the disruption estimate and the counterfactual. The third stage estimates the disruption parameter and runs the diagnostics. The fourth stage estimates the avoided-cost parameter by both routes. The fifth stage runs the robustness and placebo battery. The pipeline is built so that each stage's output is a versioned, inspectable artifact, and so that the validation gate at stage two is a hard stop rather than an advisory.

The software plan also commits to the data-handling discipline that the access constraints require. The EU SST reentry catalog is accessed through its reentry-bulletin feed under the program's vault credentials; the FAA SWIM feeds and NAS operations archives are accessed through their subscription channels; the four IAC-26 PRISMA reviews are read directly as project artifacts and supply prior ranges rather than panel observations; and ReentryFlow is run from its project repository as both the exposure-mapping instrument and the dynamic-closure simulator. The reentry-prediction surface that defines the treatment intensity is grounded in the published characterization of reentry-prediction uncertainty, its atmospheric-density, ballistic-coefficient, and reentry-angle drivers, and its tightening as the object nears reentry, so that the treatment-intensity variable is a measured quantity with a documented error model rather than a free parameter [\[21\]](#ref-21), [\[22\]](#ref-22), [\[53\]](#ref-53), [\[54\]](#ref-54), [\[55\]](#ref-55). One caution governs the computational plan: the panel does not yet exist. The plan describes the construction and the validation gates, and the dissertation is honest that the pipeline is specified and the instruments are accessed but the joined panel is not built, which is the design-stage posture the whole work maintains.

## 5.11 How this chapter advances the argument

This chapter establishes that the design addresses the causal mechanism, and it supplies the method on which the identification the contribution requires depends. That the cost is real and material is established by the literature, and the chapter does not re-argue it; it argues that the chosen design recovers the per-event cost and the avoided cost without the bias that would invalidate a naive estimator, and that it does so under assumptions that are stated, defended, and tested rather than assumed. The design dominates the alternatives in the specific sense that the Callaway and Sant'Anna estimator is provably unbiased under staggered heterogeneous timing where the TWFE estimator is provably biased, and the robustness battery demonstrates that the result does not depend on the choice among frontier estimators [\[13\]](#ref-13), [\[14\]](#ref-14), [\[15\]](#ref-15), [\[46\]](#ref-46), [\[50\]](#ref-50), [\[51\]](#ref-51). The residual risk is acceptable in the specific sense that every threat to validity is paired with a mitigation and that the most dangerous residual, low power, is met not by pretending it away but by the underpowered-null rule that refuses to read an imprecise null as confirmation of the null. The single caution carried through the chapter is the one the design-based tradition demands: the estimate is local to the observed events and is credible exactly to the degree that its identifying assumptions hold, and the dissertation's commitment is to report those assumptions' tests transparently rather than to claim a certainty the design cannot deliver. The next chapter turns this design into a fixed, pre-committed analysis plan with an explicit decision rule, so that neither the confirmation nor the falsification of the contribution can be retrofitted after the estimates are seen.


# Chapter 6: Analysis Plan and Expected Results

## 6.1 Chapter thesis and problem frame

A fixed, five-step estimation procedure paired with a decision rule committed in advance is what converts the design of Chapters 4 and 5 into a test that neither outcome can be retrofitted to confirm. The disruption parameter and the avoided-cost parameter are each given an estimation path, a sign that the literature leads us to expect, a mechanism that explains that sign, and a numerical threshold the executed estimate must clear before H1 may be declared to survive. The result tables that would carry the executed numbers are specified here in full and left empty by design, because this dissertation is presented at the design stage and no estimate has been run on the assembled panel. The contribution of this chapter is therefore not a finding. It is the demonstration that the finding, when it comes, will be interpretable as evidence rather than as the product of analytic latitude.

The problem this chapter addresses is specific and follows the shape carried through the dissertation. The current state of empirical practice in this corner of the airspace-economics literature is that quantification of launch and reentry airspace impact is done by simulation and historical-traffic accounting [\[2\]](#ref-2), [\[3\]](#ref-3), [\[4\]](#ref-4), [\[5\]](#ref-5), reported as descriptive counts of affected flights and modeled fuel-and-time penalties, with no pre-committed rule that distinguishes a confirming result from a disconfirming one. The desired state is a plan in which the estimand, the estimator, the controls, the diagnostics, and the threshold for each hypothesis are written down before estimation, so that the difference-in-differences machinery selected in Chapter 5 produces a number whose meaning was fixed before its value was known. The gap is that no prior study of reentry airspace cost has been pre-registered in this sense, and the modern staggered difference-in-differences literature warns precisely that flexible, post-hoc specification choices under heterogeneous treatment timing can manufacture the sign and significance an analyst expects to find [\[13\]](#ref-13), [\[15\]](#ref-15), [\[50\]](#ref-50). Leaving the gap open means that even a correctly specified estimate would be vulnerable to the charge that its design was tuned to its outcome, which would weaken exactly the regulatory and investment decisions the estimate is meant to inform. Closing the gap is the work of this chapter.

One standing discipline governs everything below. Every quantity that appears with a sign attached is labeled as expected, meaning it is the direction the mechanism and the prior literature lead us to anticipate, not a measured value. Every quantity that appears inside the illustrative simulation is labeled as illustrative, meaning it is a symbol or placeholder used to show the arithmetic of the design, not an estimate. The result tables in Section 6.8 are specified but unpopulated by design. This guardrail is restated here because Chapter 6 is where the temptation to present expectation as result is greatest.

## 6.2 The Five-Step Estimation Procedure

The estimation procedure has five steps, executed in order, with each step gating the next. The ordering is itself part of the pre-registration: a downstream step is not permitted to send the analyst back to re-specify an upstream step in a way that depends on the downstream result. The five steps are panel assembly, ReentryFlow validation, disruption-parameter estimation, avoided-cost estimation, and the robustness-and-placebo battery. This section specifies each.

### 6.2.1 Step one: panel assembly

The first step assembles the analysis panel by joining the EU SST reentry-event layer to the FAA NAS and SWIM flight layer through the ReentryFlow exposure mapping, producing the airspace-sector-by-time-window cells, indexed \(i\) over sectors and \(t\) over windows, that are the units of analysis fixed in the notation of the bible. Each cell carries a binary treatment indicator (whether a reentry-driven Aircraft Hazard Area, AHA, was active over that sector in that window), a treatment intensity (the prediction-uncertainty bound at the closure-decision time, drawn from the EU SST layer), the four outcome measures (delay minutes, added distance, fuel burn, and direct operating cost), and the covariate vector (baseline sector traffic density, time-of-day, day-of-week, season, concurrent weather severity, concurrent non-space traffic-management initiatives, aircraft-class mix, and the great-circle distance of the affected routes). The cohort variable \(G_i = g\) records the window in which cell \(i\) first receives a reentry closure, which is the object the Callaway and Sant'Anna estimator partitions on [\[14\]](#ref-14).

This join is feasible because the connecting instruments exist, even though they have not yet been run. The EU SST service issues per-event reentry predictions with uncertainty windows and footprint polygons, SWIM feeds and FAA operations archives carry flight-level trajectories and special-use and hazard-airspace activation records, and the Space Data Integrator platform was built by the FAA expressly to automate the situational-awareness pipeline that links a reentry vehicle's position and mission parameters to NAS operations [\[29\]](#ref-29). An exposure measure is constructible whenever a footprint polygon with a time window can be intersected with a filed-or-flown trajectory, which is the operation that the footprint-estimation literature formalizes probabilistically [\[36\]](#ref-36), [\[31\]](#ref-31). The join is described here, not built; the panel does not yet exist, and the construction in this step specifies how it will be made, not a report that it has been made. An examiner might object that the EU SST and SWIM record systems may prove non-conformable at the timestamp or polygon resolution required; the response is that conformability is exactly what step two tests before any causal estimate is trusted, so the risk is contained inside the procedure rather than assumed away.

### 6.2.2 Step two: ReentryFlow validation by precision and recall

The second step validates the ReentryFlow model as a measurement instrument before its outputs are used either as the exposure bridge or as the counterfactual simulator. ReentryFlow ingests a reentry trajectory and dispersion footprint, intersects it with an airspace traffic scenario, and returns a predicted affected-flight set and an associated cost. Its predicted exposed-flight sets are compared against the realized FAA exposure records for past events, and precision and recall are reported. Precision is the share of flights ReentryFlow predicts to be exposed that the FAA record confirms were exposed; recall is the share of flights the FAA record shows were exposed that ReentryFlow predicts. Only when these metrics clear a stated threshold is ReentryFlow trusted to generate the dynamic-closure counterfactual in step four.

This validation gate is load-bearing and is placed second on purpose. The mechanism that makes it necessary runs as follows: ReentryFlow is the deterministic mapping from a treatment (a reentry event and its footprint) to a predicted exposed-flight set, so any systematic error in that mapping propagates directly into the exposure measure that defines treatment, and from there into the disruption parameter. If the model over-predicts exposure, treated cells are contaminated with unexposed flights and the estimated cost is biased toward zero; if it under-predicts, genuinely exposed flights are mislabeled as controls and the parallel-trends comparison is corrupted. Validating precision and recall first detects both failure modes before they can enter the causal estimate. The confidence attached to the validation step itself is moderate at the design stage, because the realized-exposure ground truth depends on the completeness of the FAA activation records, which attribute reroute cause imperfectly. Evidence that would raise this confidence is a larger set of past reentry events with clean FAA exposure attribution; evidence that would lower it is a finding that the footprint polygons and the flight records disagree at the resolution the join requires.

### 6.2.3 Step three: disruption-parameter estimation

The third step estimates the disruption parameter, which is the aggregated average treatment effect on the treated (ATT) of a reentry-driven AHA closure on the cost outcomes. The primary estimator is the Callaway and Sant'Anna staggered difference-in-differences estimator, which defines the group-time average treatment effect

\[
ATT(g,t) = E\big[\,Y_{i,t}(g) - Y_{i,t}(0)\ \big|\ G_i = g\,\big],
\]

estimates each \(ATT(g,t)\) against not-yet-treated and never-treated control cells, and aggregates the estimates with valid weights into an overall ATT and into an event-study profile by time-since-closure [\[14\]](#ref-14). The disruption parameter is the aggregated effect on the four cost outcomes. Two diagnostics accompany the primary estimate and are reported regardless of whether they agree with it. The Goodman-Bacon decomposition expresses any two-way fixed-effects analogue as a weighted average of all possible two-group, two-period comparisons and exposes the forbidden comparisons that use already-treated cells as controls, which is the source of bias under staggered, heterogeneous timing [\[13\]](#ref-13). The de Chaisemartin and D'Haultfoeuille robust estimator provides a second, separately constructed estimate that does not rely on the same weighting and so functions as an independent check [\[15\]](#ref-15). The interpretation of the event-study leads and lags follows Roth's caution that the pre-treatment coefficients produced by the new methods are constructed asymmetrically from the post-treatment coefficients, so the visual heuristics developed for two-way fixed-effects event studies do not transfer unaltered [\[49\]](#ref-49).

The reason this estimator is primary rather than a conventional two-way fixed-effects regression is argued in full in Chapter 5 and is summarized here only to the extent that it shapes the analysis plan. The now-settled econometric result is that two-way fixed effects is biased under staggered timing and heterogeneous effects [\[13\]](#ref-13), [\[15\]](#ref-15), [\[50\]](#ref-50), and the Callaway and Sant'Anna construction avoids this by restricting comparisons to clean controls and aggregating with weights that cannot go negative [\[14\]](#ref-14), in the design-based tradition that lets the comparison structure rather than a functional form carry the causal claim [\[69\]](#ref-69), [\[71\]](#ref-71). This estimator buys unbiasedness at the cost of precision, because it discards the forbidden comparisons that a naive estimator would have used, so its standard errors will be wider and the power concern of Chapter 5 is sharpened, not relaxed, by the choice. Confidence that the estimator is the correct primary specification is high, because the methodological literature is unusually convergent on this point; confidence that it will yield a precisely estimated effect is only moderate and is conditioned on the sample size, which is addressed by the minimum-detectable-effect analysis carried over from Chapter 5.

### 6.2.4 Step four: avoided-cost estimation
The fourth step estimates the avoided-cost parameter by two routes designed to corroborate each other. The first route differences the realized static-closure cost against the ReentryFlow dynamic-closure counterfactual: for each real event, ReentryFlow recomputes the closure under a prediction-informed dynamic policy that sizes the withheld volume to the uncertainty available shortly before reentry rather than to a worst-case envelope fixed hours in advance, returns the implied cost, and the avoided cost per event is formed as the realized static cost minus the simulated dynamic cost, then averaged across events. The second route uses the prediction-uncertainty bound at closure-decision time as a continuous treatment intensity and estimates how the cost outcome scales with that intensity directly from the natural variation across events whose predictions were tighter or looser at the moment the closure was sized. Because the second route treats intensity as continuous, it is estimated under the difference-in-differences-with-continuous-treatment framework, which identifies treatment-on-the-treated parameters under a parallel-trends assumption analogous to the binary case but warns that comparing those parameters across intensity levels requires stronger assumptions that rule out selection on the intensity itself [\[47\]](#ref-47). The two routes are kept separate and reported separately. Agreement between a counterfactual-simulation estimate and a continuous-intensity estimate constitutes mutual corroboration, while disagreement is itself reported as a finding about the dependence of the avoided cost on the modeling route.

The mechanism that grounds the expected positive sign of this parameter is the central causal chain of the dissertation, applied to the avoided-cost lever. A reentry event with large prediction uncertainty drives a conservative static AHA sized to a worst-case dispersion envelope, which withholds a large airspace volume over a sector-time window, which forces exposed flights to delay, hold, or reroute. The avoided-cost lever acts on the first link: prediction uncertainty falls sharply as the object nears reentry, so a dynamic closure sized to the late-available uncertainty withholds a smaller volume and displaces fewer flights, and the cost difference between the two closure sizes is the avoided cost. The evidence for the premise that the late-available uncertainty is materially smaller is the short-term reentry-prediction literature, which documents that prediction error tightens as the object approaches reentry and decomposes the dominant uncertainty drivers of atmospheric density, ballistic coefficient, and reentry angle [\[22\]](#ref-22), [\[55\]](#ref-55), [\[53\]](#ref-53). The evidence for the premise that a dynamic closure can in fact exploit that tightening is the trajectory-accuracy-requirements line of work, which derives the predicted-trajectory accuracy that lets automation decide which flights clear a hazard area before activation, and the launch-coordination-center concept that begins to quantify how dynamic handling reduces the closure footprint relative to static policy [\[9\]](#ref-9), [\[33\]](#ref-33), [\[10\]](#ref-10), [\[28\]](#ref-28). The confidence in the positive sign is moderate rather than high, because all of this component evidence is launch-centric or analogue-based and there is no prior measured reentry avoided cost. This is recorded as an evidence gap and is the reason the parameter is labeled expected throughout.

### 6.2.5 Step five: the robustness-and-placebo battery

The fifth step runs the robustness-and-placebo battery specified in Chapter 5 and re-stated here as a fixed checklist so that it cannot be selectively reported. The battery comprises weather placebos (sectors and windows with comparable weather severity but no reentry closure, to confirm the effect does not appear where the mechanism is absent), adjacent-sector placebos (sectors bordering but outside the closure footprint, to confirm the effect localizes to the withheld volume), controlled-reentry exclusion (re-estimation dropping operator-scheduled reentries, whose timing is not set by orbital decay alone and could in principle correlate with airspace demand), anticipation leads (inspection of the event-study leads for pre-activation cost that would indicate flights rerouting before the closure activates), and alternative cost factors (re-computation of fuel burn and direct operating cost under different per-block-hour and per-nautical-mile assumptions by aircraft class). Each element of the battery is committed in advance, and the plan states that all elements are reported whether or not they are favorable to H1. Where the closure structure has spatial reach beyond the nominal sector, the spatial-spillover correction is applied, because treatment assigned along geographic boundaries biases a classical difference-in-differences estimate when the effect crosses the border, both by contaminating the control trend and by mixing own-treatment and neighbor-treatment effects [\[48\]](#ref-48). A synthetic difference-in-differences specification with staggered timing is reported as a further robustness check on the aggregated ATT, because it reweights controls to match pre-trends and provides a complementary identification route to the Callaway and Sant'Anna primary [\[51\]](#ref-51).

### 6.2.6 Outcome construction and inference within the procedure

Two operational matters cut across all five steps and are specified here so the procedure is reproducible rather than gestured at: how the four cost outcomes are constructed from the raw records, and how inference is conducted on the resulting estimates. Both are committed in advance because each carries a degree of analytic freedom that, if left open, would let an analyst tune the result after seeing it.

The four outcome measures are built in a fixed order from a fixed source hierarchy. Delay minutes are computed relative to two baselines, the published schedule and an undisrupted-routing counterfactual, and both are reported, because a schedule-relative delay absorbs ambient lateness that an undisrupted-routing-relative delay isolates the closure from. Added distance is computed as the difference between the flown or rerouted path and the filed or great-circle reference, taken from the SWIM trajectory records. Fuel burn is modeled from the added distance and the aircraft type using standard burn factors by aircraft class, and direct operating cost is modeled from the distance and time penalties using per-block-hour and per-nautical-mile cost factors by class. This construction is defensible because it is a documented, reproducible mapping from observable inputs rather than a discretionary scoring: the launch-era airspace-impact studies establish exactly this distance-to-fuel-to-cost translation as the standard accounting for a closure's penalty [\[2\]](#ref-2), [\[4\]](#ref-4), and holding the burn and cost factors fixed by class, rather than fitting them to the sample, removes the route by which outcome construction could be reverse-engineered to a desired sign. These factors are class averages and omit operator-specific cost structures, so the cost outcome is a standardized rather than a carrier-exact figure. The objection that the standardization understates high-cost operators is met by the alternative-cost-factors element of the robustness battery, which re-computes the outcomes under perturbed factors and reports the sensitivity. The mechanism that makes the reroute itself measurable, rather than assumed, is the realistic-reroute generation method that constructs the surrogate trajectories a flight would fly around a blocked airspace, which supplies the flown-path counterfactual that added distance is measured against [\[31\]](#ref-31), [\[35\]](#ref-35). Confidence in the outcome construction is high for the distance and delay measures, which derive directly from trajectory records, and moderate for the fuel and cost measures, which depend on the class-average factors.

Inference is conducted under the correlation structure the data impose rather than under an independence assumption the data violate. Cells in the same sector across adjacent windows are serially correlated, and cells in neighboring sectors in the same window are spatially correlated, so the plan clusters standard errors at the sector level as the primary inference and reports, where the number of treated sectors is small, the wild-cluster bootstrap as the small-sample correction. Clustering follows from the spatial and temporal dependence of the panel established in the data chapter: ignoring this dependence understates the standard errors and inflates the apparent significance, which would manufacture a confirming result for H1 out of correlated noise, and the staggered difference-in-differences methodological literature treats correct clustering as a precondition for valid inference under heterogeneous timing [\[50\]](#ref-50), [\[14\]](#ref-14). One consideration is decisive and is the same power concern that governs the decision rule: with a modest number of treated sectors, even correctly clustered inference will yield wide intervals, so the wild-cluster bootstrap is not a way to recover significance the data do not contain but a way to avoid overstating significance the data do not support. The operational consequence carried into Section 6.3 is that a near-zero estimate with a wide clustered interval is reported as underpowered and inconclusive, never as confirmation of H0. Confidence that the clustering scheme is the correct one is high; confidence that it will leave enough power to detect a moderate effect is the open empirical question the executed study resolves.

## 6.3 The Fixed Decision Rule on H0 and H1

The decision rule is the heart of the pre-registration and is stated so that it binds the interpretation of whatever the executed estimates turn out to be. It is reproduced from the contribution and hypotheses fixed in the prospectus and the bible, and it is not reworded in substance.

The null hypothesis, H0, is that spacecraft reentry and down-mass operations impose no measurable cost on the NAS, meaning that the average treatment effect of a reentry-driven AHA activation on exposed-flight cost is statistically indistinguishable from zero. The alternative, H1, is that reentry and down-mass operations impose a measurable, positive disruption cost on the NAS in delay, reroute, and closure, and that improved reentry prediction combined with dynamic airspace management yields a positive, statistically distinguishable avoided cost. H1 decomposes into the two estimable parameters: the disruption parameter (the aggregated ATT on the cost outcomes) and the avoided-cost parameter (the difference between static-closure and dynamic-closure cost).

The rule has three clauses. First, for the disruption part of H1 to survive, the executed disruption estimate must be positive, of non-trivial magnitude, and statistically distinguishable from zero under the Callaway and Sant'Anna primary specification, and it must remain so under the Goodman-Bacon and de Chaisemartin and D'Haultfoeuille diagnostics; a sign or significance that flips between the primary and the diagnostics is reported as an unresolved result, not as confirmation. Second, for the avoided-cost part of H1 to survive, the executed avoided-cost estimate must be positive and statistically distinguishable from zero under at least the counterfactual-differencing route, with the continuous-intensity route reported alongside as corroboration; an avoided cost that is positive under one route and null under the other is reported as route-dependent and downgrades the confidence in the parameter accordingly. Third, and most consequential for honesty, H0 is retained as a substantive finding only if the disruption estimate is near zero with a confidence interval tight enough to exclude economically meaningful effects, which requires the minimum-detectable-effect check to pass at the available sample size; a near-zero estimate whose confidence interval is too wide to exclude meaningful effects is reported as inconclusive and underpowered, never as confirmation of H0.

The asymmetry in the third clause is deliberate and is the rule that protects the design from the most common failure of a null-result interpretation. A failure to reject a null is not evidence for the null unless the test had the power to detect the effect it failed to find. The mechanism by which this matters here is concrete: the number of U.S.-airspace-intersecting reentry events in the current record is modest, so the staggered estimator, which already discards forbidden comparisons, may return a wide confidence interval around a small point estimate. Under a naive reading, that wide interval would be misread as support for H0. The decision rule forecloses that reading by requiring the minimum-detectable-effect threshold to be met before any null is called substantive. Confidence that this rule correctly disciplines the inference is high, because it follows directly from the logic of statistical power. What remains uncertain at the design stage is whether the sample will in fact deliver the power the rule demands, which is an empirical question the executed study will answer and which the chapter does not pretend to resolve.

## 6.4 Expected Signs with Detailed Mechanism Reasoning

This section states the signs the design leads us to expect for each parameter, attaches the mechanism that explains each sign, and labels every expectation as expected rather than estimated. The purpose is to make the predictions explicit and falsifiable, so that an executed estimate of the opposite sign or of indistinguishable-from-zero magnitude would be recognized immediately as disconfirming rather than absorbed by reinterpretation.

### 6.4.1 Expected sign of the disruption parameter

The expected sign of the disruption parameter is positive: a reentry-driven AHA closure raises the cost borne by exposed flights, measured in delay minutes, added distance, fuel burn, and direct operating cost. The launch-era record supports this. Every quantified study of launch and reentry airspace closure reports that the closure forces affected flights to reroute, which increases flown distance, delay, and fuel burn relative to the undisrupted route [\[2\]](#ref-2), [\[3\]](#ref-3), [\[4\]](#ref-4), [\[5\]](#ref-5), [\[30\]](#ref-30). The reasoning is mechanical and points in one direction only: a withheld volume of airspace removes route options, and removing route options can only weakly increase, never decrease, the cost of the constrained routing problem, so larger withheld volume, longer window, and denser displaced traffic each push the cost up. The airspace-planning and collaborative-decision-making formalization reinforces this, since a set of flights must be assigned to surrogate trajectories around a restriction and the objective accumulates rerouting, workload, and equity costs that rise with the restriction [\[35\]](#ref-35). The magnitude, not the sign, is uncertain, and the cost is expected to concentrate in a minority of exposed flights rather than spread evenly, which is addressed in Section 6.4.3. An examiner might object that operators could fully anticipate and pre-absorb the closure so that no marginal cost appears at activation; the response is that the anticipation-leads element of the placebo battery tests exactly this, and that finding pre-activation cost would relocate the effect into the pre-period rather than eliminate it. Confidence in the positive sign is high, because every mechanism in the literature points the same way and no credible mechanism points the other way; confidence in any particular magnitude is low at the design stage.

### 6.4.2 Expected sign of the avoided-cost parameter

The expected sign of the avoided-cost parameter is positive, meaning that a prediction-informed dynamic closure costs less than a static closure for the same event. This follows from the two-part empirical premise established in step four: prediction uncertainty falls as the object nears reentry [\[22\]](#ref-22), [\[55\]](#ref-55), [\[53\]](#ref-53), and a dynamic closure can exploit that fall to withhold a smaller volume [\[9\]](#ref-9), [\[33\]](#ref-33), [\[10\]](#ref-10), [\[28\]](#ref-28). Because closure cost is increasing in withheld volume, established in Section 6.4.1, a smaller withheld volume yields a smaller cost, and the difference is the avoided cost. The dynamic-airspace-management literature reinforces this, demonstrating in the launch case that real-time situational information can be used to minimize the impact of a space operation on air traffic while preserving an acceptable safety level [\[79\]](#ref-79), as does the minimizing-disruption analysis that reframes the spatial closure problem as a temporal one and argues that any given location is at risk for no more than about one minute, which is the conceptual basis for a tightly time-bounded dynamic closure [\[37\]](#ref-37). The strongest caution in the chapter attaches here: there is no prior measured reentry avoided cost, the component evidence is launch-centric or analogue-based, and the parameter is therefore labeled expected with explicit acknowledgment of the evidence gap. One objection is that a dynamic closure might trade a smaller spatial footprint for a larger residual ground or air risk, so that the cost saving is illusory once safety is held equal; the response is that the narrowing is bounded by validated prediction accuracy, so the avoided cost is defined at equal safety, not at degraded safety. Confidence in the positive sign is moderate, lower than for the disruption parameter, precisely because the supporting evidence is one analogical step removed from the reentry case.

### 6.4.3 Expected distribution of the cost across flights and aircraft classes

The expected distribution of the cost is concentrated, not uniform, and differs by aircraft class: a minority of exposed flights bears most of the cost, and the impact share differs sharply between general aviation and international carriers. The launch-era simulation evidence supports this. In a four-dimensional closure analysis at Cape Canaveral, international carriers accounted for on the order of 8 to 10 percent of impacted flights and general aviation represented roughly one third, with the precise figures sensitive to site, geometry, and timing [\[3\]](#ref-3), [\[32\]](#ref-32), [\[2\]](#ref-2). Exposure is determined by the geometric and temporal intersection of a flight's route and schedule with the closure, which is heterogeneous across the traffic population, so the cost is distributed according to who happens to be routed through the withheld volume during the active window rather than spread across all flights. The broader delay-economics result reinforces this, finding that disruption cost is unevenly borne and that a localized capacity shock concentrates on the subset of operations that must absorb it [\[11\]](#ref-11). These percentages are launch-derived priors imported to set expectations, not reentry-measured shares, and the general-aviation figure in particular is uncertain because visual-flight-rules general aviation is under-represented in the instrument-flight-rules-complete FAA records, which biases the observed share downward relative to the true one. The operational consequence retained for Chapter 7 is that the aircraft-class mix is kept as a covariate and the impact distribution is reported by class, because a policy instrument that prices the externality must know whose cost it is pricing. Confidence in the qualitative pattern (concentrated, class-differentiated) is high; confidence in the specific launch-era percentages transferring to reentry is low.

### 6.4.4 Expected behavior of the diagnostics relative to the primary estimate

A prediction is owed not only about the parameters but about how the diagnostics should behave relative to the primary estimate, because the diagnostics are what convert a single number into a defended one, and stating in advance what their agreement or disagreement would mean prevents a post-hoc choice of which estimator to believe. The expectation is that the Goodman-Bacon decomposition and the de Chaisemartin and D'Haultfoeuille robust estimator broadly agree in sign with the Callaway and Sant'Anna primary, and that any divergence is in the direction the methodological literature predicts rather than in an arbitrary direction. This follows from the structure of the bias each diagnostic isolates: the Goodman-Bacon decomposition exposes the forbidden comparisons in which already-treated cells serve as controls, and under staggered timing with effects that grow over the post-period those comparisons drag a naive estimate toward zero or even reverse its sign [\[13\]](#ref-13). The implication is directional: if the primary estimate is positive and the naive two-way-fixed-effects analogue is smaller or wrong-signed, the decomposition should attribute the gap to the forbidden comparisons, and that attribution is a confirmation of the primary, not a challenge to it. The de Chaisemartin and D'Haultfoeuille estimator, constructed without the same weighting, should land near the primary if the heterogeneity is moderate and should diverge in a quantifiable way if the heterogeneity is severe; either outcome is informative, and the plan reports the divergence rather than suppressing it [\[15\]](#ref-15), [\[50\]](#ref-50). The applied-exemplar practice reinforces this, reporting a doubly robust primary alongside a panel of robust estimators and reading their convergence as the credibility check it is [\[73\]](#ref-73), [\[72\]](#ref-72). The diagnostics are not independent oracles: they share the same panel and the same exposure measure, so a flaw in the panel assembly or the ReentryFlow exposure mapping would propagate into all three estimators alike, which is precisely why the step-two validation gate is upstream of the estimation. An examiner might object that broad agreement among the three could reflect a shared bias rather than a robust truth; the response is that the placebo battery, which probes the exposure measure against weather and adjacent-sector falsification, is the independent check that the shared-bias concern requires, so the credibility of the estimate rests on the conjunction of estimator agreement and placebo survival, not on estimator agreement alone. Confidence that the diagnostics will behave as predicted is moderate-to-high, conditioned on the panel passing the validation gate; what cannot be predicted in advance is the magnitude of any divergence, which is itself one of the reported quantities.

## 6.5 Design of the Illustrative Simulation

The illustrative simulation exists to show the arithmetic of the avoided-cost estimate without claiming any of its inputs as measured. Every quantity in it is left as a symbol, and the purpose is pedagogical: to make the structure of the calculation auditable in advance, so that when the executed estimate fills the symbols with values, the reader already knows the formula those values enter.
The simulation proceeds as follows. Let a static closure withhold a volume sized to a worst-case dispersion envelope fixed at a decision time several hours before reentry, and let the along-track prediction uncertainty at that decision time be denoted by a symbol that drives the footprint length. Let a prediction-informed dynamic closure instead withhold a volume sized to the uncertainty available at a later decision time shortly before reentry, denoted by a second, smaller symbol. The footprint length, and with it the withheld area and the population of displaced flights, scales with the along-track uncertainty. If the uncertainty falls by some fraction between the two decision times, the withheld area and the displaced-flight population fall in rough proportion. The realized static-closure cost for the event is the cost computed over the static-closure displaced-flight set; the simulated dynamic-closure cost is the cost computed over the dynamic-closure displaced-flight set; and the avoided cost per event is the first minus the second. The program-level avoided cost is the per-event avoided cost multiplied by the projected reentry cadence over the planning horizon. None of these symbols is assigned a number here, because the dissertation does not yet have the executed estimates to assign, and assigning illustrative numbers would risk their being mistaken for results.

The one external figure that enters this section enters strictly as an order-of-magnitude analogue, not as a finding. The IAC-26 governance review reports digital-licensing processing-time reductions on the order of 67 to 84 percent from process and modeling improvements in the authorization pipeline. That figure anchors a prior for what process and modeling improvement can plausibly achieve when applied to a different target, the closure footprint, and it is an analogue from licensing-paperwork time to airspace-footprint size, not a measured reduction in closure cost. The analogue is admissible because both quantities are driven by the resolution of an uncertainty, the authorization decision in one case and the reentry prediction in the other, so that tightening the uncertainty shrinks the conservative margin in both. The analogue is imperfect, and remains a prior rather than an estimate, because the elasticities of the two quantities to their respective uncertainties are not the same and have not been jointly measured. Confidence in the analogue as a loose order-of-magnitude anchor is low-to-moderate; it would be raised by a direct measurement of closure-footprint elasticity to prediction uncertainty, which is what the continuous-intensity route of step four is designed to produce.

The simulation is structured to be reported as a surface rather than a point, and this structural choice disciplines the external-validity claim of the whole dissertation. The program-level avoided cost is the product of two quantities that the executed study will estimate separately: the per-event avoided cost, which depends on how far prediction uncertainty falls between the static and dynamic decision times, and the reentry cadence over the planning horizon, which is a projection rather than a measured quantity. Because the cadence is projected and uncertain, the simulation does not collapse to a single program-level number; it is reported across a grid of cadence assumptions and a grid of prediction-uncertainty-reduction fractions, so that the program-level cost is a function of those two inputs rather than a scalar that conceals its own dependence. This is the honest representation because both inputs are genuinely uncertain at the design stage: the cadence depends on large-constellation deorbit rates and the maturation of a commercial down-mass sector, and the uncertainty-reduction fraction depends on the still-maturing dynamic-closure capability whose components are documented for launch but not yet measured for reentry [\[37\]](#ref-37), [\[9\]](#ref-9). The operational consequence carried into Chapter 7 is that the dissertation reports the cost as a conditional surface and states its extrapolation to a future high-cadence regime explicitly, rather than asserting a single figure whose apparent precision would misrepresent the design-stage evidence grade. Confidence that the surface representation is the correct one is high, because it follows from the fact that the inputs are projections; confidence in any single cell of the surface is no higher than the confidence in the two projections that define that cell, which is the property the labeling is meant to preserve.

## 6.6 Continuous-Treatment and Spillover Handling

The intensity estimate and the spatial structure of closures require two methodological adaptations beyond the binary staggered estimator, and both are specified here so that the analysis plan is complete rather than gestured at. This section argues that treating prediction uncertainty as a continuous intensity and treating the closure as a spatially extended object are not optional refinements; they are required by the structure of the data, and omitting either would bias the avoided-cost estimate in a known direction.

The continuous-treatment adaptation concerns the second avoided-cost route. When prediction uncertainty is used as a continuous treatment intensity, the relevant estimand is a dose-response relationship between intensity and cost, not a single binary contrast. The Callaway, Goodman-Bacon, and Sant'Anna continuous-treatment framework shows that a treatment-on-the-treated-type parameter is identified under a parallel-trends assumption analogous to the binary case, but that comparing the parameter across different intensity levels is not innocuous, because parallel trends does not by itself rule out selection bias in which units with different intensities were on different trends for reasons related to the intensity [\[47\]](#ref-47). The operational consequence for this plan is that the continuous-intensity estimate is reported as a level-specific dose response with its identifying assumption stated, and that cross-level comparisons are made only under the stronger, explicitly flagged assumptions the framework requires; the de Chaisemartin and D'Haultfoeuille intertemporal estimator is reported alongside as the dynamic-effects complement appropriate when the intensity varies over time within the event window [\[52\]](#ref-52). Where the heterogeneity of the cost response across the traffic distribution is of interest, the quantile-treatment-effect extension for staggered adoption provides the appropriate estimand, because the concentrated, class-differentiated cost distribution expected in Section 6.4.3 implies that the mean ATT understates the cost in the upper tail of affected flights [\[70\]](#ref-70).

The spillover adaptation concerns the spatial structure of the closure. A reentry AHA is assigned over a geographic sector, and its effect can cross the sector boundary when displaced flights reroute into adjacent sectors, raising cost there. The spatial-spillover decomposition shows that under such cross-border effects a classical difference-in-differences estimate is biased in two ways: the control sectors no longer identify the counterfactual trend because their outcomes are themselves affected by the treatment, and the treated-sector change mixes the effect of own treatment with the effect of nearby treatment [\[48\]](#ref-48). The operational consequence is that the adjacent-sector placebo in the robustness battery does double duty: it is a falsification test (the effect should localize to the withheld volume) and a spillover diagnostic (if adjacent sectors show cost, that cost is the spillover the decomposition warns about, and it is added back into the total rather than discarded). Spillover is the expected direction of leakage because of the network-propagation result carried from Chapter 2: a localized capacity shock displaces traffic that then congests neighboring nodes, so the closure's cost does not stop at the closure's edge [\[12\]](#ref-12). Confidence that spillover must be handled is high, because the displacement mechanism is well established; confidence in its magnitude is low and is one of the quantities the executed adjacent-sector analysis will estimate.

The reporting template for the staggered, intensity-aware design follows applied exemplars that present a Callaway and Sant'Anna primary with robust-estimator corroboration and an event-study profile. A staggered difference-in-differences analysis of prosecutor policy and crime deterrence and a cross-country staggered analysis of national AI strategies each demonstrate the reporting shape this dissertation adopts: a doubly robust primary estimate, a panel of additional robust estimators, an event-study plot with leads and lags, and an explicit statement of the identifying assumption [\[72\]](#ref-72), [\[73\]](#ref-73). These are cited as templates for the form of the reporting, not as substantive findings imported into the reentry case; their relevance is methodological, showing that the staggered-design reporting standard this chapter commits to is the standard the current applied literature uses.

## 6.7 Event-Study Profile Interpretation

The event-study profile is the second principal output of the disruption-parameter estimation, and its interpretation is fixed in advance here because the shape of the profile, not only its aggregated height, carries evidence. The profile plots the treatment effect by time relative to closure activation, with negative time indexing the leads (windows before activation) and positive time indexing the lags (windows after activation).

A credible result has a specific expected shape, and each feature of that shape is a testable prediction. The leads should be near zero, indicating no anticipation and no pre-trend; a jump should appear at activation, indicating the closure's onset; and the lags should decay back toward zero as displaced traffic recovers. The near-zero leads follow from the fact that reentry timing for uncontrolled objects is set by orbital decay and is not chosen with reference to airspace demand, so there should be no systematic cost movement in the windows before a closure that the affected flights had no reason to anticipate. The jump follows from the abrupt withholding of the airspace volume at activation. The decay follows from the recovery of the displaced traffic to its normal routing once the volume is released. Confidence that this is the credible shape is high, because it follows from the identification logic and the mechanism; the value of stating it in advance is that departures from it are diagnostic.

Two departures are given fixed interpretations. First, if the post-activation effect persists well beyond the active closure window rather than decaying promptly, that persistence is read as evidence that the within-window cost understates the true cost, because the closure's effect has cascaded downstream through the network. The backing for this reading is the delay-propagation literature, which establishes that a primary delay magnifies and propagates through the schedule and the network, so that a localized shock outlives its immediate window [\[12\]](#ref-12), [\[43\]](#ref-43), [\[44\]](#ref-44). The operational consequence is that a persistent lag profile raises the estimated per-event NAS cost relative to a within-window-only accounting, and the plan commits to reporting both the within-window effect and the cumulative effect over the full lag horizon so that the network-propagation component is visible rather than buried. Second, if the leads are non-zero, that is read as anticipation: flights rerouting before the closure activates, which contaminates the pre-period and relocates rather than eliminates part of the effect. The operational consequence is that the treatment window is widened to capture the pre-activation avoidance, and the anticipation is reported as part of the cost rather than discarded as a pre-trend violation. Both interpretations carry one caution: the asymmetric construction of leads and lags in the modern estimators means the lead coefficients must be read with Roth's caution and not against the two-way-fixed-effects visual heuristic [\[49\]](#ref-49). Confidence in these two conditional interpretations is high, because each follows from an established mechanism and each is committed before the profile is seen.

## 6.8 Specified-but-Unpopulated Result Tables

The result tables are specified here in full structure and left empty by design. This is the most explicit expression of the design-stage guardrail: the tables that will carry the executed numbers exist as templates, their rows and columns are fixed, and their cells are blank because no estimate has been run on the assembled panel. Populating them with illustrative numbers would invite their being read as results, so they are left unpopulated. Their templates appear in Appendix D of the backmatter; their structure is described here so the analysis plan is self-contained.

Table 6.1, the disruption-parameter table, is specified with one row per cost outcome (delay minutes, added distance, fuel burn, direct operating cost) and columns for the Callaway and Sant'Anna aggregated ATT point estimate, its clustered standard error, its confidence interval, the de Chaisemartin and D'Haultfoeuille robust estimate, and the Goodman-Bacon decomposition summary. Every cell is blank. The table is constructed so that the decision rule of Section 6.3 can be applied to it directly: the first-row, first-column cell, when filled, is the headline per-event delay cost, and its confidence interval is the quantity the minimum-detectable-effect check evaluates.

Table 6.2, the avoided-cost-parameter table, is specified with one row for the counterfactual-differencing route and one row for the continuous-intensity route, and columns for the point estimate, the standard error, the confidence interval, and the per-event-versus-program-level scaling. Every cell is blank. The two rows are placed adjacent so that the corroboration-or-divergence reading of Section 6.2.4 is immediate when the cells are filled.

Table 6.3, the event-study coefficient table, is specified with one row per time-relative-to-activation index spanning the leads and the lags, and columns for the coefficient and its confidence interval, structured to be read against the credible-shape prediction of Section 6.7. Every cell is blank.

Table 6.4, the robustness-and-placebo summary, is specified with one row per battery element (weather placebo, adjacent-sector placebo, controlled-reentry exclusion, anticipation leads, alternative cost factors, spatial-spillover correction, synthetic difference-in-differences) and columns recording whether the element was favorable, neutral, or unfavorable to H1 and the re-estimated headline effect under that element. Every cell is blank, and the plan commits to filling every row regardless of outcome, which is the mechanism that prevents selective reporting.

The empty tables are themselves a contribution, rather than an absence to be apologized for, because they fix the analytic surface before the data are seen. Pre-specifying the table structure removes the degree of freedom by which an analyst chooses, after seeing the estimates, which outcomes and which robustness checks to feature [\[50\]](#ref-50), and a result whose presentation format was committed before its value was known cannot have been formatted to flatter the value. Pre-specification disciplines presentation, not estimation quality; an underpowered sample remains underpowered no matter how carefully its tables are pre-registered, which is why the power concern is carried as a standing limitation rather than resolved by the table design. Confidence that the empty-table commitment strengthens the eventual inference is high; it is one of the few claims in the chapter that does not depend on the data, because it is a claim about analytic procedure rather than about the world.

## 6.9 Chapter synthesis

This chapter has a specific place in the argument the dissertation carries from claim to claim, and naming that place clarifies what Chapter 6 does and does not establish. The argument has five parts: the cost is real, the cost is material, the intervention acts on the mechanism, the intervention beats the alternatives, and the residual risk is acceptable. Chapters 3 and 4 establish that the cost is real and material from the descriptive and simulation literature. Chapter 5 establishes the identification by which the materiality can be measured causally. Chapter 6 does not add a new claim; it specifies the test that will, when executed, convert the third and fourth from expectation into measurement. The disruption-parameter estimate turns the materiality claim into a number; the avoided-cost-parameter estimate turns the intervention-acts-on-the-mechanism and intervention-beats-the-alternatives claims into a number; and the residual-risk claim is what bounds the avoided cost at equal safety, which is why the avoided cost is defined against an equal-safety dynamic closure and not against a riskier one.

The synthesis of the chapter is therefore a single disciplined commitment. The procedure is fixed in five gated steps, the decision rule is committed in advance with the power-asymmetry clause that prevents a null from being over-read, the expected signs are stated with their mechanisms and their confidence grades so that disconfirming estimates would be recognized as such, the illustrative simulation shows the arithmetic without claiming any input as measured, the continuous-treatment and spillover adaptations are specified because the data structure requires them, the event-study profile has its credible shape and its two diagnostic departures fixed in advance, and the result tables are specified and left empty. Nothing in this chapter is a result. Everything in it is the apparatus that makes a result interpretable. The value of that apparatus is independent of which way the executed estimates fall: a positive, significant disruption parameter and a positive avoided cost would confirm H1 and supply the externality price that the regulatory and investment decisions of Chapter 7 require, while a precisely estimated near-zero disruption parameter would falsify the contribution and be, equally, a publishable and decision-relevant result. The design is built so that both outcomes are informative, which is the mark of a falsifiable plan rather than a foregone conclusion. This property is guaranteed by the conjunction of the pre-committed decision rule and the empty pre-specified tables: together they fix what a confirming and a disconfirming result each look like before either is observed, and that fixing is what this chapter contributes to the dissertation as a whole.


# Chapter 7: Discussion

## 7.1 Chapter thesis

The value of this dissertation to NASA, to the Jet Propulsion Laboratory, and to the wider civil-space enterprise does not depend on which way the estimate lands. Whether the executed disruption parameter turns out large or small, and whether the avoided-cost parameter turns out generous or thin, the contribution is the same: a defensible, design-based price for the cost that spacecraft reentry imposes on the National Airspace System, and a defensible number for the cost that better reentry prediction combined with dynamic airspace management would avoid. That price, and that avoided cost, are the inputs a regulator needs to weigh a reentry cadence against its aviation burden, and the inputs a prediction-and-modeling investment such as ReentryFlow needs to know what avoided cost it must clear to be worth funding. The dissertation is built so that a near-zero, precisely estimated effect would falsify the contribution rather than embarrass it; that symmetry is the source of the value, because a number that can be wrong is a number a decision-maker can use.
This chapter develops that thesis along five lines. It interprets the implications of the design under both outcomes, the world in which the alternative hypothesis H1 holds and the world in which the null H0 holds, treating the second as a genuine empirical possibility rather than a formality. It traces the theoretical contribution back to each of the three frameworks that organize the dissertation, showing what the measurement returns to Rao's account of orbital-economy externalities, to North's account of institutions and transaction costs, and to the design-based identification tradition of Angrist and Pischke. It draws out the policy and mission implications for the named stakeholders, anchored in the live NASA Langley airspace-modeling line and the civil-space enterprise's interest in pricing the reentry externality. It engages the rival explanations in full, naming for each the specific feature of the research design that confronts it. It states the external-validity position honestly, locating the estimate where the design-based tradition insists it belongs: local to the cadence and event mix observed, reported as a function of cadence and prediction uncertainty so that any extrapolation is conditional and explicit rather than assumed. The design-stage discipline of the dissertation holds throughout. No estimate is claimed on the full dataset; every number that appears is labeled expected or illustrative, and the discussion concerns what the design is built to establish, not what it has established.

## 7.2 The problem this chapter addresses

The problem this chapter must resolve is interpretive rather than empirical. The current state of the argument, carried forward from Chapters 4 through 6, is a fully specified design with two estimable parameters, a fixed decision rule, and a pre-committed account of what each parameter must show for the alternative hypothesis to survive. The desired state is a clear, decision-relevant reading of what follows for the science, the policy, and the stakeholders under each admissible outcome, with rival explanations confronted and the reach of the estimate stated honestly. The gap between those two states is the work of interpretation: a design, however careful, does not by itself say what its results would mean, which rivals would remain standing, or how far its number would travel. Leaving that gap open yields a dissertation that can estimate a parameter but cannot tell a regulator or a program office what to do with it, forfeiting the practical point of having built a falsifiable measurement in the first place. This chapter closes the gap by reading the design through both outcomes, both theory and policy, and the full set of competing explanations, so that the executed study, whatever it returns, lands as evidence a decision-maker can act on.

## 7.3 Implications if H1 holds

Begin with the world in which the design returns the expected signs. If H1 holds, the dissertation supplies two operationally distinct quantities, each of which converts a standing policy ambiguity into a priced decision. This is the expected outcome because every mechanism in the airspace-disruption record points the same direction, with larger withheld volumes, longer windows, and denser displaced traffic each raising the cost of a closure [\[2\]](#ref-2), [\[3\]](#ref-3), [\[4\]](#ref-4), [\[5\]](#ref-5), [\[1\]](#ref-1). The causal chain fixed in the design carries the expectation: a reentry event carrying large prediction uncertainty forces a conservative static Aircraft Hazard Area sized to a worst-case dispersion envelope, that envelope withholds a large airspace volume over a sector-time window, exposed flights respond by delaying, holding, or rerouting, and those responses register as added delay minutes, added distance, fuel burn, and direct operating cost, with network propagation extending the cost beyond the closed window [\[12\]](#ref-12). The convergent simulation and historical-traffic evidence shows that closures touch real flights in measurable numbers, with international carriers on the order of eight to ten percent of affected flights and general aviation near one third at a major coastal site [\[3\]](#ref-3), [\[2\]](#ref-2). One caution is essential: these are expected signs derived from launch-era evidence and from the four PRISMA reviews, not executed estimates, and the confidence attached to the expectation is moderate, grounded in convergent but indirect evidence rather than in a prior causal estimate that does not exist.

If H1 holds, the disruption parameter, the aggregated average treatment effect on the treated of a reentry-driven closure on the cost outcomes, becomes the first operational quantity. Its implication is concrete: a regulator acquires the marginal aviation cost of one reentry-driven closure, expressed in the same currency, dollars of direct operating cost and minutes of delay, in which the regulator already accounts for every other source of airspace disruption. The present authorization regime has no such number. The Federal Aviation Administration integrates launch and reentry chiefly by closing large, predetermined hazard areas for windows historically measured in hours, a posture that is conservative by design and that has preserved a strong safety record [\[2\]](#ref-2), [\[4\]](#ref-4). What that posture has never had is a price on the disruption it produces, which means that the regulator weighing whether to permit a given reentry cadence has been weighing a known safety benefit against an unmeasured cost. A measured disruption parameter removes the asymmetry. It tells the regulator how much aviation cost a cadence of one reentry per week, or per day, will impose, and it does so at the margin, per event, so that the cost of an incremental increase in cadence is legible rather than guessed.

The second operational quantity, if H1 holds, is the avoided-cost parameter: the difference in cost between a static-closure regime and a prediction-informed dynamic-closure regime that narrows the withheld volume in time and space as the reentry prediction tightens. Its implication is that the dissertation prices a specific, buildable intervention rather than an abstraction. The causal lever is well identified in the design. Closure size scales with prediction uncertainty, and that uncertainty falls sharply as the object nears reentry, driven down from large atmospheric-density, ballistic-coefficient, and reentry-angle uncertainties toward a tighter late-available envelope [\[21\]](#ref-21), [\[22\]](#ref-22). A dynamic closure sized to the uncertainty available shortly before reentry therefore withholds a smaller volume and displaces fewer flights than a static closure sized to a worst-case envelope fixed hours in advance. The avoided cost is the difference, and the components of the dynamic regime whose value it captures are not hypothetical. Trajectory-accuracy requirements that let automation decide which flights clear a hazard area before activation have been derived [\[9\]](#ref-9); predicted-trajectory accuracy thresholds for reducing aviation impact have been worked out for launch and reentry [\[33\]](#ref-33); collaborative information exchange to enable tactical airspace management during space operations has been prototyped [\[28\]](#ref-28); and a launch coordination center concept has been tested in European airspace and has begun to quantify how dynamic handling shrinks the closure footprint relative to static policy [\[10\]](#ref-10). The avoided-cost parameter, on this reading, estimates the value of completing an adaptation the field has already begun for launch and extending it to reentry, not the value of inventing a capability from scratch. The confidence here is lower than for the disruption parameter, and the dissertation says so: the avoided cost has no direct empirical precedent, its support is component evidence and analogue rather than a measured reentry avoided cost, and the objection that the components do not compose into the projected saving when assembled for reentry is admitted and met by the requirement that ReentryFlow's counterfactual outputs be validated against realized exposure before they are trusted.

If both parameters return positive and statistically distinguishable from zero, the joint implication is a priced trade. The regulator can set the disruption parameter, the cost of doing nothing differently as cadence rises, against the avoided-cost parameter, the cost recovered by moving to dynamic closure, and can decide whether the investment in prediction and dynamic airspace management pays at the cadence it expects. That is the decision the dissertation is built to inform, and under H1 it informs it with two numbers that are commensurable, marginal, and falsifiable.

### 7.3.1 Why a positive disruption parameter is a congestion externality, not merely a private cost

A positive disruption parameter under H1 is more than a private inconvenience to the affected carriers, and reading it correctly is essential to the policy claims that follow. The measured cost is a congestion externality in the precise economic sense, and this classification, not the raw dollar figure, is what makes the measurement a price rather than an accounting curiosity. The cost-of-delay economics establishes the point. Air traffic delay is in part a congestion externality: when one user consumes scarce airspace capacity, it imposes delay on others that the consuming user does not bear, which is the textbook condition under which a market under-provides the avoidance of disruption [\[11\]](#ref-11). A reentry-driven closure is exactly such a consumption of scarce capacity, but with a sharper feature than ordinary congestion, because the consuming party, the space operator, is not even a participant in the aviation market that bears the cost. The cost therefore falls on third parties who neither caused nor are compensated for it, which is the defining mark of an externality and the reason its measurement is the precondition for any corrective instrument.

The classification has a second consequence that the headline number alone would hide, and the design is built to expose it. The within-window cost understates the true cost, because a localized airspace disruption propagates through the network: delay at one node raises delay downstream, so the cost of a closure outlives its active window as displaced traffic cascades [\[12\]](#ref-12). A positive disruption parameter measured on the activation window is therefore a lower bound on the externality, and the event-study profile's post-activation decay shape is itself a substantive finding about how much of the cost the within-window measure misses. The interpretive implication for the stakeholders is that a regulator pricing the externality on the within-window cost alone would under-price it, and the dissertation's event-study structure is what lets the regulator see and, if it chooses, incorporate the propagated tail. The network-propagation magnitude is bounded rather than precisely estimated in this design, because the outcome construct, delay and direct operating cost on the exposed flights, omits passenger time value and full schedule-network knock-on; the propagation evidence is used to bound the omitted downstream cost rather than to estimate it, and the confidence in the bound is moderate. Read this way, a positive disruption parameter is not a complaint on behalf of carriers; it is the empirical content of an unpriced externality on shared national infrastructure, and that is the reading the policy section carries forward.

## 7.4 Implications if H0 holds

The null is not a failure mode to be explained away; it is an admissible scientific result with its own implications, and the chapter develops them substantively. If H0 holds in its strong form, a precisely estimated disruption effect near zero with a confidence interval tight enough to exclude economically meaningful effects, the dissertation delivers a finding of equal decision value to H1, namely that the present static-closure posture imposes no material marginal cost on the NAS at the observed cadence and that the case for investing in dynamic closure on cost grounds is correspondingly weak. The reasons for taking this seriously are structural. The number of U.S.-airspace-intersecting reentry events in the current record is modest, the events are concentrated in particular sectors and seasons, and the displaced traffic may have ample slack to absorb a closure without measurable cost when cadence is low. The same causal chain read in reverse explains why: if the withheld volume is small relative to available capacity, or if exposed flights reroute at negligible marginal cost because alternative routings are nearly as efficient, the mechanism that converts a closure into a cost is throttled, and the observed effect is near zero not because the design is blind but because the cost genuinely is small at this cadence.

The interpretive discipline that the design imposes here is the distinction between a strong null and an underpowered one, and it is load-bearing. A precisely estimated near-zero effect, one whose confidence interval rules out the economically meaningful effect sizes that the launch-era evidence would lead one to expect, is a substantive finding: it says that reentry at current cadence is, on the evidence, not a material aviation-cost problem. An imprecisely estimated null, one whose confidence interval is wide enough to contain both zero and a large effect, is not a finding at all; it is the design reporting that the available sample cannot resolve the question. The dissertation refuses to let the second masquerade as the first. The minimum-detectable-effect check is the gate: the analysis plan reports the smallest effect the available sample can detect at the chosen power, and an estimated null is treated as confirmation of H0 only when that check certifies that an economically meaningful effect would have been detected had it been present. Where the check fails, the result is reported as inconclusive, the question is held open, and the path to resolving it, more events as cadence rises, is named rather than papered over. This is the honest reading of the design-based tradition's caution that an estimate is local to the events and margins observed: if the observed margins do not contain enough treatment variation to identify the effect, the design says so.

If H0 holds in its strong form, the implications for the stakeholders are not negative; they are reorienting. For the regulator, a credible strong null says that the current conservative posture is not, at present cadence, a structural drag on the NAS, and that resources earmarked for dynamic-closure capability could be deferred or redirected until cadence rises to a level at which the cost reappears. For a prediction-and-modeling investment such as ReentryFlow, a strong null on the disruption parameter, or a near-zero avoided-cost parameter, says plainly that better prediction does not pay on aviation-cost grounds at the cadence studied, which is exactly the kind of result that a responsible investment decision needs and that an advocacy posture would suppress. The dissertation's value proposition is precisely that it would report this. A near-zero avoided-cost parameter would imply that dynamic closure buys nothing over static closure in cost terms, that the safety case for the existing posture stands without an offsetting cost penalty, and that the field's investment in trajectory-accuracy and tactical-exchange capability should be justified on safety, capacity-headroom, or future-cadence grounds rather than on present cost avoidance. Each of these is a publishable, decision-relevant result. The mark of a genuinely falsifiable design, as opposed to a foregone conclusion dressed as research, is that the null outcome is as informative as the alternative, and this chapter treats it that way.

There is a third reading, intermediate between the strong null and H1, that the design accommodates and that the discussion flags. The disruption parameter may be positive and distinguishable from zero, confirming the first part of H1, while the avoided-cost parameter is near zero, falsifying the second part. The substantive meaning of that combination is specific and important: reentry imposes a real, measurable cost on the NAS, but better prediction and dynamic closure do not reduce it, perhaps because the binding constraint on closure size is not prediction uncertainty but a residual ground-and-air risk floor that no amount of prediction can shrink below, or because the dynamic regime's coordination overhead offsets its footprint saving. That reading would redirect the policy conversation from prediction investment toward the risk-acceptance threshold itself, and toward whether the regulator is willing to price residual risk differently. The design separates the two parameters precisely so that this intermediate outcome is legible rather than averaged away into a single ambiguous number.

## 7.5 Theoretical contribution back to the anchors

The dissertation borrows three frameworks and is built to give something back to each. This section states what the measurement returns to Rao's externality economics, to North's institutional economics, and to the design-based identification tradition associated with Angrist and Pischke, and it is careful to claim only what a design-stage contribution can claim.

### 7.5.1 The externality price (Rao)

Rao and colleagues analyze the orbital environment as a congestible common in which a launching party imposes collision and debris risk on others without pricing it, and they show that an orbital-use fee that internalizes the externality could raise the value of the satellite industry substantially relative to the open-access outcome [\[23\]](#ref-23). The transfer this dissertation makes is to recognize the reentry-airspace problem as the terrestrial analogue of that orbital externality. A reentry-driven closure is a capacity shock that a space operator imposes on aviation users who are not party to the authorization decision and who bear the resulting delay, reroute, and fuel cost; the operator does not face that cost, so the airspace common is over-consumed in exactly the structural sense that Rao's framework formalizes for orbit. The contribution back to the framework is the missing empirical primitive. Rao's mechanism requires a measured marginal external cost to set the fee that internalizes it; the orbital-debris literature has that primitive for collision risk, but the reentry-airspace literature has not had it for delay. The disruption parameter is that primitive for the airspace case. It supplies the marginal external cost of a reentry-driven closure, the quantity any Pigouvian instrument on reentry, a corridor-access fee, a closure-time charge, or a tradable closure right, would have to reference. One bound on the claim is essential: the dissertation provides the externality price, not the optimal fee, because the optimal fee depends on the elasticity of reentry demand and on the social cost of the residual risk, neither of which this design estimates. The contribution is the term in the numerator, not the full optimization, and the chapter says so plainly. The supporting space-economy literature reinforces that this primitive is the operative one: market formation in the new space economy is governed by the correct pricing of externalities and by institutional design rather than by technology readiness alone [\[16\]](#ref-16), [\[18\]](#ref-18), [\[17\]](#ref-17).

### 7.5.2 The institution the price enables (North)

North distinguishes institutions, the rules of the game, from organizations, the players, and holds that institutions exist to lower the transaction costs of impersonal exchange [\[24\]](#ref-24). The dissertation's transfer is to read the reentry-authorization regime as an institution and to classify its two candidate forms by their transaction costs. Static, case-by-case AHA closure is a high-transaction-cost rule: every event is adjudicated conservatively and afresh, the cost falls on third parties who do not participate in the decision, and the adjudication does not accumulate into a reusable schedule. A repeatable, prediction-informed, dynamic-closure rule is a lower-transaction-cost institution: it replaces bespoke adjudication with a standing decision procedure keyed to the prediction available at closure time. The contribution back to North's framework is the identification of the precise input that gates the move between the two institutions. North's account explains why a lower-transaction-cost rule is desirable but does not, by itself, tell a particular regulator how to justify the switch; the switch requires that the regulator be able to weigh the narrower closure's residual risk against its cost saving, and that weighing requires a measured price for the disruption the current rule imposes. The dissertation supplies that price. The avoided-cost parameter, in North's terms, is the transaction-cost saving of the institutional change, and the disruption parameter is the cost of the status quo against which that saving is measured. The claim is therefore that the measurement is institutionally load-bearing: it is not an academic ornament but the specific empirical input without which the rule change cannot be justified on its own terms. The space-economy literature again converges, locating the public-to-commercial transition in institutional design and authorization predictability rather than in engineering milestones [\[16\]](#ref-16), [\[18\]](#ref-18), [\[17\]](#ref-17). One caveat bounds the claim: the dissertation prices the inputs to the institutional choice; it does not design the dynamic-closure rule itself, which is a regulatory and architectural task that the trajectory-accuracy and tactical-exchange literatures address [\[9\]](#ref-9), [\[33\]](#ref-33), [\[28\]](#ref-28) and that lies beyond an econometric measurement.

### 7.5.3 The design-based local estimate and its honest caution (Angrist and Pischke)

The methodological anchor contributes the discipline that makes the two prior contributions credible, and the dissertation contributes back to it a worked application in a domain it has not previously reached. The design-based tradition insists that a credible source of variation and an explicit treatment-and-control definition carry the causal claim, rather than functional-form assumptions. The natural experiment here is the reentry event itself, whose timing for uncontrolled objects is set by orbital decay and is plausibly unrelated to contemporaneous demand for the affected sector, conditional on covariates. The contribution back to the tradition is twofold. First, it extends the staggered difference-in-differences frontier, the Callaway and Sant'Anna estimator with the Goodman-Bacon and de Chaisemartin and D'Haultfoeuille diagnostics, into a space-infrastructure application where treatment timing is genuinely staggered and heterogeneous and where the forbidden-comparison pathologies of two-way fixed effects would otherwise bite [\[14\]](#ref-14), [\[13\]](#ref-13), [\[15\]](#ref-15). Second, and more important for the discussion, it carries the tradition's signature epistemic caution into a policy domain that has tended to report simulation outputs as if they were estimates. The honest external-validity statement, developed in Section 7.7, is the direct expression of that caution: the estimate is local to the observed cadence and event mix, and the dissertation reports the cost as a function of cadence and prediction uncertainty so that extrapolation is conditional and stated. The contribution back to the anchor is thus not a new estimator but a demonstration that the design-based posture, identify cleanly, estimate locally, extrapolate explicitly, disciplines a space-policy question that the descriptive and simulation literatures have left under-identified. The caution is the one the tradition itself demands and that this chapter never relaxes: a clean design does not manufacture power, and where the sample cannot resolve the effect, the design reports inconclusiveness rather than a result.

## 7.6 Policy and mission implications
The policy and mission implications follow directly from the two operational quantities. The claim of this section is that the dissertation's number, in either direction, is the evidence base on which a reentry-authorization or corridor-pricing institution must rest, and the specific input that lets NASA and its partners act as economic enablers of the down-mass sector rather than as passive licensors.

The implication is most acute for the down-mass sector, and the discussion makes that case before the general regulatory one, because the down-mass case is where the externality binds hardest. The return to Earth of objects mined, refined, or manufactured in space is not launch played in reverse: reentry trajectories carry greater inherent prediction uncertainty than ascent, disperse hazard over larger geographic areas, and intersect cruise altitudes differently, so the airspace cost of a down-mass return is structurally larger and harder to predict than the airspace cost of a launch [\[4\]](#ref-4), [\[1\]](#ref-1). For a down-mass operator, the value of the returned product depends on delivering it through occupied airspace to a point near its insertion into a terrestrial supply chain, so the return leg is integral to the business rather than incidental, and the airspace cost is therefore a direct input to the operator's economics rather than an externality the operator can ignore. The interpretive consequence is that pricing the externality is not only a matter of protecting aviation users; it is also a matter of giving the down-mass operator and its public partner a legible cost to plan against. The space-economy literature treats authorization predictability and corridor access as first-order economic variables that pace market formation [\[17\]](#ref-17), [\[18\]](#ref-18), and the business-model literature on the new space economy locates the binding constraint downstream of propulsion in exactly these institutional terms [\[17\]](#ref-17). The dissertation's number is what lets that constraint be priced rather than feared, which is the precondition for the down-mass market to form at a cost its participants can anticipate.

The most immediate implication for the regulator's authorization architecture follows. The governance literature frames the choice of authorization architecture as a first-order economic variable and surveys cross-domain precedents for managing shared corridors, drawing lessons from air and maritime treaty governance for space traffic management [\[61\]](#ref-61) and assembling the emerging body of space-traffic-management doctrine [\[62\]](#ref-62). What that literature has lacked is a priced input to the choice. The disruption parameter supplies it: it tells an authorization regime how much aviation cost a given reentry corridor and cadence will generate, which is the quantity a corridor-access fee or a closure-time charge would be set against. The avoided-cost parameter complements it by telling the regime how much of that cost a move to prediction-informed dynamic closure would recover, which is the quantity that justifies, or fails to justify, the regulatory cost of standing up the dynamic regime. Where a fee instrument is contemplated, the externality price the dissertation supplies is the term the fee internalizes, in exactly the structure that the orbital-use-fee analysis uses for the orbital case [\[23\]](#ref-23) and that the economics-and-policy literature on orbital management treats as the central design problem [\[59\]](#ref-59).

The mission implication for NASA and JPL is sharper still, because it bears on a live line of work. NASA Langley and Analytical Mechanics Associates maintain an active airspace-modeling effort for reentry authorization, and the joint study from which this dissertation draws argues that government can act as an economic enabler of the down-mass sector by adapting launch-era airspace capabilities to reentry rather than building new infrastructure. A credible per-event cost and avoided cost are the evidence base for that enabling role, and the reasoning rests on the public-private-partnership logic that the space-economy literature has formalized: well-designed partnerships in which the public party reduces a binding cost or uncertainty for the private party have been shown to foster outer-space innovation [\[58\]](#ref-58). Translated to this case, the public party's enabling act is to reduce the airspace cost and the authorization uncertainty that the return leg imposes on a down-mass operator, and the dissertation supplies the measurement that tells the public party how large that cost is and how much of it a modeling-and-prediction investment can remove. For JPL and the broader civil-space enterprise, the same estimate sets a defensible price on the externality that reentry imposes on aviation, which is the quantity any future corridor-authorization or fee regime must reference, and which the benefit-sharing and investment-protection literature treats as a precondition for stable commercial space-resource activity [\[60\]](#ref-60).

The dissertation also bears on the choice among instruments, without prescribing one, and stating its bearing precisely guards against over-claiming. A measured externality price is compatible with several corrective instruments: a Pigouvian corridor-access fee set at the marginal external cost, a closure-time charge that prices the duration of the withheld volume, a tradable closure right that lets the market allocate scarce corridor access, or simply a tariff-free move to dynamic closure that shrinks the externality at source rather than pricing it. The dissertation supplies the common input all of these require, the marginal cost of the disruption, but it does not adjudicate among them, because the choice depends on the elasticity of reentry demand, the administrative cost of each instrument, and the distributional and treaty constraints that the cross-domain governance literature catalogues for shared corridors [\[61\]](#ref-61), [\[62\]](#ref-62). The benefit-sharing and investment-protection literature adds a further constraint the instrument designer must respect, namely that a stable commercial space-resource regime depends on the predictability and perceived fairness of the cost the operator faces [\[60\]](#ref-60). The interpretive claim is therefore bounded and honest: the dissertation prices the externality and so makes the instrument-design conversation possible on evidence rather than assertion, but the design of the instrument itself is a governance task that the measurement enables rather than performs. One caveat applies: an instrument set on a cadence-local price will need recalibration as cadence rises, which is the direct policy consequence of the function-valued external-validity posture developed in Section 7.7.

The implication that survives even the strong null deserves explicit statement, because it is the clearest demonstration that the dissertation's value does not depend on the sign. If the executed disruption parameter is a precise near-zero, the mission implication is that NASA and JPL can credibly assure the regulator and the nascent down-mass industry that, at present cadence, reentry is not a material aviation-cost problem, which removes a perceived barrier to commercial return operations and reorients the airspace-modeling investment toward the cadence threshold at which the cost would reappear. If the disruption parameter is positive but the avoided cost is near zero, the implication is that the enabling role shifts from prediction investment toward the risk-acceptance threshold and the question of how residual risk is priced. In every admissible outcome, the dissertation hands the stakeholders a number that changes what they should do, which is the operational test of a policy-relevant measurement.

## 7.7 Rival explanations

A causal claim is only as strong as its treatment of the explanations that would account for the same observation without the proposed mechanism. The claim of this section is that the leading rivals to a measured reentry cost are anticipated in the design and confronted by specific, named features of it, and that the design's credibility rests on those confrontations rather than on the plausibility of the headline mechanism. Each rival is stated as a competing causal account, and each is met with the design feature that would distinguish it from the reentry effect.

The leading rival is confounding by concurrent disruption, principally weather and unrelated traffic-management initiatives. The competing account holds that the disruption attributed to a reentry closure is in fact produced by a storm system or a flow-control program that happened to coincide with the closure window, so that the estimated treatment effect is the effect of weather, not of reentry. This is the most serious rival because weather is both a powerful airspace disruptor and a plausible coincident. The design confronts it on three fronts. First, weather severity and concurrent non-space traffic-management initiatives enter the specification as explicit covariates, so that the comparison holds them fixed rather than letting them load onto the reentry indicator. Second, the design exploits a structural fact that the rival cannot evade: reentry timing for uncontrolled objects is set by orbital decay and is independent of terrestrial weather, so the treatment is, by construction, orthogonal to the confounder in expectation, which is the conditional-independence basis of the identification strategy. Third, the design runs adjacent-sector placebo tests, estimating the apparent effect in sectors that border the closure footprint but lie outside it; a weather system large enough to drive the result would register in the adjacent sectors as well, whereas a reentry closure would not, so a null in the placebo sectors and an effect in the treated sector discriminate between the two accounts. This holds with one caveat: controlled reentries, whose timing operators do schedule, are not protected by the orbital-decay argument; the design conditions on schedule for those events and tests sensitivity to excluding them entirely, and the objection that some residual weather-correlated scheduling survives is met by that exclusion test rather than asserted away.

The second rival is reverse causation: that closures are scheduled into windows that are already disrupted, so that the apparent effect of reentry on cost is actually the effect of pre-existing disruption on the decision to close. The competing account would have the causal arrow run from a quiet or congested airspace state to the closure decision, rather than from the closure to the cost. This account is implausible for uncontrolled reentries, whose epochs are fixed by decay and cannot be timed to airspace conditions, and the design leans on that implausibility as its primary defense. For controlled reentries the account is admissible, and the design meets it by excluding controlled events in a robustness specification and comparing the estimate to the all-events estimate; a result that survives the exclusion is not driven by schedule-into-disruption, and a result that does not survive it is flagged as potentially contaminated by the rival rather than reported as clean. The confidence in the reverse-causation defense is high for the uncontrolled subsample and moderate for the full sample, and the chapter does not overstate it.

The third rival is full anticipation and pre-absorption: that operators and the air-traffic system anticipate the closure so completely that traffic is rerouted before activation, leaving no marginal cost to appear at the activation window and producing a spurious null, or relocating the cost into the pre-period and producing a mis-timed effect. The competing account holds that the true cost is real but is paid before the treatment window opens, so that an estimator keyed to the activation window either misses it or mis-attributes it. The design confronts this with the event-study structure itself. The leads of the event-study profile, the estimated effects in the windows before activation, are the direct test for anticipation: non-zero leads reveal pre-activation cost, and the design responds by widening the treatment window to capture the pre-activation avoidance rather than discarding it. Because anticipation relocates the cost rather than eliminating it, a design that inspects the leads and adjusts the window converts the rival from a threat into a measured component of the effect. The supporting reasoning draws on the delay-propagation evidence, which establishes that a localized airspace disruption does not stay localized in time, so that the within-window effect is a lower bound and the leads-and-lags profile is the honest accounting [\[12\]](#ref-12).

A fourth rival, less often named but material here, is measurement error in the treatment itself: that the AHA footprint polygons are model outputs carrying their own uncertainty, so that flights are mis-assigned to treated or control status and the resulting attenuation biases the estimate toward zero. This rival does not produce a spurious effect; it threatens to hide a real one. The design confronts it by propagating the EU SST prediction-uncertainty bounds into the exposure measure rather than treating the footprint as exact, and by validating ReentryFlow's predicted exposed-flight sets against realized FAA exposure, reporting precision and recall, before any exposure-dependent estimate is trusted. One caveat remains: residual mis-assignment will attenuate the estimate, so that a positive finding is conservative and a null finding must clear the minimum-detectable-effect gate before it can be read as a strong null rather than as attenuation masquerading as absence.

The cumulative force of this section is that the headline mechanism is not assumed to win; it is required to survive a battery of competing accounts, each of which has a design feature aimed at it. That is the posture the dissertation carries throughout: closures withhold real airspace and touch real flights, the cost grows with cadence and traffic density, closure size is driven by prediction uncertainty that dynamic closure can shrink, dynamic closure dominates static segregation on cost at equal safety, and the narrowing is bounded by validated prediction accuracy and by ground-and-air risk limits [\[63\]](#ref-63), [\[64\]](#ref-64). The rival-explanation analysis is where the first two of these are defended against the accounts that would dissolve them.

## 7.8 External validity

The honest reach of the estimate is the final interpretive obligation, and the claim of this section is that the dissertation's external-validity position is conditional by construction: the cost is reported as a function of cadence and prediction uncertainty, so that its extrapolation to a future down-mass regime is explicit and stated rather than assumed. This is the design-based tradition's local-effect caution taken seriously rather than waved at.

The grounds for the caution are structural. The estimate is identified from the reentry events and exposure margins actually present in the record, and that record is dominated by the current U.S. reentry mix and cadence: a modest number of events, concentrated in particular sectors and seasons, intersecting a traffic pattern that has slack the future may not have. An average treatment effect estimated on that record is local to it. The reason for not reporting a single transportable scalar is that the mechanism is explicitly cadence-dependent and uncertainty-dependent: the per-event cost rises with traffic density and with the size of the withheld volume, and the size of the withheld volume rises with prediction uncertainty, so the cost of a reentry at high future cadence, when displaced traffic has less slack to absorb it, is not the cost of a reentry today, and cannot be obtained by multiplying today's per-event cost by tomorrow's event count. The design responds by estimating the cost as a function rather than a constant: prediction uncertainty enters as a continuous treatment intensity, and cadence enters through the event-study and aggregation structure, so that the output is a relationship between cost, cadence, and prediction tightness rather than a single number to be extrapolated naively.

This has a direct consequence for how the stakeholders may use the result and for how they may not. They may read off the cost at a stated cadence and a stated prediction-uncertainty level, and they may project forward by stating the cadence and uncertainty they expect and reading the corresponding cost, with the projection's assumptions on the table. They may not treat the estimate as a cadence-invariant constant, because the design does not license that and the mechanism contradicts it. The wider context reinforces the relevance of the conditional projection: the airspace-closure burden is expected to grow with the expansion of large constellations and their deorbit-driven reentry events [\[1\]](#ref-1), and the down-mass sector's emergence would add controlled returns on top of that uncontrolled baseline, so the cadence at which the cost must be evaluated is a moving target that only a function-valued estimate can track. The function itself is estimated locally; its shape is identified over the range of cadence and uncertainty observed, and extrapolation beyond that range is an assumption, clearly labeled, not an estimate. A future high-cadence regime may exhibit nonlinearities, capacity thresholds beyond which displaced traffic can no longer be absorbed at constant marginal cost, that the present record cannot reveal, and the dissertation names this as the principal limit on the reach of its number rather than burying it.

The contribution is falsifiable in both of its parts even under this conditional reading, and stating the falsification conditions is the cleanest expression of the external-validity posture. The first part, the measurable-cost claim, is falsified by a precisely estimated near-zero disruption effect at the observed cadence, with the minimum-detectable-effect gate certifying that an economically meaningful effect would have been detected had it been present. The second part, the avoided-cost claim, is falsified by a near-zero avoided-cost effect, which would imply that better reentry prediction does not reduce airspace cost and that dynamic closure buys nothing over static closure on cost grounds. Either falsification is a publishable, decision-relevant result, and that symmetry is the guarantee that the design is a genuine test rather than a foregone conclusion. The external-validity statement is the same in either direction: whatever the sign, the number is local to the cadence and event mix observed, reported as a function of cadence and prediction uncertainty, and offered to NASA, JPL, and the civil-space enterprise as a conditional, design-based price for the cost that reentry imposes on aviation and for the cost that better prediction avoids. That conditional price, honestly bounded, is the contribution this chapter has interpreted, and it is the evidence base on which any reentry-authorization or corridor-pricing institution that follows will have to rest.

## 7.9 Synthesis

Pulling the threads together, the discussion establishes that the dissertation's worth does not depend on which way the test resolves. Under H1, it supplies a marginal externality price and a priced, buildable intervention, converting the regulator's standing trade-off between a known safety benefit and an unmeasured cost into a legible decision. Under a strong H0, it supplies an equally decision-relevant assurance that, at present cadence, reentry is not a material aviation-cost problem and that the cost case for dynamic-closure investment is weak, while its minimum-detectable-effect discipline forbids an underpowered null from masquerading as that assurance. Under the intermediate outcome, a real cost that prediction cannot shrink, it redirects the policy conversation from prediction investment to the risk-acceptance threshold. To each anchor it returns a specific primitive: the externality price Rao's mechanism needs, the transaction-cost measurement North's institutional change requires, and a disciplined design-based application that carries the identification tradition's local-and-explicit posture into a domain that has reported simulation as estimate. Against the rivals, weather and concurrent disruption, reverse causation, full anticipation, and treatment mis-measurement, it places named design features rather than assertions, and it admits where its confidence is moderate rather than high. And it states its reach honestly, as a cost function of cadence and prediction uncertainty estimated locally and extrapolated only conditionally. The conclusion to which this discussion delivers the reader is the one the conclusion chapter will formalize: a defensible, falsifiable, design-based number for the reentry-to-aviation externality and for the avoided cost of better prediction is valuable to the civil-space enterprise whichever way it lands, because a number that can be wrong is the only kind a regulator, a program office, or a partnership can responsibly act on.


# Chapter 8: Conclusion

## 8.1 The chapter thesis: a design-based number is the deliverable, whichever way the test resolves

This dissertation makes one falsifiable claim and builds the apparatus to settle it. The claim is that spacecraft reentry and down-mass operations impose a measurable disruption cost on the U.S. National Airspace System, expressed in delay, reroute, and closure, and that improved reentry prediction combined with dynamic airspace management yields a quantifiable avoided cost. The null is that reentry imposes no measurable NAS cost. The thesis of this concluding chapter is narrower and more durable than the substantive hypothesis: the enduring contribution of the work is not the sign of the treatment effect but the existence of a credible, design-based estimate of that effect, and that estimate is decision-relevant whether it comes back large, small, or indistinguishable from zero. The contribution survives the falsification of H1, because a precisely estimated near-zero effect is itself the number a regulator needs, and it survives a confirmation of H1, because a positive, bounded estimate is the price that a corridor-authorization or fee regime must reference. The dissertation is built so that its value does not depend on the outcome it expects.

The problem this chapter addresses, in the current-state, desired-state, gap, consequence frame that organizes the whole work, is a problem of closure rather than of discovery. The current state at the end of seven chapters is a fully specified research design: an estimator chosen and defended, variables operationalized, four data layers named and their joins described, threats to validity enumerated with their mitigations, an analysis plan with a pre-committed decision rule, and a discussion that develops the implications under both outcomes. The desired state is a clear statement of what has been established at the design stage, an honest accounting of what has not, and a concrete path from the present design to an executed estimate on the full dataset. The gap between them is the gap between a design and a result: nothing here has been estimated on the joined EU SST and FAA panel, because that panel does not yet exist. The consequence of leaving that gap unmarked would be to overstate the work, to let a careful design-stage dissertation read as though it carried findings it does not yet have. This chapter closes the gap by stating exactly what stands, what does not, and what must happen for the design to become an estimate.
## 8.2 What the dissertation establishes, restated as the single falsifiable contribution

The contribution, preserved in substance from the prospectus and the shared bible, is this: spacecraft reentry and down-mass operations impose a measurable disruption cost on the NAS, expressed in delay, reroute, and closure, and improved reentry prediction combined with dynamic airspace management yields a quantifiable avoided cost. It is stated as competing hypotheses so that it can fail. H0 holds that reentry imposes no measurable cost on the NAS, that the average treatment effect of a reentry-driven Aircraft Hazard Area (AHA) activation on exposed-flight cost is statistically indistinguishable from zero. H1 holds that reentry and down-mass operations impose a measurable, positive disruption cost in delay, reroute, and closure, and that improved prediction plus dynamic airspace management yields a positive, statistically distinguishable avoided cost. H1 decomposes into two estimable parameters, both fixed in their definitions: the disruption parameter, the per-event NAS cost, the aggregated average treatment effect on the treated of a reentry-driven AHA closure on exposed-flight cost outcomes; and the avoided-cost parameter, the difference in that cost between a static-closure regime and a prediction-informed dynamic-closure regime.

A set of contributions stands independent of how the test resolves. They are methodological and conceptual rather than empirical, and the act of estimation cannot retroactively invalidate them. The first is the framing of a reentry closure as a measurable causal treatment rather than a descriptive nuisance. The launch-era airspace literature establishes that closures withhold real volume and touch real flights, with international carriers on the order of 8 to 10 percent of affected flights at a major coastal site and general aviation near one third, but it does so by simulation and historical traffic accounting, not by a causal design [\[3\]](#ref-3). Reframing the reentry closure as a treatment whose effect can be identified against a credible counterfactual is a contribution that exists in the design and does not wait on the data. The reframing is novel and load-bearing, because the corpus contains no prior causal estimate of a reentry event's effect on realized flight cost, only simulation and risk assessment [\[1\]](#ref-1), [\[3\]](#ref-3); a causal target, once defined with an explicit treatment and control, changes what kind of evidence counts and what a regulator can do with it. The value of the reframing is conditional on the eventual feasibility of the identification, which Section 8.3 examines honestly. The objection, that reframing without estimation is merely relabeling, is answered by the second contribution.

The second contribution that stands is the identification strategy itself. The design takes the reentry event as a natural experiment whose timing, for uncontrolled objects, is set by orbital decay and is plausibly independent of contemporaneous demand for the affected airspace sector. It adopts the Callaway and Sant'Anna staggered difference-in-differences estimator as the primary specification, with the Goodman-Bacon decomposition and the de Chaisemartin and D'Haultfoeuille robust estimator as diagnostics against the known biases of two-way fixed effects under staggered, heterogeneous timing [\[14\]](#ref-14), [\[13\]](#ref-13), [\[15\]](#ref-15). This is a complete and defensible identification architecture, and it is a contribution whether or not it is ever run, because it specifies how the question could be answered and what assumptions the answer would rest on. The third contribution is the conceptual bridge from the airspace cost to the economic instruments that would act on it: the externality structure that makes the measured cost a price rather than a curiosity (the reentry operator imposes a capacity shock on aviation users whose cost the operator does not bear, the terrestrial analogue of the orbital-commons externality) [\[23\]](#ref-23), and the institutional argument that the measurement is the precondition for moving from a high-transaction-cost, case-by-case closure rule to a lower-transaction-cost, prediction-informed dynamic rule [\[24\]](#ref-24). These three contributions are what the dissertation has already delivered. They are the argument in miniature: the cost is real, it is structured as an externality, and the design addresses the causal mechanism. Confidence in them is high, because they rest on the cited literature and on internal logical consistency rather than on an unexecuted estimate.

## 8.3 Limitations, stated honestly

A design-stage dissertation earns its credibility by naming what it cannot yet do, and the limitations here are real, ordered roughly by how much they could change the eventual estimate. None of the limitations is disqualifying, but two of them, statistical power and the validation dependency, are serious enough that an executed result must report them in the foreground rather than the footnotes.

The first and most consequential limitation is statistical power. The number of reentry events whose predicted footprint intersects U.S. airspace and triggers an AHA in the current record is modest, and the disruption parameter is an average over those events. With few treated cells, the minimum detectable effect is large, and an estimate near zero may reflect insufficient power rather than a true absence of cost. The design confronts this with a standing rule, fixed in the analysis plan: an underpowered null is inconclusive and is never read as confirmation of H0. The power analysis reports the minimum detectable effect at the available sample size, and a null is informative only if its confidence interval is tight enough to exclude an economically meaningful effect. Power binds on the first executed wave, because the present catalog holds only a small count of U.S.-airspace-intersecting reentry events and average treatment effects estimated on few treated units have wide sampling distributions. The load-bearing point is that this limitation shrinks mechanically as reentry cadence rises with large-constellation deorbit, so the design becomes more powerful over time rather than less [\[62\]](#ref-62), [\[1\]](#ref-1). The mechanism is concrete: more reentry events produce more treated cells, more treated cells narrow the sampling distribution of the aggregated ATT, and a narrower distribution converts an inconclusive null into either a confirmed effect or a precisely estimated zero. Confidence that power will eventually suffice is moderate and rises with each year of accumulating events; confidence that the first executed wave will be fully powered is low, and the dissertation says so.

The second limitation is the coverage bias of the EU SST reentry catalog. The catalog represents tracked objects above its size threshold, which biases the treatment sample toward larger, better-characterized reentries and away from small or untracked objects. Two consequences follow. The estimated disruption parameter is local to the larger reentries that actually trigger catalogued AHAs, and may not represent the cost of the long tail of smaller events; and the prediction-uncertainty bounds that serve as the treatment intensity are themselves model outputs carrying their own error, which the design propagates into the exposure measure rather than treating as exact. This is a construct-validity and external-validity limitation jointly: the treatment is measured with error because footprint polygons are model outputs, and the sample is selected on object size. The mitigation is to report the estimate explicitly as a function of object characteristics and prediction uncertainty rather than as a single scalar, so that the restricted coverage is a stated condition on the estimate rather than a hidden bias. Confidence in this characterization is high; what would lower the limitation's bite is a catalog with a lower size threshold, which is outside the candidate program's control.

The third limitation is the secondary-synthesis status of the parameterized priors. The four IAC-26 down-mass PRISMA reviews supply the prior ranges for affected-flight fractions, per-flight cost penalties, prediction-uncertainty magnitudes, and dynamic-closure efficiency gains, and several of their anchor figures are simulation-derived or analogy-derived rather than measured on reentry. The most prominent analogue, the 67 to 84 percent processing-time reduction reported for digital-licensing systems, is used in the analysis plan only as an order-of-magnitude prior for what process and modeling improvements can achieve, applied to closure footprint rather than to paperwork, and never as a finding. Priors built on analogy can mislead if the analogy is weak; the mitigation is that the priors enter only as plausibility bounds on the estimates and as the basis for the power analysis, not as inputs that the executed estimate inherits. An executed estimate overwrites the priors; it does not average with them. Confidence that the priors are adequate for bounding plausibility is moderate; confidence that they would substitute for measurement is deliberately withheld, because they will not.

The fourth limitation is the validation dependency on ReentryFlow. The avoided-cost parameter is formed, in part, by replacing the realized static closure with a ReentryFlow dynamic-closure counterfactual and differencing the costs, and ReentryFlow is also the deterministic bridge that maps a reentry footprint to a predicted exposed-flight set. The entire counterfactual arm of the design therefore inherits ReentryFlow's accuracy. The design treats the model as an object of validation rather than a trusted oracle: before any counterfactual is believed, ReentryFlow's predicted exposed-flight sets are checked against realized FAA exposure for past events, and its precision and recall are reported. The avoided-cost estimate is only as credible as that validation, because a miscalibrated exposure map propagates directly into the counterfactual cost, and a simulator used to generate a counterfactual must first reproduce the factual. If validation fails, the avoided-cost arm is reported as unidentified rather than estimated with unwarranted confidence. The hedge against a ReentryFlow validation failure, and the reason the avoided-cost claim does not stand or fall on the model alone, is that the avoided-cost parameter is also estimable directly from natural variation in closure tightness across events with differing prediction uncertainty, using prediction uncertainty as a continuous treatment intensity. Confidence in the avoided-cost arm is therefore conditional and explicitly two-track: high if ReentryFlow validates, and still moderate via the continuous-intensity route if it does not.

The fifth and structural limitation is that the design measures a partial cost. The outcome construct is the economic cost of disruption, proxied by delay minutes, added distance, fuel burn, and direct operating cost. These proxies omit passenger time value and the schedule-network knock-on effects that propagate delay downstream of the closed window, and they omit the coordination load that rising cadence places on air navigation service providers. The network-propagation evidence implies that a closure's cost outlives its active window as delay cascades through the system, so the within-window estimate is a lower bound on the true cost [\[12\]](#ref-12). The design uses that propagation evidence to bound the omitted downstream cost rather than to ignore it, but the bound is an adjustment, not a direct measurement. The honest statement is that the disruption parameter, as defined, understates total social cost, and the direction of that understatement is known even where its magnitude is not. Confidence in the direction is high; confidence in the magnitude of the omitted component is low and is reported as a bound.

## 8.4 A concrete future-research program

The path from the present design to an executed, decision-grade estimate is short and specific, and the design's value compounds rather than decays as the reentry record grows. The future program has three tracks: execution on the full data, extension to the high-cadence regime the design anticipates, and integration with the policy instruments the estimate is meant to inform. Each track is stated as a concrete next step rather than an aspiration.

### 8.4.1 Executing the design on the full dataset

The first track is the direct execution of the design that the preceding chapters specify. It has five steps, in the order the analysis plan fixes them, and each step is a discrete unit of work rather than an open problem. Step one is to build the panel by joining the EU SST event layer to the FAA NAS and SWIM flight layer through ReentryFlow exposure mapping, producing the sector-by-window cells with their treatment indicators, treatment intensities, outcomes, and covariates. This is an engineering task with a known input and output, not a research question; the two data sources are accessed under the candidate program's vault credentials but are not yet joined, and the join is the single largest piece of remaining construction. Step two is to validate ReentryFlow by comparing its predicted exposed-flight sets to realized FAA exposure for past events and reporting precision and recall, which gates the entire counterfactual arm as Section 8.3 describes. Step three is to estimate the disruption parameter with the Callaway and Sant'Anna estimator, report the event-study profile and the aggregated ATT on the cost outcomes, and run the Goodman-Bacon and de Chaisemartin and D'Haultfoeuille diagnostics [\[14\]](#ref-14), [\[13\]](#ref-13), [\[15\]](#ref-15). Step four is to estimate the avoided-cost parameter by differencing realized static-closure cost against the ReentryFlow dynamic-closure counterfactual and, independently, by using prediction uncertainty as a continuous treatment intensity. Step five is to run the robustness and placebo battery: weather placebos, adjacent-sector placebos, controlled-reentry exclusion, anticipation leads, and alternative cost factors. This is a tractable program, not a speculative one, because every step has a defined method and a named data source already specified in Chapters 4 through 6: a design whose every estimating equation, variable, and threat is written out can be executed by following it. The power constraint of Section 8.3 determines not whether the steps run but how informative their output is on the first wave.

A practical sequencing point follows from the power limitation. Because the design becomes more powerful as the reentry record grows, the most informative execution strategy is not a single one-shot estimate but a staged one: run the design on the present record to establish the pipeline, validate ReentryFlow, and report the minimum detectable effect, then re-run as the event count accumulates, treating the estimate as a quantity that sharpens over successive waves. This converts the power limitation from a defect into a schedule. The first wave delivers the validated pipeline and an honest, possibly inconclusive, estimate; later waves deliver the precision. The mechanism is the same one named in Section 8.3: accumulating events narrow the sampling distribution of the aggregated ATT, so the same design yields tighter intervals over time without any change to its identification.

### 8.4.2 Extension to higher-cadence down-mass regimes

The second track addresses the gap between the cadence the design will first observe and the cadence the contribution ultimately concerns. The estimate produced on the present record is local to the current reentry mix and frequency, and the design-based tradition is explicit that such an estimate does not automatically extrapolate to a future high-cadence down-mass regime. The future program treats this as a modeling task with a defined structure rather than an unbounded uncertainty. Because the design reports the cost as a function of cadence and prediction uncertainty, the extrapolation to a higher-cadence regime is conditional and stated: the program-level cost is the per-event cost multiplied by a projected reentry cadence over a planning horizon, with the per-event cost itself allowed to vary with traffic density and closure size. The forward research question is the shape of the cadence-to-cost relationship at frequencies the current record does not reach, where network congestion may make the marginal closure more costly than the average closure observed at low cadence. The mechanism for that nonlinearity is identified in the delay-propagation literature: as the baseline NAS load rises and as closures arrive more often, each closure displaces traffic into an already more congested network, and the propagated cost grows faster than the count of closures [\[12\]](#ref-12). The cost function is plausibly convex in cadence, because the propagation evidence and the congestion-externality structure [\[11\]](#ref-11), [\[12\]](#ref-12) imply that convexity is the generic signature of a congestible resource under rising demand. One point is essential: the present record cannot estimate the convexity directly because it does not contain the high-cadence variation, so the convexity is a hypothesis for the high-cadence regime rather than a measured feature, and confidence in it is moderate and downgraded accordingly. This is the honest external-validity posture: the design extrapolates explicitly, conditionally, and with its confidence calibrated to the design-stage evidence grade, rather than assuming the low-cadence estimate carries to high cadence unchanged.

A parallel extension concerns the down-mass sector specifically, as distinct from the broader population of uncontrolled reentries. The down-mass case is the one in which the return leg is integral to the economic activity rather than incidental, because the value of a returned product depends on delivering it through occupied airspace to its insertion into a terrestrial supply chain. The present record contains few or no commercial down-mass events, so the disruption parameter is estimated on launch-stage debris and controlled reentries and then applied to a down-mass future. The forward program must therefore treat the down-mass extension as a transferability question: under what conditions does a cost estimated on today's reentry mix represent the cost a commercial down-mass operator's controlled, scheduled, repeated return would impose? Controlled, scheduled down-mass returns may be more predictable than uncontrolled decay, which would lower the prediction-uncertainty intensity and shrink the closure, but they would also be more frequent and more clustered around specific recovery sites, which would raise local traffic density. The net sign is an open empirical question that only a record containing real down-mass events can settle. Stating it as open, rather than assuming the launch-era estimate transfers, is the discipline the design imposes.

### 8.4.3 Integration with corridor-pricing and authorization design

The third track connects the measured cost to the institutions that would act on it, which is the point of the measurement in North's terms: the estimate is the input that lets the authorization rule change, and the rule change is where the economic value is [\[24\]](#ref-24). The future program here is the design of the corridor-authorization and fee instruments that the estimate would parameterize. A measured per-event NAS cost is the price on the reentry-to-aviation externality, and once that price exists, the institutional question becomes how to internalize it: through a corridor-access fee that confronts the reentry operator with the aviation cost it imposes, through a prediction-accuracy standard that conditions a narrower closure on demonstrated trajectory precision [\[9\]](#ref-9), or through a dynamic-closure rule that narrows the withheld volume as the prediction tightens and prices the residual risk against the cost saving. The estimate is the necessary precondition for any of these instruments, because a regulator cannot weigh a narrower closure's residual risk against its cost saving without a measured cost of the wider closure. North's transaction-cost logic makes the point: a lower-cost institution can only displace a higher-cost one when the price of the disruption the higher-cost rule imposes is known [\[24\]](#ref-24), and the broader space-economy literature locates market formation in institutional design and externality pricing rather than technology readiness alone [\[16\]](#ref-16), [\[18\]](#ref-18), [\[54\]](#ref-54). Designing the instrument is a separate research program from estimating the cost, and this dissertation delivers the estimate, not the instrument.

This integration track also has a governance and registration dimension that the forward agenda must engage, because a corridor-pricing or dynamic-closure regime presupposes the data infrastructure to identify and characterize reentering objects in near-real time. The space-traffic-management literature is converging on the recognition that the registration and tracking practices underpinning reentry prediction are themselves incomplete, and that improving them is a precondition for the kind of prediction-informed dynamic management whose avoided cost this dissertation estimates [\[80\]](#ref-80), [\[79\]](#ref-79). The mechanism is direct: a dynamic closure can only be sized to a late, tight prediction if the object is tracked well enough to produce that prediction, so the avoided-cost lever depends on the same registration and observation improvements that the governance literature is independently calling for. The forward program should therefore treat the prediction surface and the registration regime as joint, not separable: the value of the avoided-cost parameter to a policymaker is partly a value of investing in the tracking and registration that make tight predictions possible. Confidence that the registration agenda is complementary to the avoided-cost mechanism is high, because both act on the same prediction-uncertainty quantity; the open question the forward program inherits is the cost-effectiveness frontier of registration investment against closure-cost savings, a benefit-cost study the present design does not perform but clearly motivates.

A final element of the integration track is the cross-domain transfer of governance lessons. The air-traffic-management system solved, decades ago, the problem this dissertation studies in a different guise: how to integrate heterogeneous, intermittently hazardous operations into a shared, congestible airspace without segregating them statically and permanently. The analyses that draw cross-domain lessons from air traffic management for space traffic management are the natural reference for an authorization institution that wants to adapt rather than reinvent [\[60\]](#ref-60), and the emerging manual and handbook treatments of space traffic management governance are the venues in which a measured externality price would enter policy [\[59\]](#ref-59), [\[54\]](#ref-54). The forward program is to carry the estimate into those governance discussions as the quantitative anchor they currently lack. The contribution of this dissertation to that program is specific and bounded: it supplies the number, design-based and falsifiable, on which the institutional design can rest.

## 8.5 Closing

The answer this dissertation carries, restated one last time and unchanged from where it began, is that reentry imposes a measurable, positive, per-event cost on the National Airspace System, and that better prediction combined with dynamic closure avoids a quantifiable share of it. The argument is pressing because reentry cadence is rising with large-constellation deorbit and the emergence of a commercial down-mass sector, so a policy of large static closures becomes a structural drag on the NAS rather than a rare contingency [\[1\]](#ref-1), [\[62\]](#ref-62). The design anticipates, as an expectation and not as an executed result, a positive disruption parameter and a positive avoided-cost parameter, because every mechanism in the cited literature points the same direction: a larger withheld volume, a longer window, and denser displaced traffic all raise the cost, and prediction uncertainty, which sizes the closure, falls sharply as the object nears reentry, which is the lever the avoided-cost parameter is built to measure [\[4\]](#ref-4), [\[9\]](#ref-9), [\[21\]](#ref-21), [\[22\]](#ref-22). The action that follows is to give regulators a defensible price for the reentry-to-aviation externality and to give a prediction-and-modeling investment such as ReentryFlow a target avoided cost to clear.
The uncertainty posture is the one this chapter has insisted on throughout. No estimate has been executed on the full dataset; the panel the design joins does not yet exist; every quantity in the analysis plan is a symbol awaiting a number, and the numbers that appear in the preceding chapters are priors and illustrations, never findings. What stands at the design stage is the apparatus: the reframing of a reentry closure as a causal treatment, the Callaway and Sant'Anna identification strategy with its staggered-timing diagnostics, the four-layer data architecture, the pre-committed decision rule, and the externality-and-institutions logic that makes the eventual estimate a price rather than a curiosity [\[14\]](#ref-14), [\[23\]](#ref-23), [\[24\]](#ref-24). The argument holds at the design stage on its first three parts: that the cost is real, that it is material, and that the design addresses the causal mechanism. It commits the executed work to demonstrate the remaining two: that dynamic closure beats static segregation on cost at equal safety, and that the residual risk of a narrower closure is acceptable and bounded by validated prediction accuracy.

The design cannot confirm itself by construction. A precisely estimated near-zero disruption effect would falsify the first part of the contribution, and a near-zero avoided-cost effect would falsify the second, implying that better reentry prediction does not reduce airspace cost and that dynamic closure buys nothing over static closure on cost grounds. Either outcome is a publishable, decision-relevant result, which is the mark of a falsifiable design. The value of the work to NASA, to JPL, and to the wider civil-space enterprise is the same whether the measured effect is large or small: a defensible, design-based number for the cost that reentry imposes on aviation, and for the cost that better prediction avoids, is the evidence base on which any reentry-authorization or corridor-pricing institution must rest. This dissertation does not claim to hold that number yet. It claims to have built, in full and in the open, the means to obtain it, and to have shown why obtaining it is worth doing in either direction. The work is offered in that spirit of stewardship. The skies are a shared inheritance, and the duty to govern their growing traffic wisely, neither closing them in needless caution nor opening them in haste, is one that only an honest, measured number can fulfill.
# References

The reference list is compiled directly from the project bibliography in a single consistent IEEE-numbered style. The numbering is canonical across the whole dissertation: a citation [N] in any chapter resolves to entry [N] here, and every in-text citation used in the chapters appears in this list. Every entry is a real, deduplicated source carrying a clickable digital object identifier, or, where a DOI was never minted, a clickable resolver URL; no entry is invented. Of the 89 entries, 60 are grade-A peer-reviewed sources and 29 are grade-B reports, preprints, or working papers, with the grade governing the epistemic weight each source carries in the argument.

<span id="ref-1"></span>[1] S. Wright, et al., "Airspace closures due to reentering space objects," *Scientific Reports*, 2025. DOI: [10.1038/s41598-024-84001-2](https://doi.org/10.1038/s41598-024-84001-2).

<span id="ref-2"></span>[2] A. Srivastava, T. J. St. Clair, S. Zobell, D. Fulmer, "Assessing impact of space launch and reentry operations on the national airspace system (NAS) using historical traffic patterns," *Symposium on Dependable, Autonomic and Secure Computing (DASC)*, 2015. DOI: [10.1109/dasc.2015.7311651](https://doi.org/10.1109/dasc.2015.7311651).

<span id="ref-3"></span>[3] J. K. Tinoco, C. Yu, R. Firmo, C. A. Castro, M. Moallemi, R. J. Babb, "Sharing airspace: Simulation of commercial space horizontal launch impacts on airlines and finding solutions," *Journal of Space Safety Engineering*, 2021. DOI: [10.1016/j.jsse.2021.02.001](https://doi.org/10.1016/j.jsse.2021.02.001).

<span id="ref-4"></span>[4] D. Young, "Effects of Future Launch and Reentry Operations on the National Airspace System," *Journal of Air Transportation*, 2017. DOI: [10.2514/1.d0039](https://doi.org/10.2514/1.d0039).

<span id="ref-5"></span>[5] N. Robson, T. Bolic, A. Cook, "ATM strategies for, and impacts of, space launches," *Journal of Physics: Conference Series*, 2024. DOI: [10.1088/1742-6596/2716/1/012083](https://doi.org/10.1088/1742-6596/2716/1/012083).

<span id="ref-6"></span>[6] T. Rabu, "Handling of external risks, including launch and re-entry events, in the aviation and maritime sector," *Journal of Space Safety Engineering*, 2024. DOI: [10.1016/j.jsse.2024.04.003](https://doi.org/10.1016/j.jsse.2024.04.003).

<span id="ref-7"></span>[7] S. Hilton, R. Sabatini, A. Gardi, H. Ogawa, P. Teofilatto, "Space traffic management: Towards safe and unsegregated space transport operations," *Progress in Aerospace Sciences*, 2019. DOI: [10.1016/j.paerosci.2018.10.006](https://doi.org/10.1016/j.paerosci.2018.10.006).

<span id="ref-8"></span>[8] K. Thangavel, N. E.-D. Safwat, A. Gardi, R. Sabatini, "Multi-domain traffic management: Toward integrated air and space transport operations," *IEEE Aerospace and Electronic Systems Magazine*, 2025. DOI: [10.1109/maes.2025.3555246](https://doi.org/10.1109/maes.2025.3555246).

<span id="ref-9"></span>[9] L. A. Weitz, T. Gruber, N. E. Rozen, "Predicted trajectory accuracy requirements to reduce aviation impact of space launch operations," *Journal of Air Transportation*, 2024. DOI: [10.2514/1.d0426](https://doi.org/10.2514/1.d0426).

<span id="ref-10"></span>[10] S. Kaltenhaeuser, et al., "Towards Efficient Integration of Rocket Launches and Re-Entry Operations in European Airspace," *IAF Symposium on Commercial Spaceflight Safety Issues*, 2024. DOI: [10.52202/078377-0003](https://doi.org/10.52202/078377-0003).

<span id="ref-11"></span>[11] C. Mayer, T. Sinai, "Network Effects, Congestion Externalities, and Air Traffic Delays: Or Why Not All Delays Are Evil," *American Economic Review*, 2003. DOI: [10.1257/000282803769206269](https://doi.org/10.1257/000282803769206269).

<span id="ref-12"></span>[12] J. J. Rebollo, H. Balakrishnan, "Characterization and prediction of air traffic delays," *Transportation Research Part C*, 2014. DOI: [10.1016/j.trc.2014.04.007](https://doi.org/10.1016/j.trc.2014.04.007).

<span id="ref-13"></span>[13] A. Goodman-Bacon, "Difference-in-differences with variation in treatment timing," *Journal of Econometrics*, 2021. DOI: [10.1016/j.jeconom.2021.03.014](https://doi.org/10.1016/j.jeconom.2021.03.014).

<span id="ref-14"></span>[14] B. Callaway, P. H. C. Sant'Anna, "Difference-in-Differences with multiple time periods," *Journal of Econometrics*, 2020. DOI: [10.1016/j.jeconom.2020.12.001](https://doi.org/10.1016/j.jeconom.2020.12.001).

<span id="ref-15"></span>[15] C. de Chaisemartin, X. D'Haultfoeuille, "Two-way fixed effects and differences-in-differences estimators with heterogeneous treatment effects," *Journal of Econometrics*, 2023. DOI: [10.1016/j.jeconom.2023.105480](https://doi.org/10.1016/j.jeconom.2023.105480).

<span id="ref-16"></span>[16] M. Weinzierl, "Space, the Final Economic Frontier," *Journal of Economic Perspectives*, 2018. DOI: [10.1257/jep.32.2.173](https://doi.org/10.1257/jep.32.2.173).

<span id="ref-17"></span>[17] W. Peeters, "The Paradigm Shift of NewSpace: New Business Models and Growth of the Space Economy," *New Space*, 2024. DOI: [10.1089/space.2023.0060](https://doi.org/10.1089/space.2023.0060).

<span id="ref-18"></span>[18] N. Adilov, P. Alexander, "Engineering the new space economy: Market creation and institutional design," *Journal of Industrial and Business Economics*, 2026. DOI: [10.1007/s40812-026-00398-z](https://doi.org/10.1007/s40812-026-00398-z).

<span id="ref-19"></span>[19] K. Handley, N. Limao, "Trade and investment under policy uncertainty: Theory and firm evidence," *American Economic Journal: Economic Policy*, 2015. DOI: [10.1257/pol.20140068](https://doi.org/10.1257/pol.20140068).

<span id="ref-20"></span>[20] M. Vergaaij, C. R. McInnes, M. Ceriotti, "Comparison of material sources and customer locations for commercial space resource utilization," *Acta Astronautica*, 2021. DOI: [10.1016/j.actaastro.2021.03.010](https://doi.org/10.1016/j.actaastro.2021.03.010).

<span id="ref-21"></span>[21] A. D. Hayes, R. Caverly, "Atmospheric density-compensating model predictive control for targeted reentry of drag-modulated spacecraft," *Journal of Guidance, Control, and Dynamics*, 2024. DOI: [10.2514/1.g008665](https://doi.org/10.2514/1.g008665).

<span id="ref-22"></span>[22] J. Geul, E. Mooij, R. Noomen, "Analysis of Uncertainties and Modeling in Short-Term Reentry Predictions," *Journal of Guidance, Control, and Dynamics*, 2018. DOI: [10.2514/1.g003258](https://doi.org/10.2514/1.g003258).

<span id="ref-23"></span>[23] A. Rao, M. G. Burgess, D. Kaffine, "OPUS: An integrated assessment model for satellites and orbital debris / Orbital-use fees and the value of the space industry," *arXiv (working paper)*, 2023. DOI: [10.48550/arxiv.2309.10252](https://doi.org/10.48550/arxiv.2309.10252).

<span id="ref-24"></span>[24] D. C. North, "Institutions, Institutional Change and Economic Performance," *Cambridge University Press*, 1990. DOI: [10.1017/cbo9780511808678](https://doi.org/10.1017/cbo9780511808678).

<span id="ref-25"></span>[25] H. Jones, "The recent large reduction in space launch cost," *48th International Conference on Environmental Systems (ICES)*, 2018. Available: [https://ttu-ir.tdl.org/handle/2346/74082](https://ttu-ir.tdl.org/handle/2346/74082).

<span id="ref-26"></span>[26] S.-Y. Kang, M. Jo, J.-Y. Choi, S. S. Yang, "Cost effectiveness of reusable launch vehicles depending on the payload capacity," *Aerospace*, 2025. DOI: [10.3390/aerospace12050364](https://doi.org/10.3390/aerospace12050364).

<span id="ref-27"></span>[27] Jonas Radtke, Daniel Lück, Lorenz Losensky, Sven Kaltenhaeuser, Christopher Brain, Augustin Udristioiu, "Integrating Air Traffic Management and Space Traffic Management: Concepts, Challenges, and Solutions for the Evolving Aerospace Landscape in Europe," 2024. DOI: [10.2514/6.2024-4824](https://doi.org/10.2514/6.2024-4824).

<span id="ref-28"></span>[28] Bill Lash, Thomas St. Clair, Catherine Bolczak, Constance Morgan, Duane Freer, Jeffrey Richards, "Enabling Tactical Airspace Management During Space Launch and Reentry Operations Through Collaborative Information Exchange," *AIAA AVIATION 2020 FORUM*, 2020. DOI: [10.2514/6.2020-2852](https://doi.org/10.2514/6.2020-2852).

<span id="ref-29"></span>[29] Laurence H. Mutuel, Daniel Murray, "Space Data Integrator: FAA's Innovative Platform for Launch and Reentry Operations," *54th AIAA Aerospace Sciences Meeting*, 2016. DOI: [10.2514/6.2016-0218](https://doi.org/10.2514/6.2016-0218).

<span id="ref-30"></span>[30] Oliver J. Bojorquez, Jun Chen, "Risk Level Analysis for Hazard Area During Commercial Space Launch," 2019. DOI: [10.1109/dasc43569.2019.9081702](https://doi.org/10.1109/dasc43569.2019.9081702).

<span id="ref-31"></span>[31] Amal Srivastava, "Generating Realistic Reroutes to Assess Air Traffic Impact of Blocked Airspaces," 2019. DOI: [10.1109/dasc43569.2019.9081707](https://doi.org/10.1109/dasc43569.2019.9081707).

<span id="ref-32"></span>[32] Janet K. Tinoco, Noah D Eudy, Maxwell Cannon, "Simulation and Analysis of 4-D Airspace Closures due to Commercial Space Operations: Impacts on Airlines and General Aviation," *The Journal of Aviation/Aerospace Education and Research*, 2020. DOI: [10.15394/jaaer.2020.1853](https://doi.org/10.15394/jaaer.2020.1853).

<span id="ref-33"></span>[33] Tim Gruber, Lesley A. Weitz, Nicholas Rozen, "Determining Predicted Trajectory Accuracy Requirements to Reduce the Aviation Impact of Space Launch and Reentry Operations," *ASCEND 2023*, 2023. DOI: [10.2514/6.2023-4632](https://doi.org/10.2514/6.2023-4632).

<span id="ref-34"></span>[34] Zheng Tao, Ganghuai Wang, Jon Semanek, Ashley Williams, Jonathan L. Schwartz, "Assessing Factors that Affect the Safety of Space Launch and Reentry Operations in the National Airspace System," *AIAA SPACE 2016*, 2016. DOI: [10.2514/6.2016-5322](https://doi.org/10.2514/6.2016-5322).

<span id="ref-35"></span>[35] Hanif D. Sherali, Raymond W. Staats, Antonio Trani, "An Airspace Planning and Collaborative Decision-Making Model: Part I: Probabilistic Conflicts, Workload, and Equity Considerations," *Transportation Science*, 2003. DOI: [10.1287/trsc.37.4.434.23272](https://doi.org/10.1287/trsc.37.4.434.23272).

<span id="ref-36"></span>[36] Alessandro Falsone, Maria Prandini, "A Randomized Approach to Probabilistic Footprint Estimation of a Space Debris Uncontrolled Reentry," *IEEE Transactions on Intelligent Transportation Systems*, 2017. DOI: [10.1109/tits.2017.2654511](https://doi.org/10.1109/tits.2017.2654511).

<span id="ref-37"></span>[37] Irina Beatrice Stefanescu, Cristian Emil Constantinescu, Octavian Thor Pleter, "Minimizing Air Traffic Disruption from Uncontrolled Space Debris Reentries," *EASN 2024*, 2025. DOI: [10.3390/engproc2025090075](https://doi.org/10.3390/engproc2025090075).

<span id="ref-38"></span>[38] Charlotte Hook, Ewan Wright, Michael Byers, Aaron C. Boley, "Uncontrolled reentries of space objects and aviation safety," *Acta Astronautica*, 2024. DOI: [10.1016/j.actaastro.2024.05.026](https://doi.org/10.1016/j.actaastro.2024.05.026).

<span id="ref-39"></span>[39] Karthik Gopalakrishnan, Hamsa Balakrishnan, "Control and Optimization of Air Traffic Networks," *Annual Review of Control Robotics and Autonomous Systems*, 2021. DOI: [10.1146/annurev-control-070720-080844](https://doi.org/10.1146/annurev-control-070720-080844).

<span id="ref-40"></span>[40] Pablo Fleurquin, José J. Ramasco, Victor M. Eguı́luz, "Characterization of Delay Propagation in the US Air-Transportation Network," *Transportation Journal*, 2014. DOI: [10.5325/transportationj.53.3.0330](https://doi.org/10.5325/transportationj.53.3.0330).

<span id="ref-41"></span>[41] Haoyu Zhang, Weiwei Wu, Shengrun Zhang, Frank Witlox, "Simulation Analysis on Flight Delay Propagation Under Different Network Configurations," *IEEE Access*, 2020. DOI: [10.1109/access.2020.2999098](https://doi.org/10.1109/access.2020.2999098).

<span id="ref-42"></span>[42] Qinggang Wu, Minghua Hu, Xiaozhen Ma, Yaniun Wang, Wei Cong, Daniel Delahaye, "Modeling Flight Delay Propagation in Airport and Airspace Network," *2018 21st International Conference on Intelligent Transportation Systems (ITSC)*, 2018. DOI: [10.1109/itsc.2018.8569657](https://doi.org/10.1109/itsc.2018.8569657).

<span id="ref-43"></span>[43] Pablo Fleurquin, José J. Ramasco, Victor M. Eguı́luz, "Systemic delay propagation in the US airport network," *Scientific Reports*, 2013. DOI: [10.1038/srep01159](https://doi.org/10.1038/srep01159).

<span id="ref-44"></span>[44] Chi Li, Jianfeng Mao, Lingyi Li, Jingxuan Wu, Lianmin Zhang, Jianyu Zhu, Zibin Pan, "Flight delay propagation modeling: Data, Methods, and Future opportunities," *Transportation Research Part E Logistics and Transportation Review*, 2024. DOI: [10.1016/j.tre.2024.103525](https://doi.org/10.1016/j.tre.2024.103525).

<span id="ref-45"></span>[45] Christopher Mayer, Todd Sinai, "Network Effects, Congestion Externalities, and Air Traffic Delays: Or Why All Delays Are Not Evil," *National Bureau of Economic Research*, 2002. DOI: [10.3386/w8701](https://doi.org/10.3386/w8701).

<span id="ref-46"></span>[46] Coady Wing, Seth Freedman, Alex Hollingsworth, "Stacked Difference-in-Differences," *National Bureau of Economic Research*, 2024. DOI: [10.3386/w32054](https://doi.org/10.3386/w32054).

<span id="ref-47"></span>[47] Brantly Callaway, Andrew Goodman-Bacon, Pedro H. C. Sant’Anna, "Difference-in-Differences with a Continuous Treatment," *arXiv (Cornell University)*, 2021. DOI: [10.48550/arxiv.2107.02637](https://doi.org/10.48550/arxiv.2107.02637).

<span id="ref-48"></span>[48] Kyle Butts, "Difference-in-Differences Estimation with Spatial Spillovers," *arXiv (Cornell University)*, 2021. DOI: [10.48550/arxiv.2105.03737](https://doi.org/10.48550/arxiv.2105.03737).

<span id="ref-49"></span>[49] Jonathan Roth, "Interpreting Event-Studies from Recent Difference-in-Differences Methods," *arXiv (Cornell University)*, 2024. DOI: [10.48550/arxiv.2401.12309](https://doi.org/10.48550/arxiv.2401.12309).

<span id="ref-50"></span>[50] Andrew C. Baker, David F. Larcker, Charles C. Y. Wang, "How much should we trust staggered difference-in-differences estimates?," *Journal of Financial Economics*, 2022. DOI: [10.1016/j.jfineco.2022.01.004](https://doi.org/10.1016/j.jfineco.2022.01.004).

<span id="ref-51"></span>[51] Zachary Porreca, "Synthetic difference-in-differences estimation with staggered treatment timing," *Economics Letters*, 2022. DOI: [10.1016/j.econlet.2022.110874](https://doi.org/10.1016/j.econlet.2022.110874).

<span id="ref-52"></span>[52] Clément de Chaisemartin, Xavier D’Haultfœuille, "Difference-in-Differences Estimators of Intertemporal Treatment Effects," *The Review of Economics and Statistics*, 2021. DOI: [10.1162/rest_a_01414](https://doi.org/10.1162/rest_a_01414).

<span id="ref-53"></span>[53] Jaedong Seong, Okchul Jung, Youeyun Jung, Daewon Chung, "Analysis of Reentry Prediction of CZ-5B Rocket Body," *Journal of Space Technology and Applications*, 2021. DOI: [10.52912/jsta.2021.1.2.149](https://doi.org/10.52912/jsta.2021.1.2.149).

<span id="ref-54"></span>[54] Carmen Pardini, Luciano Anselmo, "Performance Evaluation of Atmospheric Density Models for Satellite Reentry Predictions with High Solar Activity Levels.," *TRANSACTIONS OF THE JAPAN SOCIETY FOR AERONAUTICAL AND SPACE SCIENCES*, 2003. DOI: [10.2322/tjsass.46.42](https://doi.org/10.2322/tjsass.46.42).

<span id="ref-55"></span>[55] David J. Gondelach, Roberto Armellin, Aleksander Lidtke, "Ballistic Coefficient Estimation for Reentry Prediction of Rocket Bodies in Eccentric Orbits Based on TLE Data," *Mathematical Problems in Engineering*, 2017. DOI: [10.1155/2017/7309637](https://doi.org/10.1155/2017/7309637).

<span id="ref-56"></span>[56] Denny M. Oliveira, E. Zesta, Katherine Garcia‐Sage, "Tracking reentries of Starlink satellites during the rising phase of solar cycle 25," *Frontiers in Astronomy and Space Sciences*, 2025. DOI: [10.3389/fspas.2025.1572313](https://doi.org/10.3389/fspas.2025.1572313).

<span id="ref-57"></span>[57] Anelí Bongers, José L. Torres, "Orbital debris and the market for satellites," *Ecological Economics*, 2023. DOI: [10.1016/j.ecolecon.2023.107831](https://doi.org/10.1016/j.ecolecon.2023.107831).

<span id="ref-58"></span>[58] Gordon C. Rausser, Elliot Choi, Alexandre M. Bayen, "Public-private partnerships in fostering outer space innovations," *Proceedings of the National Academy of Sciences*, 2023. DOI: [10.1073/pnas.2222013120](https://doi.org/10.1073/pnas.2222013120).

<span id="ref-59"></span>[59] R. J. Briggs, Mann Virdee, "The Economics and Policy of Space Debris and Orbital Management," *The Oxford Handbook of the New Space Economy*, 2026. DOI: [10.1093/oxfordhb/9780198881049.013.0015](https://doi.org/10.1093/oxfordhb/9780198881049.013.0015).

<span id="ref-60"></span>[60] Petr Bohacek, Simon P. Worden, Kyran Grattan, "Benefit-Sharing as Investment Protection for Space Resource Utilization," *New Space*, 2022. DOI: [10.1089/space.2021.0050](https://doi.org/10.1089/space.2021.0050).

<span id="ref-61"></span>[61] Anon., "Cross-Domain Lessons for Space Traffic Management: An Analysis of Air and Maritime Treaty Governance Mechanisms," 2023. DOI: [10.7249/rra2208-2](https://doi.org/10.7249/rra2208-2).

<span id="ref-62"></span>[62] Stephan Hobe, "The Cologne Manual on Space Traffic Management (CM-STM) : co-operation project between the Institute of Air Law, Space Law and Cyber Law at the University of Cologne and the German Aerospace Center (DLR)," *Kölner Universitäts PublikationsServer (Universität zu Köln)*, 2026. DOI: [10.18716/cm25](https://doi.org/10.18716/cm25).

<span id="ref-63"></span>[63] C. Pardini, L. Anselmo, "On the need to assess and mitigate the risk from uncontrolled re-entries," *Acta Astronautica*, 2024. DOI: [10.1016/j.actaastro.2024.03.057](https://doi.org/10.1016/j.actaastro.2024.03.057).

<span id="ref-64"></span>[64] Pardini C., "The risk on the ground and in the airspace posed by uncontrolled re-entries: Should the growth observed in recent years be considered worrying?," *Acta Astronautica*, 2026. DOI: [10.1016/j.actaastro.2026.01.049](https://doi.org/10.1016/j.actaastro.2026.01.049).

<span id="ref-65"></span>[65] S.-H. Park, J. Navarro Laboulais, P. Leyland, S. Mischler, "Re-entry survival analysis and ground risk assessment of space debris," *Acta Astronautica*, 2020. DOI: [10.1016/j.actaastro.2020.09.034](https://doi.org/10.1016/j.actaastro.2020.09.034).

<span id="ref-66"></span>[66] L. Niccolai, G. Mengali, "Decay time estimate for LEO spacecraft," *Acta Astronautica*, 2024. DOI: [10.1016/j.actaastro.2024.09.045](https://doi.org/10.1016/j.actaastro.2024.09.045).

<span id="ref-67"></span>[67] M. Sorge, D. Alary, F. Lacomba, H. Tung, B. Singh, "Space traffic management: Large constellations," *Acta Astronautica*, 2025. DOI: [10.1016/j.actaastro.2025.01.043](https://doi.org/10.1016/j.actaastro.2025.01.043).

<span id="ref-68"></span>[68] C. Colombo, P. Martinez, F. Letizia, B. Cattani, E. David, D. Escobar, "Space capacity management and its interaction with space traffic management," *Acta Astronautica*, 2025. DOI: [10.1016/j.actaastro.2025.01.069](https://doi.org/10.1016/j.actaastro.2025.01.069).

<span id="ref-69"></span>[69] Susan Athey, Guido Imbens, "Design-based Analysis in Difference-In-Differences Settings with Staggered Adoption," 2018. DOI: [10.3386/w24963](https://doi.org/10.3386/w24963).

<span id="ref-70"></span>[70] Xiaofeng Li, Lihua Lin, "Identification of quantile treatment effects in difference-in-differences settings with staggered adoption," *Economics Letters*, 2024. DOI: [10.1016/j.econlet.2024.111792](https://doi.org/10.1016/j.econlet.2024.111792).

<span id="ref-71"></span>[71] Xuliang Wang, "Modern Staggered Difference-in-Differences: From the Pitfalls of Two-Way Fixed Effects (TWFE) to Robust Estimation," 2025. DOI: [10.2139/ssrn.5456874](https://doi.org/10.2139/ssrn.5456874).

<span id="ref-72"></span>[72] Samantha Bielen, "Prosecutors and crime deterrence: Evidence from a difference-in-differences analysis with staggered treatment," *Journal of Criminal Justice*, 2024. DOI: [10.1016/j.jcrimjus.2023.102147](https://doi.org/10.1016/j.jcrimjus.2023.102147).

<span id="ref-73"></span>[73] Phuc V. Nguyen, "National AI Strategies and Macroeconomic Performance A Staggered Difference-in-Differences Approach," 2026. DOI: [10.2139/ssrn.6580199](https://doi.org/10.2139/ssrn.6580199).

<span id="ref-74"></span>[74] Roger Beatty, Rose Hsu, Lee Berry, James Rome, "Preliminary Evaluation of Flight Delay Propagation through an Airline Schedule," *Air Traffic Control Quarterly*, 1999. DOI: [10.2514/atcq.7.4.259](https://doi.org/10.2514/atcq.7.4.259).

<span id="ref-75"></span>[75] Nabin Kafle, Bo Zou, "Modeling flight delay propagation: A new analytical-econometric approach," *Transportation Research Part B: Methodological*, 2016. DOI: [10.1016/j.trb.2016.08.012](https://doi.org/10.1016/j.trb.2016.08.012).

<span id="ref-76"></span>[76] Vaggelis Giannikas, Anna Ledwoch, Goran Stojković, Pablo Costas, Alexandra Brintrup, Ahmed Ali Saeed Al-Ali, Vinod Kumar Chauhan, Duncan Mcfarlane, "A data-driven method to assess the causes and impact of delay propagation in air transportation systems," *Transportation Research Part C Emerging Technologies*, 2022. DOI: [10.1016/j.trc.2022.103862](https://doi.org/10.1016/j.trc.2022.103862).

<span id="ref-77"></span>[77] Mazhar Arıkan, Vinayak Deshpande, Milind Sohoni, "Building Reliable Air-Travel Infrastructure Using Empirical Data and Stochastic Models of Airline Networks," *Operations Research*, 2013. DOI: [10.1287/opre.1120.1146](https://doi.org/10.1287/opre.1120.1146).

<span id="ref-78"></span>[78] Ralph Tüllmann, Christian Arbinger, Stuart Baskcomb, Jens Berdermann, Hauke Fiedler, Erich Klock, Thomas Schildknecht, "On the Implementation of a European Space Traffic Management System I. A White Paper.," *Bern Open Repository and Information System (University of Bern)*, 2018. DOI: [10.7892/boris.111870](https://doi.org/10.7892/boris.111870).

<span id="ref-79"></span>[79] James Torres, P. Röhl, Jimmy Krozel, Terry L. Thompson, "A Dynamic Air Traffic Management Approach to Operationally Responsive Space," *AIAA Atmospheric Flight Mechanics Conference and Exhibit*, 2007. DOI: [10.2514/6.2007-6632](https://doi.org/10.2514/6.2007-6632).

<span id="ref-80"></span>[80] Sven Kaltenhäuser, Tobias Rabus, D W Freer, Dan Bogdan, "Uncontrolled Re-entry Risk for Aviation and the benefits of real time information services," *elib (German Aerospace Center)*, 2026. DOI: [10.60575/e98z-c874](https://doi.org/10.60575/e98z-c874).

<span id="ref-81"></span>[81] WanTong Chen, ZeRui Cao, ShiYu Ren, "Enhanced Covariance Propagation for Asymmetric High-Order Uncertainty in Reentry Debris Risk Prediction," 2025. DOI: [10.21203/rs.3.rs-6731968/v1](https://doi.org/10.21203/rs.3.rs-6731968/v1).

<span id="ref-82"></span>[82] E. K. Sutton, J. M. Forbes, R. S. Nerem, "Global thermospheric neutral density and wind response to the severe 2003 geomagnetic storms from CHAMP accelerometer data," *Journal of Geophysical Research Atmospheres*, 2005. DOI: [10.1029/2004ja010985](https://doi.org/10.1029/2004ja010985).

<span id="ref-83"></span>[83] G. M. Keating, R. H. Tolson, M. S. Bradford, "Evidence of long term global decline in the Earth's thermospheric densities apparently related to anthropogenic effects," *Geophysical Research Letters*, 2000. DOI: [10.1029/2000gl003771](https://doi.org/10.1029/2000gl003771).

<span id="ref-84"></span>[84] Ayisha M Ashruf, Ankush Bhaskar, Vineeth, C, Pant, Tarun Kumar, "Deciphering Solar Cycle Influence on Long-Term Orbital Deterioration of Low-Earth Orbiting Space Debris," *arXiv (Cornell University)*, 2024. DOI: [10.48550/arxiv.2405.08837](https://doi.org/10.48550/arxiv.2405.08837).

<span id="ref-85"></span>[85] Kenneth A. Small, José A Gómez-Ibáñez, "Road Pricing for Congestion Management: The Transition from Theory to Policy," 2005. DOI: [10.4324/9780203985359-27](https://doi.org/10.4324/9780203985359-27).

<span id="ref-86"></span>[86] Ian Parry, Kenneth A. Small, "Does Britain or the United States Have the Right Gasoline Tax?," *American Economic Review*, 2005. DOI: [10.1257/0002828054825510](https://doi.org/10.1257/0002828054825510).

<span id="ref-87"></span>[87] Smriti Tandon, Antonia Puccio, Subhankar Das, "Space Accommodation Operations: Maximizing Efficiency in Orbital Facilities," *Pioneering the New Space Economy through AI and Immersive Technologies*, 2025. DOI: [10.1007/978-981-96-5977-7_8](https://doi.org/10.1007/978-981-96-5977-7_8).

<span id="ref-88"></span>[88] N. Sanchez-Ortiz, D. Wood, L. J. Smith, C. Gregori De La Malla, M. Strah, M. Huelsmann, "Outreach on Space Traffic Management," *Acta Astronautica*, 2025. DOI: [10.1016/j.actaastro.2025.01.031](https://doi.org/10.1016/j.actaastro.2025.01.031).

<span id="ref-89"></span>[89] T. Masson-Zwaan, P. Martinez, F. Letizia, C. Melograna, M. Reynders, R. Rovetto, "The need to improve registration practices in the context of space traffic management," *Acta Astronautica*, 2024. DOI: [10.1016/j.actaastro.2024.06.052](https://doi.org/10.1016/j.actaastro.2024.06.052).


# Appendices

## Appendix A: Variable and Data Dictionary

This appendix expands the notation block of Chapter 4 into a complete dictionary so that the construction of every quantity in the estimating equation is unambiguous. The dictionary is the load-bearing artifact of the backmatter, because the falsifiability of the contribution rests on it: a reader who disputes the disruption parameter must be able to see exactly how the outcome was built, from which source, on which scale, and with which expected sign, and only a dictionary at this resolution permits that. The unit of analysis throughout is the airspace-sector-by-time-window cell, indexed \(i\), with windows indexed \(t\); the cell is the row of the panel, and every variable below is measured at the cell level except the treatment-defining event attributes, which are inherited from the reentry event that generates the closure intersecting that cell.

| Symbol / name | Construct | Operational definition | Source dataset | Scale | Expected sign |
|---|---|---|---|---|---|
| \(Y_{i,t}\) (delay) | Disruption cost in time | Realized delay minutes for the cell's flights relative to schedule and to an undisrupted baseline | FAA NAS / SWIM | Minutes, continuous, non-negative | Positive under treatment |
| \(Y_{i,t}\) (distance) | Disruption cost in routing | Added flown distance relative to the great-circle or filed route, summed over the cell | FAA NAS / SWIM | Nautical miles, continuous, non-negative | Positive under treatment |
| \(Y_{i,t}\) (fuel) | Disruption cost in fuel | Modeled fuel burn from added distance and aircraft type | FAA NAS / SWIM + aircraft-class factors | Kilograms or pounds, continuous | Positive under treatment |
| \(Y_{i,t}\) (DOC) | Disruption cost in dollars | Direct operating cost from per-block-hour and per-nautical-mile factors by aircraft class | FAA NAS / SWIM + standard cost factors | US dollars, continuous | Positive under treatment |
| \(D_{i,t}\) | Treatment indicator | One if a reentry-driven Aircraft Hazard Area is active over the cell's sector during the window, else zero | EU SST footprint intersected with cell via ReentryFlow | Binary | Defines treatment |
| \(G_i\) | Treatment cohort | Window in which cell \(i\) first receives a reentry closure | Derived from EU SST event sequence | Categorical (window index) | Grouping variable |
| \(U_{i}\) | Treatment intensity | Prediction-uncertainty bound (along-track and cross-track dispersion) at closure-decision time, which scales closure size | EU SST prediction bulletin | Kilometers, continuous, non-negative | Larger value to larger cost |
| \(\rho_{i,t}\) | Baseline traffic density | Sector flight count per unit time in the undisrupted baseline | FAA NAS / SWIM | Flights per hour | Larger value to larger cost |
| \(W_{i,t}\) | Weather severity | Concurrent convective or wind-shear severity index over the sector-window | FAA / external weather feed | Ordinal or continuous index | Confounder, controlled |
| \(M_{i,t}\) | Concurrent TMIs | Indicator and count of non-space traffic-management initiatives active in the cell | FAA NAS / SWIM | Count | Confounder, controlled |
| \(A_{i,t}\) | Aircraft-class mix | Shares of general aviation, domestic carrier, and international carrier flights in the cell | FAA NAS / SWIM | Proportions summing to one | Effect modifier |
| Calendar controls | Temporal fixed structure | Time-of-day, day-of-week, season indicators | Derived from window timestamp | Categorical | Nuisance, absorbed |

The disruption parameter is the aggregated average treatment effect on the treated across the cost outcomes; the avoided-cost parameter is the per-event difference between realized static-closure cost and the ReentryFlow-simulated dynamic-closure cost, averaged over events, and estimated again using \(U_i\) as a continuous intensity. The aircraft-class mix \(A_{i,t}\) is retained as an effect modifier rather than a mere control because the launch-era simulation evidence shows the impact distribution differs sharply across classes, with international carriers near 8 to 10 percent and general aviation near one third of affected flights [\[3\]](#ref-3), [\[32\]](#ref-32); treating that heterogeneity as a control would discard a finding the design is built to recover. The treatment intensity \(U_i\) carries measurement error because the footprint and its dispersion are model outputs rather than observations, so the EU SST uncertainty bound is propagated into the exposure construction rather than treated as exact; the expected sign on \(U_i\) is positive because a larger dispersion forces a larger conservative closure, which is the mechanism the avoided-cost lever later reverses.

## Appendix B: Derivations

This appendix derives the two estimands so that the chapter equations are self-contained. The first derivation is the aggregation of the group-time average treatment effects into the disruption parameter. Following the design that Chapter 5 fixes, the primitive object is the group-time effect \(ATT(g,t) = E[\,Y_{i,t}(g) - Y_{i,t}(0) \mid G_i = g\,]\), the expected difference between the realized outcome for cells first treated in window \(g\) and their never-treated potential outcome \(Y_{i,t}(0)\), among cells in cohort \(g\). Each \(ATT(g,t)\) is identified against not-yet-treated and never-treated control cells under conditional parallel trends, which is what makes the comparison clean and avoids the forbidden comparisons that bias two-way fixed-effects estimation under staggered timing [\[13\]](#ref-13), [\[14\]](#ref-14), [\[50\]](#ref-50). The overall disruption parameter is a weighted average \(\theta = \sum_{g}\sum_{t \ge g} w(g,t)\, ATT(g,t)\), where the weights \(w(g,t)\) are non-negative, sum to one, and are chosen to reflect cohort sizes and the event-time structure rather than the implicit and possibly negative regression weights of two-way fixed effects. The event-study profile is the same family of effects re-indexed by time-since-treatment \(e = t - g\), giving \(\theta_e = \sum_{g} w_g\, ATT(g, g+e)\), so that the leads \(e < 0\) test for anticipation and pre-trends and the lags \(e \ge 0\) trace the dynamic response and its decay. Reading \(\theta\) as the per-event NAS cost is justified because, under the maintained identification, each \(ATT(g,t)\) is a causal contrast and their valid-weighted average is therefore causal; the caution is that the average is over the observed cohorts and event times, so \(\theta\) is local to the realized cadence and event mix, which is the design-based external-validity caution carried throughout the dissertation.

The second derivation is the avoided-cost differencing identity. Let \(C^{S}_{k}\) denote the realized cost of event \(k\) under the static-closure regime, constructed from the cost outcomes above over all cells the event's static Aircraft Hazard Area intersects, and let \(C^{D}_{k}\) denote the cost of the same event under the prediction-informed dynamic-closure counterfactual, recomputed by ReentryFlow with the hazard volume sized to the late-available uncertainty rather than the worst-case envelope. The per-event avoided cost is \(\Delta_k = C^{S}_{k} - C^{D}_{k}\), and the avoided-cost parameter is the average \(\bar{\Delta} = \frac{1}{K}\sum_{k=1}^{K}\Delta_k\) over the \(K\) events with U.S.-airspace-intersecting footprints. This identity is exact by construction given the two cost evaluations, so the inferential burden falls entirely on the validity of the counterfactual \(C^{D}_{k}\), which is why ReentryFlow's precision and recall against realized exposure are reported before any \(\bar{\Delta}\) is trusted; \(\bar{\Delta}\) measures the value of the dynamic regime because \(C^{D}_{k}\) holds the event fixed and varies only the closure policy, isolating the policy contrast. The program-level avoided cost is \(\bar{\Delta}\) multiplied by the projected reentry cadence over the planning horizon, an extrapolation whose confidence is moderate and downgraded because the cadence-to-cost relationship is plausibly convex at frequencies the present record does not contain [\[12\]](#ref-12). The continuous-intensity estimate substitutes for this differencing where events vary naturally in \(U_i\): there the avoided cost is read off the estimated dose-response of cost to prediction uncertainty rather than from a paired counterfactual, and the two estimates serve as a mutual check.

## Appendix C: Instrument and Query Details

The EU SST reentry catalog is queried for the fields that build the treatment: the predicted reentry epoch and its uncertainty window, the predicted ground-track polygon and dispersion footprint, the object identity and physical characteristics where known, and the realized reentry time and location issued in the post-event bulletin. Access is through the EU SST portal and its reentry-bulletin feed under the candidate program's vault credentials. The FAA layer draws on the System Wide Information Management feeds and the associated NAS operations archives for filed and flown trajectories, departure and arrival times against schedule, ground and airborne delay, reroute advisories and traffic-management initiatives, and the activation records of special-use and hazard airspace; the Space Data Integrator platform is the FAA capability that automates the launch-and-reentry data path into the NAS and is the documented basis for treating these feeds as the authoritative operational record [\[29\]](#ref-29). ReentryFlow's validation metrics are defined as the precision and recall of its predicted exposed-flight set against the realized FAA exposure for past events: precision is the share of flights ReentryFlow predicts exposed that were in fact exposed, and recall is the share of truly exposed flights it captures, with both reported before the model's dynamic-closure counterfactuals are used.

The research sweep that produced this bibliography is logged for transparency. Local brains were queried first: the Hall of Shoulders anchor brains for Rao, North, and Angrist-Pischke framing; the ACTA_Papers brain, which contributed the verified Acta Astronautica DOIs on uncontrolled reentry, ground risk, demise, and space-traffic management [\[38\]](#ref-38), [\[63\]](#ref-63), [\[64\]](#ref-64), [\[65\]](#ref-65), [\[66\]](#ref-66), [\[67\]](#ref-67), [\[68\]](#ref-68), [\[88\]](#ref-88), [\[89\]](#ref-89); the Space Economy Papers brain, which confirmed the Weinzierl and space-economy framing already present in the seed; and the Space Brain, which returned internal NORTH STAR and ReentryFlow context flagged as non-citable per the brain-trust catalogue. Vault APIs were then swept across nineteen facet queries: OpenAlex and Crossref (both reachable) carried the bulk of recovery and DOI validation, NTRS added NASA grey-literature coverage, and the keyed Scopus channel contributed one unique selection. Semantic Scholar, arXiv, and Springer Metadata were attempted but unavailable in the sweep environment (an expired certificate, a read timeout, and a 401 entitlement response respectively); their unavailability is recorded here for completeness and did not block coverage, because OpenAlex and Crossref captured the arXiv and Springer DOIs by other routes. Two non-Crossref DOIs (the Cologne Manual [\[62\]](#ref-62) and the real-time reentry-risk service [\[80\]](#ref-80)) were confirmed against the DataCite registry with matching titles, and IEEE DOIs that returned anti-bot status codes on publisher HEAD requests were confirmed registered via the Crossref works API. The final corpus is 26 prospectus-seed references carried verbatim plus 63 sweep-selected references, deduplicated by lowercased DOI and then normalized title, for the 89 entries listed above.

## Appendix D: Supplementary Tables (Specified-but-Unpopulated Templates)

The result tables below are specified at the design stage and deliberately left unpopulated, in keeping with the standing guardrail that no estimate is executed on the full dataset. They are reproduced here as the templates the executed analysis will fill, so that the reporting format is fixed in advance and cannot be retrofitted to the outcome.

**Table D.1. Disruption parameter, aggregated ATT by cost outcome (template).** Columns: outcome (delay minutes, added distance, fuel burn, direct operating cost); aggregated ATT point estimate; cluster-robust standard error (sector-level); 95 percent confidence interval; minimum detectable effect at the available sample size. Rows: one per cost outcome, plus a row reporting the Goodman-Bacon decomposition share and the de Chaisemartin and D'Haultfoeuille robust estimate as diagnostics [\[13\]](#ref-13), [\[15\]](#ref-15).

**Table D.2. Event-study profile (template).** Columns: event time \(e\) from a defined pre-window to a defined post-window; \(\theta_e\) point estimate; confidence interval. The pre-treatment rows test anticipation and parallel pre-trends; the post-treatment rows trace the response and its decay, whose persistence beyond the closed window is read against the delay-propagation evidence [\[12\]](#ref-12), [\[44\]](#ref-44).

**Table D.3. Avoided-cost parameter (template).** Columns: estimation route (static-versus-dynamic differencing; continuous prediction-uncertainty intensity); per-event mean \(\bar{\Delta}\); standard error; implied program-level avoided cost at the projected cadence. A row reports the ReentryFlow precision and recall that gate the counterfactual.

**Table D.4. Robustness and placebo battery (template).** Rows: weather placebo, adjacent-sector placebo, controlled-reentry exclusion, anticipation-lead inspection, alternative cost factors. Columns: the re-estimated disruption parameter and whether the conclusion is unchanged. A near-zero placebo and a stable main estimate across rows are the conditions under which the disruption parameter survives.

The four templates together encode the pre-committed decision rule of Chapter 6: for the alternative hypothesis to survive, Table D.1 must show a positive, non-trivial, statistically distinguishable disruption parameter that is robust across Table D.4, and Table D.3 must show a positive avoided-cost parameter; for the null to be retained as more than an underpowered non-finding, Table D.1 must show a near-zero estimate with a confidence interval tight enough to exclude economically meaningful effects, which is exactly what the minimum-detectable-effect column is there to certify. The tables are empty by design, and their emptiness is the honest signature of a design-stage dissertation that has built the apparatus in full and in the open without claiming the numbers it is built to produce.