# Deep Space Network as a Queue: Contention, Wait-Time, and the Science-Throughput Penalty of Antenna Scheduling **Candidate:** JPL_INSTRUMENTS_NAV_05 **Program:** COLLEGIUM 1st Battalion **JPL / NORTH STAR category:** Navigation and Guidance **Date:** 2026-06-15 --- ## Abstract The Deep Space Network (DSN) is the shared ground asset that provides tracking, telemetry, and command for nearly every NASA deep-space mission and for many partner missions. Its three complexes at Goldstone, Madrid, and Canberra hold a fixed number of antennas, while the population of supported missions and their aggregate data-return requirements continue to grow. Demand for tracking passes therefore exceeds supply during recurring windows, and the schedule that allocates antenna time becomes a contested resource. This dissertation treats the DSN as a queueing system in which mission requests wait for allocation and some requested passes are never served. The central question is whether the science-throughput penalty caused by this contention is uniformly distributed across requests, as a simple congestion picture would imply, or whether it is heavy-tailed and concentrated on identifiable orbital-geometry and mission-phase windows. The falsifiable contribution is stated as a pair of hypotheses. The null holds that scheduling friction is roughly uniform and that a small share of conflicts does not account for a disproportionate share of lost downlink. The alternative holds that the wait-time and pass-loss distributions are heavy-tailed and that a minority of geometry and phase windows drive a majority of the lost science return. The method combines queueing-theory characterization of request-to-allocation wait times, survival analysis of the time from request submission to either allocation or expiry, heavy-tail distribution fitting, and regression of pass-loss events on orbital geometry, frequency band, and concurrent-mission load. The named data are the DSN scheduling and tracking logs available through the DSN service catalog and Service Preparation Subsystem schedule archives, NTRS DSN loading and forecasting reports, and mission downlink-requirement records. The work is presented as a design-stage analysis plan. No empirical results are claimed as executed; expected and illustrative magnitudes are labeled as such throughout. The intended payoff for NASA and JPL is a defensible, data-grounded basis for where added aperture, schedule policy changes, or demand shaping would recover the most science return per dollar. --- ## 1. Introduction and Contribution ### 1.1 The problem The Deep Space Network is the single most heavily shared navigation and communications asset in NASA's portfolio. Every mission beyond near-Earth orbit depends on it for radiometric tracking, for telemetry downlink, and for command uplink. The physical supply is fixed over any planning horizon of interest: a bounded set of large-aperture antennas distributed across three longitudes so that a deep-space target is in view of at least one complex at all times. Demand, by contrast, is not fixed. The number of active deep-space missions has grown, the per-mission data-return requirement has grown as instruments produce more data, and crewed exploration plans add high-rate, time-critical traffic. The result is structural oversubscription: in recurring windows the sum of requested antenna time exceeds the available antenna time, and the schedule must decide which requests are served, which are reduced, and which are dropped (Johnston et al., 2014; Hackett et al., 2018; Claudet et al., 2022). When a request is not served, or is served late, or is shortened, the mission loses downlink. Lost downlink is lost science return and, for navigation-critical phases, lost tracking that can degrade orbit-determination accuracy. The aggregate of these losses across the portfolio is a science-throughput penalty attributable to contention rather than to any hardware failure. The penalty is real and is acknowledged in the operational literature, which has motivated two decades of work on automated and optimization-based DSN scheduling (Johnston and Clement, 2008; Johnston et al., 2012; Claudet et al., 2022; Goh et al., 2021). That literature has concentrated on producing better schedules. It has paid less attention to characterizing the statistical shape of the penalty that scheduling friction imposes, and in particular to whether that penalty is spread evenly or concentrated. ### 1.2 The gap The DSN scheduling literature is strong on mechanism and on optimization. It frames the allocation problem as constraint satisfaction, mixed-integer programming, multi-objective optimization under uncertainty, or reinforcement learning, and it reports schedule-quality metrics such as the number of conflicts resolved or the fraction of requests satisfied (Johnston et al., 2014; Guillaume et al., 2021; Claudet et al., 2022; Goh et al., 2021). The capacity and traffic-forecasting literature is strong on aggregate loading and on the projection of future demand (Abraham et al., 2016; the heliophysics observatory expansion analysis of Malphrus and collaborators, 2023). What is missing is a statistical treatment that asks a distributional question: given the realized stream of requests and allocations, how is wait-time distributed, how is pass-loss distributed, and do a small number of structural conditions explain a large share of the loss. The aggregate satisfaction rate, even when reported, is a mean. A mean conceals whether the unserved requests are scattered thinly across all missions and all times or are bunched onto specific orbital-geometry and mission-phase windows. The policy implications of the two cases differ sharply. If the loss is uniform, the remedy is uniform: more aperture, applied broadly. If the loss is concentrated, the remedy is targeted: aperture, policy, or demand-shaping aimed at the specific windows that carry the load. ### 1.3 The single falsifiable contribution This dissertation contributes a falsifiable claim about the shape of the science-throughput penalty. - **H0 (null):** DSN scheduling friction is roughly uniform. Request-to-allocation wait times and pass-loss events are adequately described by a light-tailed (for example exponential or near-exponential) distribution, and no small subset of orbital-geometry, band, or mission-phase windows accounts for a disproportionate share of lost downlink. Under H0, a Gini-type concentration measure of lost downlink across windows is low, and the share of total loss carried by the top decile of windows is close to the ten percent that uniformity would predict. - **H1 (alternative):** DSN scheduling friction is heavy-tailed and concentrated. The wait-time distribution and the pass-loss distribution exhibit heavy tails that are better described by a power-law, log-normal, or generalized-Pareto upper tail than by an exponential, and a minority of identifiable geometry and phase windows (for example conjunction-driven viewing overlaps, critical mission events, and high-load concurrency intervals) account for a majority of the lost downlink. The contribution is falsifiable because each clause has a pre-registered test with a pre-registered decision rule, described in Section 5. Heavy-tail claims are tested by likelihood-ratio comparison of candidate distributions against the exponential and by goodness-of-fit on the fitted tail (Clauset et al., 2009; Alstott et al., 2014; Gillespie, 2015). Concentration claims are tested by the share of loss carried by the top decile of windows and by the slope of the loss-versus-window-rank curve. If the wait-time tail is statistically indistinguishable from exponential and the top-decile loss share is near ten percent, H0 survives and the contribution is refuted. ### 1.4 Why it matters for NASA and JPL NASA invests in DSN aperture, in scheduling software, and in mission-side data-management policy. Each of these is a lever on the science-throughput penalty, and each has a cost. If the penalty is concentrated, then a small, targeted intervention can recover a large share of lost science, which is the highest-return use of constrained aperture and operations budget. If the penalty is uniform, targeted interventions buy little and only broad capacity expansion helps. The distinction is decision-relevant at the level of NORTH STAR navigation and guidance planning, because tracking passes lost in specific mission phases also degrade orbit-determination products. A defensible characterization of the penalty's shape, grounded in the actual scheduling logs rather than in aggregate satisfaction rates, is therefore directly useful to capacity planning, to scheduling-policy design, and to the negotiation of mission downlink requirements. --- ## 2. Background and Literature ### 2.1 The DSN scheduling problem in the literature The DSN allocates antenna time through a multi-stage process that moves from long-range forecasting, through mid-range negotiation among missions, to near-real-time conflict resolution and execution. The canonical description of the automated scheduling environment is given by Johnston and colleagues, who describe the deployed DSN scheduling engine and the integrated planning workflow that spans forecasting to real-time operations (Johnston et al., 2014; Hackett et al., 2018). Earlier work framed the allocation as multi-objective optimization under uncertainty and applied evolutionary computation to it (Johnston and Clement, 2008; Guillaume et al., 2007). More recent work formulates the allocation as mixed-integer linear programming, including the Delta-MILP formulation that targets schedule changes rather than full re-solves (Claudet et al., 2022), and as a sequential decision problem addressed with deep reinforcement learning (Goh et al., 2021). The common thread is that all of these methods take demand as given and seek a better allocation of fixed supply. They optimize the schedule; they do not characterize the statistical penalty that remains after optimization. A parallel strand addresses capacity and traffic. Abraham and colleagues model DSN traffic for the human-exploration era and project the loading that crewed missions will add (Abraham et al., 2016). Analyses of network expansion to support specific science portfolios, such as the heliophysics system observatory, quantify the gap between projected demand and current supply (Malphrus et al., 2023). The radio-science and planetary-radar community has separately documented how DSN time competes across mission classes in the coming decade (the next-decade radio-science assessment of 2025). These studies establish that oversubscription is real and growing. They report it as an aggregate shortfall. They do not decompose it into a distribution over requests or over structural windows. ### 2.2 The queueing lens Treating a contested allocation system as a queue is standard operations-research practice. The DSN request stream is an arrival process; the antennas are servers; a request that cannot be served immediately waits, and a request whose viewing window passes before service is lost, which is a form of customer abandonment or deadline expiry. Kendall's notation and the embedded-Markov-chain analysis of single-server queues with general service distributions provide the formal vocabulary (Kendall, 1953; the embedded-Markov-chain bulk-queue analysis of 1964), and Kleinrock's two-volume treatment supplies the standard results for waiting-time distributions and their tails (Kleinrock, 1976). The relevant feature for this work is that queueing systems do not generally produce light-tailed waiting times. Under heavy-tailed service requirements, under bursty arrivals, or under priority disciplines that systematically defer a subclass of customers, waiting times become heavy-tailed even when the mean load is moderate. A line of work integrates queueing theory directly with scheduling to study dynamic scheduling problems of exactly this kind, where queue-discipline choices shape the realized wait distribution (Terekhov et al., 2014). The DSN schedule is a priority-and-preference discipline negotiated among missions, which is precisely the structure that can convert moderate average load into a heavy upper tail of waits for the deferred subclass. Three features of the DSN map onto known heavy-tail mechanisms in queueing systems, and naming them makes the H1 prediction concrete rather than speculative. First, the DSN viewing geometry imposes a hard deadline on each request: a target is visible from a given complex only for a bounded window, so a request that is not served before its window closes is lost rather than merely delayed. Deadline-driven queues with abandonment are known to produce a wait distribution with a defected mass and a heavy conditional tail among the survivors, because the requests that do get served late are systematically the ones the discipline could afford to defer. Second, the arrival process is bursty rather than smooth: requests cluster on the geometry windows when many targets are simultaneously well-placed, so the instantaneous load far exceeds the average load during those intervals. Burstiness in the arrival process is one of the classical routes by which an otherwise moderately loaded queue develops heavy waiting-time tails (Kleinrock, 1976). Third, the service requirement is itself variable: a high-rate critical-event pass occupies an antenna far longer and more rigidly than a routine cruise pass, so the service-time distribution is skewed, and skewed service is the third classical route to heavy waits. The point is not that any single mechanism must dominate but that all three are present in the DSN by construction, so the burden of proof should sit on the light-tailed null, not on the heavy-tailed alternative. This is the queueing-theoretic reason to test for, rather than assume away, the fat tail. ### 2.3 Forrester's lens: stocks, flows, feedback, and delay Jay Forrester's system dynamics provides the structural reading of why the DSN penalty might concentrate rather than spread (Forrester, 1961, as developed in the system-dynamics literature). In Forrester's terms the state of the DSN allocation system lives in stocks: the backlog of unserved or partially served requests, and the accumulated unmet downlink requirement per mission. These stocks change through flows: the arrival rate of new requests and the service rate at which passes are allocated and executed. The two are coupled through information feedback with delay. When backlog rises, missions respond by submitting requests earlier, by inflating requested durations to hedge against being cut, and by escalating priority for critical phases. Each of these responses is a balancing action from the mission's local point of view, but in aggregate they are a reinforcing pressure on contention during the windows that everyone wants. Forrester's central claim is that system behavior is dominated by loop structure and delay, not by the intentions of the actors, and that this produces counterintuitive concentration. Applied here, the lens predicts that contention will not smear evenly across the calendar; it will pile up on the geometry and phase windows where many missions' local optimizations collide, because the feedback that would relieve those windows operates with a delay longer than the window itself. This is a structural argument for H1 and against the uniform null. Forrester's related result that the temporal shape of an inflow, not merely its total, determines whether a stock stabilizes or runs away reinforces the point: it is the distribution of request timing, not the average load, that should drive the penalty (the launch-rate-distribution dynamical analysis of D'Ambrosio et al., 2022, applies the same inflow-shape logic in a space-systems setting). ### 2.4 Taleb's lens: fat tails, Extremistan, and fragility Nassim Taleb's program supplies the statistical discipline and the risk reading. Taleb's distinction between Mediocristan, where the mean and variance of a process are stable and a single observation cannot dominate the aggregate, and Extremistan, where a few extreme realizations dominate the mean and the historical record undersamples the tail, is the exact distinction at issue in H0 versus H1 (Taleb, drawing on the heavy-tail and fat-tail literature). If DSN wait-time and pass-loss live in Mediocristan, the aggregate satisfaction rate is an adequate summary and uniform remedies follow. If they live in Extremistan, the mean is misleading, the sample undersamples the worst windows, and any planning that uses the average loss understates the true exposure during the critical windows. Taleb's methodological insistence is twofold and is adopted directly in the analysis plan. First, do not assume light tails by default; test for them, and prefer the heavier-tailed model unless the data reject it, because the cost of falsely assuming thin tails is asymmetric. Second, characterize the second-order response of the system to the dose of the stressor rather than forecasting the stressor: the relevant question is not when the next high-contention window will occur but how the lost-downlink response curves as concurrent load rises. A convex (accelerating) response is the signature of fragility and is itself evidence for concentration. The heavy-tail estimation tools used in this work, the Clauset-Shalizi-Newman likelihood framework and the powerlaw and poweRlaw packages, are the standard implementation of Taleb's first instruction (Clauset et al., 2009; Alstott et al., 2014; Gillespie, 2015), and generalized-Pareto peaks-over-threshold modeling supplies the tail estimator for the second (the generalized-Pareto-process and peaks-over-threshold literature). ### 2.5 Synthesis Forrester explains why concentration is structurally plausible: looped, delayed feedback among locally optimizing missions piles contention onto shared windows. Taleb explains how to test for it without assuming it away and why getting the tail right matters for risk-correct planning. The DSN scheduling literature supplies the mechanism and the data. The queueing literature supplies the formal model of waits and losses. The contribution sits in the gap none of these has filled: a distributional, falsifiable characterization of the science-throughput penalty. --- ## 3. Data ### 3.1 Named datasets The analysis draws on three named data sources, all of which are real artifacts of DSN operations and the NASA technical-reporting system. 1. **DSN scheduling and tracking logs.** These are the request, allocation, and execution records produced by the DSN scheduling system and exposed through the DSN service catalog and the Service Preparation Subsystem (SPS) schedule archives. Each record documents a requested tracking activity: the requesting mission, the requested antenna or antenna class, the requested time window, the frequency band, the activity type (telemetry, command, radiometric tracking), the negotiated allocation if any, and the executed pass if any. The scheduling workflow that produces these records is documented in the operational literature (Johnston et al., 2014; Hackett et al., 2018). 2. **NTRS DSN loading and forecasting reports.** The NASA Technical Reports Server hosts the DSN loading analyses, traffic-forecasting studies, and capacity assessments that report aggregate oversubscription by complex, by band, and by epoch (for example Abraham et al., 2016, on human-exploration-era traffic, and the network-expansion analyses). These reports provide the load context and the validated demand projections against which the log-derived load measures are checked. 3. **Mission downlink-requirement records.** Each supported mission carries a stated data-return requirement and a set of critical-event windows during which tracking is mission-critical. These records, maintained in mission service agreements and reflected in the request stream, convert a lost pass into a quantified lost-downlink magnitude and identify the mission-phase windows used as covariates. ### 3.2 Access path The DSN scheduling and tracking logs are accessed through the DSN service-catalog and SPS schedule-archive interfaces under the appropriate data-use arrangements; the NTRS reports are publicly retrievable from ntrs.nasa.gov; the mission downlink-requirement records are obtained from mission service agreements. Where direct log access is constrained, the published loading reports provide aggregate cross-checks sufficient to calibrate the load covariates. ### 3.3 Unit of analysis The primary unit of analysis is the **tracking-pass request**: one requested activity for one mission on one antenna class over one viewing window. Each request is the subject of the survival model (time from submission to allocation or expiry) and the unit over which pass-loss is defined. A secondary unit is the **geometry-phase window**: a discretized cell defined by orbital-geometry condition, mission phase, complex, and band, over which lost-downlink is aggregated to test concentration. ### 3.4 Variable construction - **Wait time:** elapsed time from request submission to the moment of confirmed allocation. Right-censored for requests still pending at the observation cutoff. - **Pass-loss indicator:** binary, set to one when a requested pass is not executed (dropped, expired, or reduced below a science-useful threshold) and zero when executed as requested. - **Lost-downlink magnitude:** for a lost or reduced pass, the shortfall between the requested data return and the executed data return, computed from band, requested duration, and the mission's nominal data rate. This is the science-throughput penalty in physical units. - **Orbital-geometry covariates:** target declination and viewing-window overlap among concurrently supported targets, elevation profile at each complex, and conjunction or occultation conditions that compress the usable window. - **Band:** the requested frequency band (for example S, X, Ka), which governs antenna eligibility and contention with other band users. - **Concurrent-mission load:** the count and aggregate requested antenna-time of other missions contending for the same complex and band in the same interval, the realized load covariate that operationalizes Forrester's inflow-shape variable. - **Mission-phase window:** categorical indicator of critical phase (for example launch, maneuver, encounter, entry-descent-landing) versus cruise. ### 3.5 Coverage The intended coverage is a multi-year span of the request and tracking logs sufficient to include several cycles of the recurring high-contention windows and a representative set of mission phases across mission classes. Multi-year coverage is required so that the tail of the wait-time and loss distributions is observed often enough to fit, consistent with Taleb's warning that short records undersample the tail. ### 3.6 Limitations Four limitations are recognized at the design stage. First, request records reflect strategic mission behavior: missions inflate requested durations and submit early to hedge against cuts, so requested time overstates true need and must be interpreted as a hedged quantity, not a clean demand signal. Second, reduced (not fully dropped) passes require a science-usefulness threshold to classify, and that threshold is a modeling choice that must be varied in sensitivity analysis. Third, lost-downlink magnitude depends on assumed nominal data rates that vary with link conditions, introducing measurement error in the dependent variable. Fourth, access constraints on the raw logs may force reliance on aggregated loading reports for some intervals, which coarsens the load covariate. None of these defeats the design, but each bounds the strength of the eventual claim and is carried into the threats-to-validity discussion. --- ## 4. Research Design and Identification ### 4.1 Estimands The design targets three estimands. First, the **shape of the wait-time distribution**, in particular whether its upper tail is heavy. Second, the **shape and concentration of the lost-downlink distribution** across geometry-phase windows. Third, the **conditional association between pass-loss and its drivers**, namely orbital geometry, band, and concurrent-mission load, holding mission and epoch fixed. ### 4.2 Estimators Three estimators are used, matched to the three estimands. 1. **Survival model of request-to-allocation time.** Wait time is right-censored (pending requests, requests whose window expires) and is therefore time-to-event data. The Kaplan-Meier estimator gives the nonparametric survival function of the wait (Kaplan and Meier, 1958), and a Cox proportional-hazards model gives the covariate-adjusted hazard of allocation as a function of geometry, band, and load (Cox, 1972; Therneau and Grambsch, 2000). Time-varying covariates are admitted because concurrent load changes while a request waits (the time-dependent-covariate extension of the Cox model). The hazard of allocation is the inverse-intuition complement of wait: covariates that lower the allocation hazard lengthen the wait. 2. **Heavy-tail distribution fitting.** The empirical wait-time and lost-downlink distributions are fit to candidate families: exponential (the light-tailed null benchmark), log-normal, power-law, and generalized Pareto for the upper tail. Fitting and model comparison follow the Clauset-Shalizi-Newman maximum-likelihood-plus-goodness-of-fit procedure, implemented in the powerlaw and poweRlaw packages, with likelihood-ratio tests choosing among non-nested candidates (Clauset et al., 2009; Alstott et al., 2014; Gillespie, 2015). The upper tail is additionally estimated by peaks-over-threshold with a generalized-Pareto fit, the standard extreme-value estimator for tail behavior above a high threshold. 3. **Pass-loss regression.** The binary pass-loss indicator is regressed on geometry, band, and concurrent load using a logistic specification with mission and epoch fixed effects; the continuous lost-downlink magnitude is modeled with a generalized linear model appropriate to its skewed, non-negative support. Concentration is summarized by a Gini coefficient of lost-downlink across windows and by the share of total loss carried by the top decile of windows. ### 4.3 Identification strategy The identifying variation is the within-mission, within-epoch difference in pass-loss and wait as concurrent load and geometry vary. Mission fixed effects absorb time-invariant mission characteristics (instrument data rate, priority class, strategic-padding behavior) so that the load and geometry coefficients are not confounded by which missions tend to be busy. Epoch fixed effects absorb network-wide shocks (an antenna in maintenance, a network-wide demand surge) common to all missions in a period. The remaining variation used for identification is that a given mission, in a given period, faces different contention on different requests because of the geometry of its target and the coincidental load from others, which is plausibly exogenous to that mission's own value of the specific pass. The concentration test does not require this conditional-exogeneity assumption: it is a description of the realized distribution and stands on the completeness of the log, not on an identification claim. ### 4.4 Model specification For pass-loss probability, the index specification is a logistic model in which the log-odds of loss for request i of mission m in epoch t is a linear function of geometry covariates, band indicators, concurrent-load measures, mission-phase indicators, and mission and epoch fixed effects. For the survival model, the Cox partial likelihood is specified with the same covariate vector entered as (where appropriate) time-varying. For the tail, the generalized-Pareto scale and shape parameters are estimated above a threshold chosen by standard mean-residual-life and parameter-stability diagnostics; a positive shape parameter is the formal signature of a heavy (Pareto-type) tail. ### 4.5 Threats to validity - **Internal validity.** The principal threat is that requested time is endogenous to expected contention: missions that anticipate a hard window pad and escalate, so load and loss are jointly determined by an unobserved expectation. Mission and epoch fixed effects absorb the systematic component of this; an instrument or a placebo using geometry-driven (not behavior-driven) window compression is used as a robustness check, because the geometry of a target's viewing window is set by celestial mechanics and is not chosen by the mission. - **External validity.** The penalty's shape may be specific to the studied epoch's mission mix. The crewed-exploration era will change the mix toward high-rate, time-critical traffic (Abraham et al., 2016), which the design cannot observe in advance. Claims are therefore conditioned on the studied mix, and the load-response curve, rather than a single point estimate, is reported so that extrapolation is explicit. - **Construct validity.** Lost downlink is a proxy for lost science return; not all bits carry equal scientific value, and a navigation-critical tracking pass that is lost has a value not captured by a data-volume measure. The construct is therefore reported both in data-volume units and, where mission records permit, weighted by mission-declared criticality. - **Statistical-conclusion validity.** Heavy-tail fitting is prone to false positives when sample sizes are small or thresholds are chosen post hoc. The Clauset framework's goodness-of-fit step, pre-registered threshold-selection diagnostics, and likelihood-ratio comparison against the exponential are used precisely to control this, and the decision rule is fixed before estimation (Clauset et al., 2009; Gillespie, 2015). --- ## 5. Analysis Plan and Findings **This section is a design-stage analysis plan. The numbers below are illustrative placeholders that describe the form the results would take. They are not empirical findings and have not been computed on the DSN logs. They are marked as illustrative wherever they appear.** ### 5.1 Estimation procedure The estimation proceeds in five pre-registered steps. 1. **Assemble the request-level panel.** Join the DSN scheduling and tracking logs to mission downlink-requirement records and to derived geometry covariates, producing one row per tracking-pass request with wait time, censoring indicator, pass-loss indicator, lost-downlink magnitude, and the covariate vector. 2. **Estimate the wait-time distribution.** Compute the Kaplan-Meier survival curve of request-to-allocation time overall and stratified by load quartile, band, and mission phase. Fit exponential, log-normal, power-law, and generalized-Pareto models to the wait distribution and to its upper tail. Record the likelihood-ratio test of each heavier-tailed candidate against the exponential and the goodness-of-fit p-value of the chosen model. 3. **Estimate the loss concentration.** Aggregate lost-downlink to geometry-phase windows. Compute the Gini coefficient across windows and the share of total loss carried by the top decile. Plot lost-downlink against window rank. 4. **Estimate the conditional drivers.** Fit the Cox model for allocation hazard and the logistic and GLM models for pass-loss and lost-downlink magnitude, with mission and epoch fixed effects. Report covariate associations for geometry, band, and concurrent load. 5. **Estimate the load-response curve.** Plot expected lost-downlink against concurrent-mission load to test for convexity, the Taleb fragility signature. ### 5.2 Decision rules - H1 wait-time clause is accepted if the likelihood-ratio test prefers a heavy-tailed model over the exponential at the pre-set significance level and the generalized-Pareto shape parameter is significantly positive; otherwise H0 is retained for waits. - H1 concentration clause is accepted if the top-decile loss share is materially above the ten percent uniformity benchmark (a pre-registered threshold, for example a top-decile share at or above forty percent) and the Gini coefficient is correspondingly high; otherwise H0 is retained for concentration. - The contribution as a whole is accepted only if both clauses are accepted; either clause failing refutes it. ### 5.3 Expected findings (illustrative, not computed) Under the structural reasoning of Sections 2.3 and 2.4, the design anticipates the following pattern. These figures are **illustrative placeholders** chosen to show the form of a positive result; they are not estimates. - The wait-time survival curve would decline slowly in its tail, with a fitted generalized-Pareto shape parameter greater than zero (illustrative value: a shape near 0.3 to 0.5), and the likelihood-ratio test would prefer a power-law or log-normal upper tail over the exponential. - Lost-downlink would concentrate: an illustrative top-decile window share on the order of half to two-thirds of total lost downlink, against the ten percent that uniformity predicts, with a correspondingly elevated Gini coefficient. - The pass-loss regression would show higher loss probability in conjunction-driven viewing-overlap windows, in the most contested band, and at the highest concurrent-load quartile, with mission-phase critical windows carrying elevated loss conditional on load. - The load-response curve would be convex: lost-downlink would rise faster than linearly as concurrent load increases, the fragility signature. If instead the wait tail were statistically indistinguishable from exponential, the top-decile loss share were near ten percent, and the load-response curve were linear, the null would survive and the contribution would be refuted. Both outcomes are reportable; the design does not presuppose the alternative. ### 5.4 What a robust positive result would and would not establish A positive result would establish that the penalty is concentrated and heavy-tailed and would localize the carrying windows. It would not by itself establish causation of the concentration, which the Forrester feedback account supplies as theory but which the cross-sectional design tests only through the convex load-response signature. That limitation is stated plainly and is the boundary of the claim. --- ## 6. Discussion ### 6.1 Implications If the heavy-tail-and-concentration result holds, the operational implication is that the highest-return interventions are targeted at the carrying windows rather than spread across the network. Three levers follow. First, capacity: added aperture, or array combination of existing antennas, deployed to relieve the specific complexes, bands, and epochs that the analysis localizes. Second, policy: scheduling-discipline changes that protect the deferred subclass, since the heavy wait tail is partly a product of a priority discipline that systematically defers some requests (Terekhov et al., 2014). Third, demand shaping: negotiating mission downlink requirements and request timing so that locally optimal padding and early submission, the Forrester feedback, do not pile onto the shared windows. Each lever is testable by re-running the design on post-intervention logs. ### 6.2 Rival explanations Three rival explanations must be addressed. First, the concentration could be an artifact of a few atypical missions rather than a structural property; mission fixed effects and a leave-one-mission-out robustness check address this. Second, the heavy tail could be an artifact of threshold choice in the tail fit; the pre-registered diagnostics and goodness-of-fit step address this (Clauset et al., 2009). Third, the apparent concentration on geometry windows could reflect that those windows are simply when the most valuable science is requested, so the loss is high-value by selection rather than by contention; the criticality-weighted construct and the within-mission identification address this by separating value from contention. ### 6.3 External validity and the exploration era The crewed-exploration era will shift the mission mix toward high-rate, time-critical traffic and is projected to raise loading substantially (Abraham et al., 2016; Malphrus et al., 2023). The reported load-response curve, rather than a single satisfaction rate, is the externally valid object: it predicts how the penalty would scale as load rises, which is the quantity a planner needs for the era the data do not yet contain. A convex curve implies that the exploration-era penalty would be worse than a linear extrapolation of today's average loss, which is itself a planning-relevant finding. ### 6.4 What would falsify the contribution The contribution is falsified if any of the following holds on the full data: the wait-time tail is statistically indistinguishable from exponential; the top-decile window share of lost downlink is near the uniformity benchmark; or the load-response curve is linear or concave. Each is a concrete, pre-registered outcome. The design is built so that the null is a live possibility and a clean refutation is achievable, which is the property that makes the contribution scientific rather than rhetorical. --- ## 7. Contribution and Conclusion This dissertation reframes a known operational fact, that the Deep Space Network is oversubscribed, as a measurable distributional question, and states a single falsifiable answer. The contribution is the claim that the science-throughput penalty from DSN antenna contention is heavy-tailed and concentrated on identifiable orbital-geometry and mission-phase windows, against the null that it is uniform. The claim is operationalized with a queueing and survival-analysis design over the DSN scheduling and tracking logs, NTRS loading reports, and mission downlink-requirement records, with pre-registered tests and decision rules. The methodological frame is Forrester's: backlog and unmet-requirement stocks, request and service flows, and delayed feedback among locally optimizing missions, which makes concentration structurally plausible. The statistical discipline is Taleb's: test for fat tails rather than assume them away, and characterize the convex load-response rather than forecast the next bad window. The work is presented honestly as a design-stage plan; no empirical results are claimed. Its value to NASA and JPL is to convert a known shortfall into a localized, decision-relevant target, so that constrained aperture, scheduling policy, and demand negotiation can be aimed where they recover the most science return, and so that planning for the crewed-exploration era uses the right object, the load-response curve, rather than a misleading average. If the data refute the contribution, that too is useful: it would justify broad rather than targeted capacity expansion. Either way, the penalty's shape, not merely its average, becomes a measured quantity. --- ## References 1. Cox, D. R. (1972). Regression Models and Life-Tables. *Journal of the Royal Statistical Society: Series B*, 34(2), 187-220. https://doi.org/10.1111/j.2517-6161.1972.tb00899.x 2. Kaplan, E. L., and Meier, P. (1958). Nonparametric Estimation from Incomplete Observations. *Journal of the American Statistical Association*, 53(282), 457-481. https://doi.org/10.1080/01621459.1958.10501452 3. Therneau, T. M., and Grambsch, P. M. (2000). *Modeling Survival Data: Extending the Cox Model*. Springer. https://doi.org/10.1007/978-1-4757-3294-8 4. Clauset, A., Shalizi, C. R., and Newman, M. E. J. (2009). Power-Law Distributions in Empirical Data. *SIAM Review*, 51(4), 661-703. https://doi.org/10.1184/r1/6586835.v1 5. Alstott, J., Bullmore, E., and Plenz, D. (2014). powerlaw: A Python Package for Analysis of Heavy-Tailed Distributions. *PLOS ONE*, 9(1), e85777. https://doi.org/10.1371/journal.pone.0085777 6. Gillespie, C. S. (2015). Fitting Heavy Tailed Distributions: The poweRlaw Package. *Journal of Statistical Software*, 64(2). https://doi.org/10.18637/jss.v064.i02 7. Kleinrock, L. (1976). *Queueing Systems, Volume 1: Theory* (reviewed). *SIAM Review*, 18(4). https://doi.org/10.1137/1018095 8. Kendall, D. G. (1953/1955). On the Theory of Queues with Many Servers. *Transactions of the American Mathematical Society*. https://doi.org/10.2307/1992945 9. Terekhov, D., Tran, T. T., Down, D. G., and Beck, J. C. (2014). Integrating Queueing Theory and Scheduling for Dynamic Scheduling Problems. *Journal of Artificial Intelligence Research*, 50, 535-572. https://doi.org/10.1613/jair.4278 10. Johnston, M. D., Tran, D., Arroyo, B., Sorensen, S., Tay, P., Carruth, J., Coffman, A., and Wallace, M. (2014). Automated Scheduling for NASA's Deep Space Network. *AI Magazine*, 35(4), 7-25. https://doi.org/10.1609/aimag.v35i4.2552 11. Claudet, T., Alimo, R., Goh, E., and Johnston, M. D. (2022). Delta-MILP: Deep Space Network Scheduling via Mixed-Integer Linear Programming. *IEEE Access*, 10, 41330-41340. https://doi.org/10.1109/access.2022.3164213 12. Johnston, M. D., and Clement, B. J. (2008). Deep Space Network Scheduling Using Multi-Objective Optimization with Uncertainty. *SpaceOps 2008 Conference*. https://doi.org/10.2514/6.2008-3580 13. Johnston, M. D., Tran, D., Arroyo, B., and Page, C. (2012). Automating Mid- and Long-Range Scheduling for NASA's Deep Space Network. *SpaceOps 2012 Conference*. https://doi.org/10.2514/6.2012-1296235 14. Guillaume, A., Lee, S., Wang, Y.-F., Zheng, H., Hovden, R., Chau, S., Tung, Y.-W., and Terrile, R. J. (2007). Deep Space Network Scheduling Using Evolutionary Computational Methods. *2007 IEEE Aerospace Conference*. https://doi.org/10.1109/aero.2007.352900 15. Goh, E., Venkataram, H. S., Hoffmann, M., Johnston, M. D., and Wilson, B. (2021). Scheduling the NASA Deep Space Network with Deep Reinforcement Learning. *2021 IEEE Aerospace Conference*. https://doi.org/10.1109/aero50100.2021.9438519 16. Hackett, T. M., Bilen, S. G., Johnston, M. D., et al. (2018). Integrated Planning and Scheduling for NASA's Deep Space Network - from Forecasting to Real-time. *SpaceOps 2018 Conference*. https://doi.org/10.2514/6.2018-2728 17. Abraham, D. S., MacNeal, B. E., Heckman, D. P., et al. (2016). Traffic Modeling for Deep Space Network in the Human Exploration Era. *SpaceOps 2016 Conference*. https://doi.org/10.2514/6.2016-2523 18. Malphrus, B. K., et al. (2023). Expanding the Deep Space Network to Support the Heliophysics System Observatory. *Frontiers in Astronomy and Space Sciences*, 9, 1051527. https://doi.org/10.3389/fspas.2022.1051527 19. D'Ambrosio, A., et al. (2022). A Dynamical Systems Analysis of the Effects of the Launch Rate Distribution on the Stability of the Orbital Environment. *Journal of Space Safety Engineering*. https://www.semanticscholar.org/search?q=Dynamical%20Systems%20Analysis%20Launch%20Rate%20Distribution 20. Lewis, H. G. (2020). Understanding Long-Term Orbital Debris Population Dynamics. *Journal of Space Safety Engineering*, 7(3), 164-170. https://doi.org/10.1016/j.jsse.2020.06.006 21. Forrester, J. W. (1961). *Industrial Dynamics* (and the system-dynamics tradition). *System Dynamics Review*. https://doi.org/10.1002/sdr.284 22. Deep Space Network Radio Science and Ground-Based Planetary Radar in the Next Decade (2025). *Radio Science*. https://doi.org/10.1029/2025rs008296