# Interrogation mind-map: PHD-08

Nodes: 130 | questions: 48 | grounded claims: 42 | gaps: 40

## Questions

- **[measurement]** Does a single latency scalar built from outcome-records confound delays produced by three distinct decision games (Model I rational pause, Model II organizational-routine queue clearing, Model III bureaucratic side-payment), and does the headline latency coefficient survive hand-coding a stratified sample of authorization intervals by process type and re-estimating with process-type interaction terms? (raised by allison)
- **[identification]** Do the two leading instruments (authorizing-office workload, proximity to the appropriations calendar) violate the exclusion restriction because each is itself the operating variable of a specific game (workload = Model II organizational congestion; appropriations clock = Model III OMB/committee bargaining), acting on cost and schedule directly rather than only through scalar latency? Can the seat each instrument operates from be documented and a placebo/overid test (e.g., no-year-funded programs insulated from the appropriations cycle) distinguish a true instrument from a channel? (raised by allison)
- **[rival]** Is the latency-outcome association a Model III common-cause artifact, in which long latency and high cost growth are co-produced 'resultants' of the same coalition fight (pulling and hauling over a contested program) with neither causing the other, and does the latency coefficient survive an independent documentary proxy for bureaucratic-bargaining intensity per program-phase? (raised by allison)
- **[rival]** Program-and-era FE treats NASA as one transaction-cost machine, but an identical latency value can be produced by structurally different seats holding the clock (Mission Directorate HQ, field-center program office, OMB passback, appropriations subcommittee). Partition each authorization interval by which seat owns the pending decision and report whether the latency-to-cost-and-schedule association holds WITHIN a seat or is only a between-seat composition effect. (raised by allison)
- **[identification]** The surviving instrument, authorizing-office workload, is in bureaucratic-politics terms a contested stake, not exogenous traffic: queue order is a pull-and-haul resultant in which politically backed programs jump ahead. Build a measure of each program's standing (directorate champion, congressional-district interest, flagship designation) and test whether workload predicts cost/schedule DIFFERENTLY for portfolio winners vs losers; if the path runs only through deprioritized losers, the exclusion restriction fails. (raised by allison)
- **[governance]** Era FE absorb the reorganizations and procurement reforms that compress/lengthen reporting layers, but in Model III a reorganization is the live treatment: it re-seats players, abolishes a clearance, or moves an authorization. Isolate documented discrete RE-SEATING events (clearance layer eliminated, authority delegated downward, review office stood up/disbanded) and show in an event study that the SAME programs' latency and cost/schedule outcomes move together after re-seating; if they do not, latency is a passive marker of regime, not a lever. (raised by allison)
- **[identification]** Continuous-treatment framing does not dissolve the heterogeneity problem; it relocates it into selection-into-dose. Parallel trends identifies ATT(latency) at observed doses but does not license a cross-dose slope. State the stronger assumption that makes beta a marginal effect and show empirically whether ATT(latency) is flat across the dose support or driven by selection into high latency. (raised by callaway_santanna)
- **[identification]** No-anticipation is the load-bearing assumption and the candidate's own Section 4.5 baseline-gaming check concedes it is likely violated: if agencies set optimistic baselines BECAUSE they anticipate slow authorization, measured growth responds to expected latency before authorization resolves. Show the documentary distribution of the foreseeability-to-realization gap and report pre-period placebo ATT(g,t) on cost growth and schedule slip for cohorts entering each discrete regime. Non-zero pre-period placebos mean no-anticipation fails. (raised by callaway_santanna)
- **[empirics]** Under heterogeneous and dynamic effects a 2SLS LATE is a weights-driven average over compliers that can carry negative or non-convex weights, the same pathology TWFE suffers. Show the implied weights of the INSTRUMENTED estimand: is the 2SLS latency coefficient a convex-weighted average of program-phase causal effects or does it place negative weight on some program-time cells? Provide the Goodman-Bacon-style decomposition of the instrumented estimand, not the OLS TWFE one, and defend the exclusion restriction against dynamic portfolio-difficulty spillovers. (raised by callaway_santanna)
- **[empirics]** Produce the implied-weight (Goodman-Bacon / effective-sample) decomposition of the preferred FE-IV latency coefficient: each program's and each era's share of the identifying variation, ranked, and show whether a handful of long-cycle flagships (Apollo, Shuttle, Constellation, SLS, JWST) supply the majority of the weight, in which case the estimand must be restated as the effect for that dominant subset. (raised by callaway_santanna)
- **[rival]** Partition the panel by era regime (Apollo/Cold-War, Shuttle, post-Challenger, post-Constellation/commercial) and report ATT(latency) era-by-era rather than pooled; show whether the latency-cost association holds sign and magnitude within the modern regime a current manager operates in, or whether the pooled headline is carried by extinct early cohorts. (raised by callaway_santanna)
- **[identification]** State the target population for the policy claim ('compress authorization latency to improve execution') and demonstrate, from the panel's entry/exit/censoring structure, that the estimand's implicit weights match it: do cancelled/censored program-phases (Constellation, SLS upper stage, second mobile launcher) get weight proportional to managerial relevance, or does survivorship tilt the effective sample toward programs that completed despite their latency, breaking transportability to in-flight and future programs? (raised by callaway_santanna)
- **[identification]** Latency and cost/schedule trouble form a closed feedback loop; a single-equation FE+IV beta recovers an open-loop slice. Can the candidate's own panel show that latency Granger-precedes trouble (and not vice versa) at the within-phase level, with an asymmetric cross-lag? If trouble predicts next-period latency as strongly as the reverse, the beta is a loop coefficient mislabeled as one-way cause. (raised by forrester)
- **[governance]** Forrester's policy-resistance result: complex social systems counteract parameter interventions through delayed compensating loops. After documented latency-compressing regime switches in the panel (reorganizations cutting reporting layers; procurement reforms shortening authorization), does the post-regime cost-growth and schedule-slip trajectory bend as the static beta predicts, or does latency reappear elsewhere (longer informal pre-decision staffing, more rework, reasserted reviews) so total elapsed time does not fall? (raised by forrester)
- **[measurement]** Latency is a per-phase elapsed interval treated as the unit, but cost growth and schedule slip are accumulations integrated over whole program life, and the delay from an authorization action to its effect on a stock can exceed one phase. Can the panel show the latency-to-outcome delay is shorter than the phase window, i.e. that a phase's latency consequences land within that phase rather than being carried into later phases of the same program? If the dominant delay exceeds the phase, the phase-level beta attributes to phase p an accumulation phase p only seeded (a stock-flow / bathtub timing error). (raised by forrester)
- **[mechanism]** A level-on-level FE regression of an acquisition outcome on latency assumes an open one-pass causal path, but the cited practitioner mechanism is a loop (delay -> requirements drift -> re-reviews -> added authorization events -> longer latency). On the modern KDP-dated subperiod, build a within-program count of authorization events per phase and test whether it rises endogenously as a phase's cost-growth accumulates, then report whether beta is stable once you condition on the loop's own state. (raised by forrester)
- **[measurement]** Cost growth and schedule slip are stock accumulations (integrals of rates over a phase), yet the estimator treats each program-phase as a static snapshot. On programs with dated interim cost/schedule re-baselines within a phase, reconstruct the accumulation trajectory and test whether latency predicts the RATE of cost accrual at the moment it occurs versus the END-of-phase stock, demonstrating the phase-end beta is not a mechanical integration artifact (long phases score higher cost growth merely because the bathtub had more time to fill). (raised by forrester)
- **[empirics]** The headline coefficient is a within-program-within-era (actor-level) estimate, but the agency-level behavior of interest (cadence collapse, chronic overrun) is a system property of the shared authorization queue, where one office's latency raises the workload stock that lengthens every other program's latency. Aggregate to the authorizing-office-by-year level, regress total office latency on the contemporaneous count of programs in the queue to test for queue-congestion dynamics, then show the program-level beta summed across the panel reproduces the observed agency-level cadence series rather than committing a fallacy of composition. (raised by forrester)
- **[measurement]** On the overlap window where both the coarse milestone-to-milestone latency (1958-2026) and the fine key-decision-point latency (modern subperiod) can be built from the same programs, report their correlation and the systematic level gap (coarse minus fine, in months), and show whether that gap is constant or drifts over time. A drifting gap means the coarse series is not measuring the same latency in 1962 as in 2024. (raised by maddison)
- **[empirics]** Re-estimate the latency-cost association under at least two independent deflators (NASA New Start Inflation Index vs an economy-wide GDP deflator) and report how much beta moves; if the headline coefficient depends on the chosen inflation series, the deflator was measured, not the latency effect. (raised by maddison)
- **[measurement]** For programs that received a re-baseline, show the distribution of latency and cost-growth computed against the original versus the re-set baseline, and state which baseline the panel uses by era; if early-era baselines are informal practitioner targets and late-era ones are GAO/JCL-anchored commitments, numerator and denominator are jointly mis-scaled and the within-era fixed effect cannot net out a regressor-correlated distortion. (raised by maddison)
- **[identification]** Publish the exact era-boundary rule. Prove the periodization is derived from a documented break in a measured series (cost-growth, cadence, or appropriations), not from a chosen reform calendar co-determined with the authorization-regime treatment, and show the latency coefficient under at least two materially different defensible periodizations so the result does not depend on where the epochs are drawn. (raised by maddison)
- **[mechanism]** Decompose phase cost growth into a time-rate standing-cost component (priced from documented monthly burn during idle authorization intervals) and a residual, and show that latency is associated with the residual (genuine productivity loss: rework, requirements churn, re-baselining) and not merely with the standing-cost term that elapsed schedule would predict by construction. (raised by maddison)
- **[measurement]** Produce per-decade counts of program-phase observations and the share of latency values that are point-identified versus bounded, then re-estimate the latency association on the pre-1980 segment alone, so the panel can see whether the 'long-run series' claim survives outside the data-dense modern (GAO / Aerospace-Conference) window or is a modern-NASA finding in historical costume. (raised by maddison)
- **[measurement]** Latency runs from a documented trigger to a documented authorization, and you call the authorization the decision. Emergent strategy says realized choices form as patterns before any apex ratifies them. Reconstruct, from the contemporaneous documentary trail, the date substantive commitment crystallized versus the date the recorded authorization fired; show the divergence distribution (in months) before I believe the recorded authorization is the decision. (raised by mintzberg)
- **[mechanism]** Your construct fires only on events that left a dated formal record. Real executive decision-making is fragmented, interruption-driven, verbally mediated, exactly the activity that leaves no dated artifact. A hallway decision ratified slowly on paper and a genuinely agonized months-long review can produce identical panel entries. Using process-rich cases (oral histories, NASA institutional histories, Apollo management-control record), show that documentary latency correlates with an independent, process-derived measure of how long substantive deliberation actually took. (raised by mintzberg)
- **[identification]** Eras differ in how much of the real decision is made inside the formally-recorded review machinery versus outside it before ratification. If that ratio is era-dependent, era fixed effects do not save you: latency means a different thing in 1965 than in 2020. Produce, per era, an estimate of what fraction of substantive decision time your documentary trigger-to-authorization interval captures, and show that this capture fraction is stable enough that a within-era coefficient compares like with like. (raised by mintzberg)
- **[mechanism]** Pooled latency coefficient (Sec 4.2) presumes one slope across NASA, but Structure-in-Fives treats NASA as a hybrid of distinct configurations each coordinated by a different mechanism; in a machine-bureaucratic phase latency-as-review-time buys conformance, in an adhocratic phase the same latency destroys the mutual adjustment novel work needs. Stratify the panel by dominant coordinating configuration per program-phase (routine production/operations vs. novel first-of-kind development) and show whether the latency-cost slope is heterogeneous, or is the pooled beta averaging a productive and a destructive coefficient into a meaningless mean? (raised by mintzberg)
- **[measurement]** The latency clock has two datable endpoints, a documentary 'trigger event' (Sec 3.3) and an 'authorization event'. Managerial-work realism says real decision-making is fragmented, interruption-driven, and verbally mediated; the substantive due moment and resolution are negotiated in corridor conversation and enter the record only when ratified. On a hand-coded sample with oral histories or contemporaneous correspondence, demonstrate that the documentary timestamps are not the LATE ratifying bookends of a process that substantively opened and closed earlier, and that this ratification lag is uncorrelated with phase difficulty rather than a measure of how much administrative ceremony a hard phase generates. (raised by mintzberg)
- **[identification]** The design uses era fixed effects to absorb rule regimes and reorganizations, then identifies beta off within-program, within-era variation. But under the candidate's own path-dependence premise the rule regime IS the treatment: a reorganization compressing reporting layers is the natural experiment that reveals whether latency CAUSES cost growth or whether long latency and high cost are co-products of a single ill-fitting machine-bureaucratic configuration imposed on novel work. By soaking regime variation into delta_t, is the design absorbing the exact variation that distinguishes 'latency is the disease' from 'latency is a symptom of configuration mismatch'? On documented latency-compressing reorganizations, show post-regime cost-and-schedule improvement that exceeds the mechanical removal of the compressed waiting time, the only test ruling out the remedy being a downstream symptom. (raised by mintzberg)
- **[identification]** Within-program-within-era latency, the residual your coefficient is identified off, is precisely where the institutions-versus-organizations distinction says the institutional signal is NOT, since both the rule regime (era FE) and the standing organization (program FE) have been differenced out. Can you show from the assembled panel that residual latency tracks a documented rule-of-the-game change (a coded register of rule-text changes) with non-trivial R-squared, rather than idiosyncratic case-handling noise? If residual latency is orthogonal to documented rule changes, the coefficient cannot speak to the institutional mechanism claimed. (raised by north)
- **[measurement]** North's economic-history method exists to detect the gap between de jure rules and de facto rules-in-use. The latency measure is built from a single documentary rule over milestone and key-decision-point records, but the documented authorization date is a de jure artifact that can lag or lead the de facto moment a decision was effectively settled. Can you validate, on a subset of programs, that documentary latency correlates with an independent de facto measure of when the decision was actually made (internal correspondence timestamps, oral-history dates, working-group minutes)? Without it, you may be measuring the cadence of paperwork, not the transaction cost of deciding. (raised by north)
- **[rival]** The proposed instruments, the authorizing office's contemporaneous workload and appropriations-calendar timing, are themselves products of the institutional matrix rather than exogenous shocks to it: workload is set by the same rule regime under test (how many actions the rules route to that office), and the appropriations cycle is a formal institution with its own path-dependent effects on cost and schedule. Can you demonstrate the exclusion restriction empirically, that workload affects cost growth, schedule slip, and cadence ONLY through latency and not through a direct congestion or rationing channel, e.g. by showing zero partial association of workload with outcomes conditional on latency? If workload moves outcomes through any other path, the instrument is invalid and the claim collapses back to the difficulty-driven correlation. (raised by north)
- **[identification]** Anchor the NASA latency-to-cost/schedule coefficient against at least one out-of-NASA institutional regime (DoD milestone-decision-authority SAR programs, ESA, or commercial fixed-price firm-commitment contracts) and report whether sign and magnitude survive a change of the funding-commitment rules; absent a benchmark, you have measured one institutional matrix and called it a law. (raised by north)
- **[governance]** State the scope condition formally: is H1 (latency drives overrun) a portable causal law or a property of NASA's institutional matrix that would weaken or reverse under fixed-price firm-commitment rules where the time cost of authorization is borne by the contractor? Name the specific rule-of-the-game feature (cost-reimbursement risk allocation, annual re-authorization, multi-layer review) the mechanism requires, and predict the coefficient's direction when that feature is removed. (raised by north)
- **[empirics]** Partition the NASA panel by documented internal rule-regime (named reorganizations and procurement-reform statutes; the appropriations/review/reserve rules of 1965, 1990, 2020 differ) and show the latency coefficient is stable across regimes rather than a composite averaging over distinct transaction-cost structures; if beta differs by regime, report it per regime. (raised by north)
- **[identification]** Both instruments (authorizing-office workload, appropriations-calendar timing) load on an undrawn upstream node: the agency-wide fiscal-political regime (CRs, caps, shutdown risk), which has an arrow into office workload AND a plausible direct arrow into cost growth. Era FE only block this back-door if 'era' is coarser than the year-to-year fiscal variation the instrument exploits and finer than the regime shifts it must absorb. Specify the era partition granularity and demonstrate from the NASA appropriations record whether within-era residual variation in the instrument is statistically independent of the funding-instability index. If they co-move within era, the exclusion restriction fails through the fiscal-regime back-door and no first-stage F can save it. (raised by pearl)
- **[measurement]** Outcome 'schedule slip' and treatment 'authorization latency' are measured from the same documentary milestone intervals, and Section 5.3 concedes schedule slip will be 'the largest and most robust' coefficient 'because authorization latency is itself a component of elapsed schedule.' That is a definitional path, not a strong-mechanism prediction: an arrow runs Latency -> Schedule-slip by construction. Show, from the milestone taxonomy in the NASA program records, the exact set-theoretic decomposition of a phase's elapsed time into authorization intervals vs engineering-execution intervals, and prove the two are disjoint. If any authorization month is also in the schedule-slip numerator, the schedule-slip result is mechanically tautological and must be dropped. (raised by pearl)
- **[identification]** The reverse-causation rival ('programs already overrunning generate more authorization events and longer latency') is answered by 'using latency measured early in each phase.' This temporal-ordering defense works only if the early-phase latency clock starts before any signal of trouble. Foreseen trouble at phase entry (known-immature TRL, contested baseline) inflates both early latency and eventual cost growth via an anticipated-difficulty arrow that program FE cannot remove (it varies by phase). From GAO/NTRS records, exhibit for a sample of program-phases the dated sequence showing the latency-clock-start event precedes the first documented risk-elevation, review-board finding, or requirements-change request in that phase. If the risk signal systematically predates the latency-start event, the early-measurement defense collapses and the arrow runs difficulty -> latency. (raised by pearl)
- **[identification]** Draw the explicitly mediated DAG (latency -> mediator [standing-cost accrual / requirements churn / workforce idle] -> cost, with an unobserved program-difficulty confounder) and state whether the latency-to-cost effect is front-door identifiable THROUGH the mediator without the contested IV. If difficulty also arrows directly into the mediator, front-door fails and neither door is open: which is it, and what in the three named datasets would populate the mediator node? (raised by pearl)
- **[measurement]** State the error structure of the two-resolution latency scalar on the DAG. Is the coarse-era measurement error CLASSICAL (independent of true latency, attenuating beta) or NON-CLASSICAL and correlated with the era node and the outcome (sparse-documentation eras also being looser-baseline, larger-overrun eras)? If non-classical, era FE conditions on a variable downstream of both mismeasured latency and outcome (collider-adjacent) and does not purge it. Show, on the documented overlap window where both rules compute for the same program-phases, the joint distribution of (coarse latency, fine latency, outcome). (raised by pearl)
- **[empirics]** A DAG commits to refutable conditional-independence statements. Name one your graph entails whose violation falsifies the graph (not merely the estimate). Your graph asserts the two instruments affect outcomes ONLY through latency, implying a vanishing partial correlation (instrument independent of outcome residual given latency and controls). Will you pre-register and report that conditional-independence test on the assembled panel and commit that a non-vanishing partial correlation FALSIFIES the exclusion restriction (and the IV), not merely flags a robustness footnote? If not, on what basis is the relationship causal rather than associational? (raised by pearl)
- **[measurement]** Before committing to a mean-regression frame, fit the tail of the cost-growth and schedule-slip distributions on the assembled panel and estimate the tail exponent alpha (Hill estimator on upper order statistics + max-to-sum ratio convergence plot). If alpha <= 2, variance is undefined, clustered SEs estimate a non-existent moment, and beta does not converge to a stable population mean. Have you measured alpha, and if alpha < 2 what is the coefficient an estimate OF, given its target sample mean is itself non-convergent? (raised by taleb)
- **[empirics]** Run a jackknife-by-extreme-deletion on the headline latency coefficient: delete the single most extreme cost-growth program (Constellation: ~$9B, zero operational flights), then top two, then top five, and plot beta and its interval vs number of tail programs excluded. If beta flips sign, loses significance, or moves by more than its own SE when the three most catastrophic programs are removed, the finding IS those three programs and a within-program mean estimator is the wrong instrument. Will you pre-register this as a falsification condition alongside the 5.2 decision rule, and what instability threshold concedes the mean frame has failed? (raised by taleb)
- **[identification]** A Kessler-grade or Constellation-grade outcome is an absorbing/ruin event, not a marginal slip, and the H1 narrative is implicitly that compressing latency lowers tail exposure. A mean coefficient cannot establish that: a process can have a flat or favorable mean response to latency while the upper tail of cost growth gets fatter with latency. Re-pose the hypothesis on the tail, estimate whether latency shifts the tail exponent or the conditional probability of extreme overrun (quantile regression at the 90th/95th percentile, or a tail-index regression), not the conditional mean. Have you tested whether latency's TAIL effect matches its mean effect in sign and significance, and if the mean effect is null but the tail effect is positive, is that a rejection of H0 or a confirmation? (raised by taleb)
- **[measurement]** Enumerate the population of NASA programs (1958-2026) that never reached authorization or were terminated during a long latency interval (cancelled, zeroed, pre-authorization limbo) from GAO new-start/cancellation records, NTRS, and budget justifications, and report what fraction of candidate phase-observations is missing from the survivor panel for that reason. (raised by taleb)
- **[identification]** Produce a Heckman-style selection model or Manski-style bounds argument from the survivor panel plus the enumerated cancellation set that quantifies the sign and magnitude of survivorship bias on the latency coefficient, showing whether survivor-only beta is an upper or lower bound and how wide the bound is once dead programs re-enter. (raised by taleb)
- **[rival]** Demonstrate that the policy claim 'compress latency to reduce overrun' is estimable when the worst latency outcome (cancellation/ruin) is structurally excluded from every dependent variable, or restate the contribution as a within-survivor association that is silent on, and possibly opposite to, latency's effect on program-killing ruin. (raised by taleb)

## Grounded claims

- **[measurement]** The premise that a single latency scalar can conflate three causally distinct processes is well-founded: Essence of Decision establishes that the same observed outcome admits three different generative explanations, Model I (unitary value-maximizing choice, e.g., a deliberate top-management pause for technical information), Model II (organizational output of SOPs/routines/repertoires, e.g., a review queue clearing on schedule), and Model III (a political 'resultant' of bargaining among players pulling and hauling from different seats, e.g., a budget fight settled by a side payment). Because these three games have opposite reform implications, a latency measure read off outcome-records cannot, on its own, identify which game produced any given interval, so the demand to hand-code process type from the surrounding documentary record before trusting the coefficient is the correct methodological move. The decomposition logic is sound; whether the coefficient empirically survives the disaggregation is an open empirical question this retrieval cannot settle.
    - Jones, C. M. (2010), 'Bureaucratic Politics and Organizational Process Models,' Oxford Research Encyclopedia of International Studies (Allison Models I/II/III survey) | https://doi.org/10.1093/acrefore/9780190846626.013.2 | grade A
    - Roberts (2025), simulating Allison's three foreign-policy decision-making models (RAM, Organizational Process, Bureaucratic Politics) as distinct processes | https://doi.org/10.1017/S1049096525101650 | grade A
    - Hall, J. L. (2016), 'Columbia and Challenger: Organizational failure at NASA,' Space Policy, Model II organizational routine/culture (not individual irrationality) producing space outcomes inside a leading space institution | https://doi.org/10.1016/j.spacepol.2016.11.001 | grade B
- **[identification]** The identification worry is conceptually valid on Allison's own terms: authorizing-office workload is the operating variable of a Model II organizational-congestion mechanism (organizations produce outputs via routines and finite processing repertoires), and proximity to the appropriations calendar is the clock of a Model III bargaining game among the agency, OMB, and appropriations committees ('where you stand depends on where you sit'; outcomes reflect deadlines and deals). If both are themselves channels of the very organizational and bureaucratic-politics dynamics through which latency would affect cost and schedule, then treating them as exogenous nudges on a unitary actor mis-specifies them, they can plausibly hit cost/schedule directly, not only through the scalar. So the demand for a documented, seat-by-seat exclusion argument plus a falsification test (e.g., no-year-funded programs insulated from the appropriations cycle) is warranted. The retrieved corpus grounds the seat-identification of each instrument as a game variable but does NOT contain the candidate's panel cases, any overidentification/placebo result, or any IV diagnostic, so whether the exclusion restriction actually holds is unresolved here.
    - Jones, C. M. (2010), Oxford Research Encyclopedia, Model II organizational-process (SOP/routine output) and Model III governmental-politics (bargaining resultants among seated players) | https://doi.org/10.1093/acrefore/9780190846626.013.2 | grade A
    - Hall, J. L. (2016), 'Columbia and Challenger: Organizational failure at NASA,' Space Policy, empirical anchor that organizational routine/congestion drives space-program outcomes | https://doi.org/10.1016/j.spacepol.2016.11.001 | grade B
- **[rival]** The bureaucratic-politics common-cause rival is exactly the confound Allison's Model III predicts and is a live threat to the design: in Model III, observed government action is a 'resultant' of pulling and hauling among players bargaining from different seats, so a contested, multi-center program can co-produce both long latency AND high cost growth as joint residues of one underlying coalition fight, with neither causing the other. Allison's interrogation discipline explicitly warns against 'over-attributing intentionality', taking an outcome that is better explained by an organizational SOP or a bureaucratic resultant and crediting it to deliberate strategy (here, a controllable process-speed lever). Era fixed effects absorb common shocks but not program-specific bargaining intensity, which is the within-program variation the design leans on; hence the request to build an independent documentary proxy for bargaining intensity (distinct reprogramming requests, count of competing center stakeholders, congressional-mark reversals) and show robustness is the right falsification. Whether the coefficient survives that proxy is an empirical claim no retrieved source establishes.
    - Jones, C. M. (2010), Oxford Research Encyclopedia, Model III bargaining 'resultants', 'pulling and hauling', 'where you stand depends on where you sit' | https://doi.org/10.1093/acrefore/9780190846626.013.2 | grade A
    - Hall, J. L. (2016), 'Columbia and Challenger: Organizational failure at NASA,' Space Policy, outcomes as residue of organizational/political dynamics rather than individual choice | https://doi.org/10.1016/j.spacepol.2016.11.001 | grade B
- **[rival]** The bureaucratic-politics (Model III) lens predicts that pooled 'latency' is not one variable: decisions are political resultants of bargaining games among players in distinct positions ('where you stand depends on where you sit'), so a single program/era FE cannot separate a beta that is a weighted mix of seats with opposite mechanisms. This is a grounded conceptual REQUIREMENT the candidate must satisfy (a Model-III seat decomposition), NOT an empirical finding: no retrieved corpus source supplies a NASA seat-partitioned latency dataset or a within-seat vs between-seat beta to settle whether the association is homogeneous or compositional.
    - Allison Hall-of-Shoulders dossier (Model III, 'where you stand depends on where you sit'; pull-and-haul resultants), Essence of Decision ch.3 | hos://brains/allison/dossier | grade A
    - Jones, C.M., 'Bureaucratic Politics and Organizational Process Models', Oxford Research Encyclopedia | https://doi.org/10.1093/acrefore/9780190846626.013.2 | grade A
- **[identification]** Model III gives a principled reason to doubt the exclusion restriction: when an office is congested, the clearing order is a bargaining resultant by political standing, so workload can reach outcomes through coalition standing (queue priority) rather than solely through imposed latency. This makes the demanded winner-vs-loser heterogeneity test a legitimate identification challenge. This is a grounded conceptual claim about the mechanism, NOT evidence that the exclusion restriction in fact fails: no retrieved source supplies a NASA portfolio-standing variable or a winner/loser interaction estimate.
    - Allison Hall-of-Shoulders dossier (Model III: decisions as resultants of pull-and-haul among players in positions; relative power, deadlines, deals) | hos://brains/allison/dossier | grade A
    - Jones, C.M., 'Bureaucratic Politics and Organizational Process Models', Oxford Research Encyclopedia | https://doi.org/10.1093/acrefore/9780190846626.013.2 | grade A
- **[governance]** Allison's framework supports converting era FE from nuisance into identifying variation: a reorganization is precisely a Model-III intervention that changes who must say yes, so a re-seating event study is the most direct natural experiment available to test whether latency is a causal lever moving with the seat or a passive correlate of administrative regime. This is a grounded methodological claim about the right design, NOT an empirical result: no retrieved source supplies a coded set of NASA re-seating events or an event-study estimate of co-movement between latency and outcomes.
    - Allison Hall-of-Shoulders dossier (Model III reorganizations re-seat players / abolish clearances; outcomes as resultants of who holds which seat) | hos://brains/allison/dossier | grade A
    - Jones, C.M., 'Bureaucratic Politics and Organizational Process Models', Oxford Research Encyclopedia | https://doi.org/10.1093/acrefore/9780190846626.013.2 | grade A
- **[identification]** Callaway, Goodman-Bacon and Sant'Anna (2024) prove that under a continuous treatment, a parallel-trends assumption analogous to the binary case identifies treatment-on-the-treated-type parameters at a given dose, but comparing those parameters ACROSS dose levels is not licensed by parallel trends because selection-into-dose generates bias: units selecting higher doses differ systematically. Reading a cross-dose slope (beta per +1 month) as a marginal effect therefore requires an additional stateable assumption (e.g. homogeneity of the dose-response across the latency distribution, or strong parallel trends in every potential-dose outcome). Absent that assumption and absent an empirical demonstration that the ATT(latency) curve is flat across the dose support, the claim must be restricted to ATT-at-observed-dose, not a marginal effect. The methodological standard is settled by the literature; the empirical ATT(latency) curve is the candidate's burden and is not in any retrieved source.
    - Callaway, Goodman-Bacon & Sant'Anna, 'Difference-in-Differences with a Continuous Treatment' (2024 working paper) | https://doi.org/10.2139/ssrn.4716682 | grade B
    - Callaway & Sant'Anna, 'Difference-in-Differences with multiple time periods,' Journal of Econometrics (2021) | https://doi.org/10.1016/j.jeconom.2020.12.001 | grade A
- **[identification]** The correct test for the anticipation/baseline-gaming worry is pre-treatment (placebo) ATT(g,t) estimates: a flat pre-trend in the periods before the authorization event is what an honest event-study must surface, and anticipation effects (e.g. land speculation before a launch) violate the no-anticipation assumption unless explicitly modeled. Two caveats from the cited literature bind the inference. First, Roth (2022) shows a flat pre-trend is NECESSARY BUT NOT SUFFICIENT: conventional pre-trends tests have low power, and conditioning the analysis on having passed a pre-test can itself distort the estimate, so a clean placebo plot does not by itself rescue no-anticipation. Second, Roth, Sant'Anna, Bilinski & Poe (2023) require stating the canonical assumptions, naming which are relaxed, and reporting sensitivity to parallel-trends/anticipation violations rather than a single point estimate. The candidate's baseline-gaming concession is, in this framework, an anticipation violation, and only pre-period placebo ATT(g,t) plus a sensitivity analysis can settle whether the latency coefficient is contaminated. The methodological resolution is grounded; the candidate's actual placebo estimates are not in retrieval.
    - Roth, 'Pretest with Caution: Event-Study Estimates after Testing for Parallel Trends,' American Economic Review: Insights (2022) | https://doi.org/10.1257/aeri.20210236 | grade A
    - Roth, Sant'Anna, Bilinski & Poe, 'What's Trending in Difference-in-Differences? A Synthesis of the Recent Econometrics Literature,' Journal of Econometrics (2023) | https://doi.org/10.1016/j.jeconom.2023.03.008 | grade A
- **[empirics]** Goodman-Bacon (2021) establishes that a TWFE/2x2 estimator under staggered timing is a weighted average of all possible two-group/two-period comparisons, and when effects grow over time the comparisons that use already-treated units as controls enter with negative weights, which can drive the headline estimate toward zero or the wrong sign. The candidate's IV does not escape this by construction: instrumenting a staggered, dynamic, heterogeneous treatment inherits the same forbidden-comparison weighting unless the instrumented estimand is itself decomposed and shown convex. The Callaway-Sant'Anna corrective (excluding already-treated units from the control pool, estimating disaggregated ATT(g,t), and aggregating with transparent researcher-chosen weights) is the design standard the candidate must meet for the instrumented estimand as well. The dossier's review-lens makes the demand explicit: if any already-treated units serve as controls or any weight is negative, the headline coefficient is a contaminated average and the cohort-by-period ATT(g,t) is required instead. This grounds the demand and the corrective; it does not supply the candidate's decomposition or an exclusion-restriction defense, both of which are empirical/argumentative burdens absent from retrieval.
    - Goodman-Bacon, 'Difference-in-differences with variation in treatment timing,' Journal of Econometrics (2021) | https://doi.org/10.1016/j.jeconom.2021.03.014 | grade A
    - Callaway & Sant'Anna, 'Difference-in-Differences with multiple time periods,' Journal of Econometrics (2021) | https://doi.org/10.1016/j.jeconom.2020.12.001 | grade A
- **[empirics]** PARTIAL/REFUSED on the quantitative core. The dissertation is a design-stage proposal that explicitly does not report estimates from the assembled panel: 'The study is a design-stage proposal and analysis plan. It does not report estimates from the full assembled dataset... an illustrative output table appears in the analysis-plan chapter with its cells deliberately left unpopulated' (Table 6.1, 'Illustrative coefficient table, deliberately unpopulated by design'). The Goodman-Bacon decomposition is named only as a diagnostic to be 'reported' once estimated, not a computed result. Because no headline FE-IV beta has been fitted, no per-program or per-era implied-weight decomposition exists to report or rank; producing one would require fabricating both the coefficient and its weights, which the no-confabulation contract forbids. The design does acknowledge the underlying concern: it pre-commits to wild-cluster bootstrap and alternative-clustering checks 'so that conclusions are not driven by a few large modern programs,' which is the within-sample acknowledgment of exactly the weight-concentration risk the question raises, but it is a planned safeguard, not a delivered decomposition.
    - PHD-08 dissertation.md, Section 1.7 (Scope and delimitations), line 158 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md, abstract/overview, line 21 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md, list of tables (line 53) and Section 4.2 (line 549) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Hall-of-Shoulders thinker brain hos-callaway_santanna dossier | https://doi.org/10.1016/j.jeconom.2021.03.014 | grade A
- **[rival]** The panel DOCUMENTS the era-regime partition structure but cannot yet report era-by-era ATT(latency). Era fixed effects map to a five-band regime sequence (delta_1 pre-formal-baseline; delta_2 early program-management formalization; delta_3 NPR 7120.5 life-cycle-review; delta_4 joint-confidence-level; delta_5 Standing-Review-Board), and coverage band B.1 (1958-1969, 1970-1989, 1990-2009, 2010-2026) records that early bands enter at coarse resolution only. The design also pre-commits to report descriptive latency and outcome distributions for each era regime before estimation. But both Table C.1 and Table B.1 are explicitly 'illustrative bands; exact boundaries fixed at assembly' / 'entries populated at panel assembly' -- the partition is defined, the era-disaggregated estimates are not produced. So the disaggregation-before-aggregation discipline the panelist demands is structurally anticipated, yet whether the modern-regime ATT holds sign/magnitude vs the pooled number is unsettleable from the current artifact: no pooled estimate and no era-specific estimate exist.
    - PHD-08 dissertation.md, Appendix C (Era-Regime Definition Table), lines 1368-1380 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md, Appendix B coverage table (line ~1355) and Section 6 describe-before-estimate (line 831) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Hall-of-Shoulders thinker brain hos-callaway_santanna dossier | https://doi.org/10.1016/j.jeconom.2020.12.001 | grade A
- **[identification]** The panel STATES the target population and DOCUMENTS the cancelled/censored-cohort structure as a named limitation, but cannot yet demonstrate that the aggregated estimand's implicit weights transport to it. Target population: NASA programs with documented baselines, 1958-2026, with the policy lever ('compress authorization latency') aimed at program execution management for in-flight and future programs (NORTH STAR / JPL category); the dissertation explicitly confines generalization to NASA and refuses generalization to other agencies or commercial programs. The cancelled/censored cases the panelist names are present and flagged: Constellation (~$9B over ~5 years, no operational missions before cancellation) and the SLS upgraded upper stage plus second mobile launcher (terminated across the ~41-month Artemis I-to-II interval) enter only as program-record-derived anecdotes, not measurements. The design concedes the panel is unbalanced by construction, that some cadence outcomes are 'censored or partially observed' (notes but does not adopt nonparametric censored-panel estimators as a sensitivity lane, citing Yoon 2024), and that NTRS coverage-density bias over-represents programs that produced more formal reporting (correlated with size and era) -- which is the survivorship/selective-attrition tilt the question targets. But the demonstration that the estimand's implicit weights MATCH the in-flight/future target population requires a fitted estimand whose weights can be computed; none exists. So transportability is acknowledged as a threat and partially bounded by design (balanced-modern-subpanel reporting, censored-outcome sensitivity lane, coverage-density caveat), not demonstrated.
    - PHD-08 dissertation.md, Section 1.7 (line 154) and Section 1.6 significance (lines 144-148) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md, Chapter context, line 92 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md, Sections 4.2 (line 549), 6 estimation choices (line 680), 4 known biases (line 525) | https://doi.org/10.2139/ssrn.4953073 | grade B
- **[identification]** Forrester's apparatus holds that complex program behavior is dominated by closed loops of causation, so a relationship like latency->trouble must be presumed bidirectional until tested: trouble (a growing backlog of pending actions) is itself an inflow to review activity that extends latency. This is a falsifiable demand a system-dynamics reviewer is entitled to make, and the canonical project literature operationalizes exactly this bidirectional structure (rework, schedule-pressure, and ripple-effect loops feeding back on schedule). HOWEVER, no source retrieved this turn contains PHD-08's panel or any cross-lagged Granger test on it; the candidate's data are not shown to establish (or refute) cross-lag asymmetry. The grounded conclusion is methodological: a single-equation FE+IV slope cannot, by construction, distinguish an open-loop coefficient from a closed-loop one, so the asymmetry test is required before the beta can be read as one-way causation.
    - Forrester, Industrial Dynamics (1961) / Urban Dynamics, forrester dossier hall-of-shoulders | https://doi.org/10.2307/214050 | grade A
    - Lyneis et al., 'Quantifying the impacts of rework, schedule pressure, and ripple effect loops on project schedule performance', System Dynamics Review | https://doi.org/10.1002/sdr.1551 | grade A
    - Lyneis & Ford, 'System dynamics applied to project management: a survey, assessment, and directions', System Dynamics Review | https://doi.org/10.1002/sdr.377 | grade A
- **[governance]** Forrester's 'Counterintuitive Behavior of Social Systems' establishes that complex social systems are insensitive to most chosen policy levers and tend to respond by counteracting them (policy resistance), because actors see local short-term cause-and-effect but not the closed-loop delayed structure; Urban Dynamics gives the worked case where a well-intentioned low-leverage parameter intervention deepened the condition it targeted. The project-dynamics literature supplies the compensating mechanism in this specific domain: cutting authorized review time raises schedule pressure, which increases rework and ripple-effect loops, partially restoring elapsed schedule. So a measured latency reduction need not compress the closed-loop outcome. BUT no source retrieved this turn contains PHD-08's regime-switch trajectories or any post-intervention event study on them. The grounded result is that the static association beta cannot reveal whether the lever moved the system or the system absorbed it; only a post-regime trajectory test can, and that test is absent from the corpus.
    - Forrester, 'Counterintuitive Behavior of Social Systems' (1971) | https://doi.org/10.1007/bf00148991 | grade A
    - Forrester, Urban Dynamics (low-leverage policy deepens stagnation), forrester dossier hall-of-shoulders | https://doi.org/10.2307/214050 | grade A
    - Lyneis et al., 'Quantifying the impacts of rework, schedule pressure, and ripple effect loops on project schedule performance', System Dynamics Review | https://doi.org/10.1002/sdr.1551 | grade A
- **[measurement]** Sterman's bathtub-dynamics result empirically demonstrates the stock-flow failure: even expert subjects systematically misinfer an accumulation's trajectory from a contemporaneous flow, because material and information delays between a flow and its effect on a stock are not mentally simulable. Cost growth and schedule slip are stocks (accumulations); per-phase latency is a flow. Forrester's delay principle requires that an identification respect the delay structure: if the dominant latency-to-outcome delay exceeds the phase window, the within-phase contemporaneous beta mis-assigns to phase p an accumulation seeded in p but realized later, contaminating the within-program identification. The rework/ripple-effect project literature confirms that the consequences of a schedule action propagate across later stages rather than landing contemporaneously. HOWEVER, no source retrieved this turn contains a measurement of PHD-08's latency-to-outcome lag relative to its phase window; whether the delay is sub-phase or supra-phase is not established by the corpus. The grounded conclusion is that the test (estimating the impulse-response lag of accumulated cost/schedule to a phase's latency) is mandatory and the phase-as-unit choice is unjustified until it is run.
    - Sterman, 'Bathtub Dynamics: initial results of a systems thinking inventory', System Dynamics Review | https://doi.org/10.1002/sdr.198 | grade A
    - Forrester, delays/limits-of-mental-model principle, forrester dossier hall-of-shoulders (Industrial Dynamics) | https://doi.org/10.2307/214050 | grade A
    - Lyneis et al., 'Quantifying the impacts of rework, schedule pressure, and ripple effect loops on project schedule performance', System Dynamics Review | https://doi.org/10.1002/sdr.1551 | grade A
- **[mechanism]** The objection is methodologically valid and is exactly the test Forrester's apparatus prescribes. A reinforcing latency->rework->latency loop is a closed internal feedback structure; a static level-on-level regression with additive fixed effects encodes an open one-pass path and structurally cannot represent or detect it. Forrester's own endogeneity test demands the analyst reproduce the troublesome behavior using only the system's internal feedback structure with no exogenous shock, and if an external driver is needed to get the behavior, identify the missing loop; the reinforcing-loop archetype (self-sustaining collisional cascade) is the canonical form this takes in the space domain. If the candidate's narrative posits a loop, latency is a flow inside that loop rather than an exogenous regressor, and a treating it as exogenous biases beta. I can ground that the critique is correct in principle, but retrieval contains NO KDP-dated within-program authorization-event count panel for this candidate's data, so I CANNOT settle whether such a count actually rises with accumulated cost-growth or whether beta is stable once conditioned on the loop state. That empirical test is unresolved by retrieval.
    - forrester dossier (review lens, Hall of Shoulders brain hos-forrester) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/forrester/ | grade C
    - Forrester, Industrial Dynamics (system-dynamics foundation) | https://doi.org/10.1002/sdr.284 | grade A
    - Kessler & Cour-Palais, Collision frequency of artificial satellites: the creation of a debris belt (1978) | https://doi.org/10.1029/JA083iA06p02637 | grade A
- **[measurement]** The measurement objection is sound and rests on documented evidence. Cost growth and schedule slip are stocks built from their flows; Sterman's bathtub-dynamics result empirically demonstrates that even highly educated subjects systematically misread a stock built from its flows (the stock-flow failure), which is precisely why a phase-end stock regressed on latency risks confounding a real rate effect with mechanical integration over a longer elapsed window. Forrester's apparatus insists on a stock-versus-flow decomposition and on separating timescales (the slow accumulation constraint bounding the fast flow). The correct discriminating test is to regress the dated within-phase accrual RATE on latency rather than the end-of-phase stock; if latency predicts the instantaneous accrual rate, the effect is real, whereas if it predicts only the end stock, the result may be an integration artifact of elapsed time. I can ground that this is the right test and that the artifact risk is real, but retrieval contains NO dated within-phase re-baseline accrual series for this candidate's programs, so I CANNOT settle whether latency predicts the rate or merely the integrated stock. That discrimination is unresolved by retrieval.
    - Sterman, Bathtub Dynamics: initial results of a systems thinking inventory (2000), System Dynamics Review | https://doi.org/10.1002/sdr.198 | grade A
    - Forrester, Industrial Dynamics (system-dynamics foundation) | https://doi.org/10.1002/sdr.284 | grade A
- **[empirics]** The aggregation-discipline objection is well founded. Forrester's review lens explicitly raises aggregation discipline: an analysis that evaluates actors individually must be re-run against the aggregate, because the stock integrates all of them and an actor-level estimate need not compose into the system behavior. The space literature documents the congestion mechanism the question invokes: the coordination/workload burden grows with the number of objects and operators, so system-level conjunction and unresolved-encounter rates scale super-linearly with population, a rising-load-into-a-delayed-control-loop pattern that produces the oscillation and overshoot Forrester's delay analysis predicts; and economists describe the same defect as an un-priced congestion externality where each actor's contribution is imposed on others, so the decision loop is open where it should close. This is the structural basis for a fallacy-of-composition warning: individually estimated effects need not aggregate to the emergent queue behavior. I can ground that the office-by-year queue-congestion test is the correct discipline and that congestion feedback is documented, but retrieval contains NO authorizing-office-by-year aggregate panel for this candidate's data, so I CANNOT settle whether agency latency exhibits queue-congestion dynamics empirically or whether the summed program-level betas reproduce the observed cadence series. That composition test is unresolved by retrieval.
    - forrester dossier (review lens, Hall of Shoulders brain hos-forrester) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/forrester/ | grade C
    - Colombo, Martinez, Letizia et al., Space capacity management and its interaction with space traffic management, Acta Astronautica (2025) | https://doi.org/10.1016/j.actaastro.2025.01.069 | grade A
    - Weinzierl, Space, the Final Economic Frontier, Journal of Economic Perspectives (2018) | https://www.aeaweb.org/articles?id=10.1257/jep.32.2.173 | grade B
- **[measurement]** GROUNDED REFUSAL OF THE EMPIRIC, MEASUREMENT PRINCIPLE CONCEDED. The dissertation is explicitly design-stage and reports no executed estimates (Ch1.7: 'The study is a design-stage proposal and analysis plan. It does not report estimates from the full assembled dataset'; Ch6 'reports no executed estimates'). The two-resolution design is described (Sec 4.3.2: coarse milestone-to-milestone for the full 1958-2026 span; fine key-decision-point for the modern subperiod only) and the candidate even names splice-validation as the intended falsification test ('if a latency effect appears only in the fine measure... it could be an artifact of denser documentation; if it appears in the coarse measure across the full span as well, that artifact explanation is much weaker'). But the specific artifact Maddison demands, the overlap-window Pearson correlation, the coarse-minus-fine level gap in months, and a test of whether that gap is constant or drifts, is NOT computed anywhere in the dissertation. No overlap correlation, no level offset, no drift estimate exists to report. The candidate's own Maddison single-rule standard (Sec 4.3.1, 'the same operational definition... must be applied identically in 1962 and in 2024, or the resulting series is not comparable') makes the demand legitimate and currently unmet: comparability is asserted via era fixed effects and the codebook mapping of pre-modern milestones, not demonstrated by a stable level offset on the overlap. I therefore cannot supply the correlation, the gap, or the drift verdict from retrieval; they would have to be produced by executing the panel. The canonical method for the principle Maddison invokes, holding measurement on a common real footing across time before any cross-era comparison, is the Maddison/Penn-World-Table real-comparison tradition the candidate cites as ref [16].
    - PHD-08 dissertation.md Sec 1.7, 1.8, 4.3.1-4.3.2, 6 (Chapter 6 pre-registered plan, no estimates) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Feenstra, Inklaar & Timmer, The Next Generation of the Penn World Table (via maddison thinker brain dossier source) | https://doi.org/10.1257/aer.20150954 | grade A
- **[measurement]** GROUNDED REFUSAL OF THE EMPIRIC, SHARED-DENOMINATOR CONTAMINATION ACKNOWLEDGED BUT NOT QUANTIFIED. The dissertation concedes exactly the defect Maddison names. Limitation 1 (Sec 4.7): 'the documentary definition of a baseline changed over time'; Sec 4.6 states plainly 'A baseline set informally in 1965 and a baseline set at the seventy-percent joint confidence level in 2015 are not the same object, and a cost-growth number computed against each is not strictly comparable,' and Sec 1.7 flags baseline optimism/gaming as endogenous. Baseline is indeed the denominator of cost growth and schedule slip (Sec 1.8 definitions) and the candidate's stated remedy is era fixed effects (delta_t) plus the two-resolution design plus a baseline-conservatism check (Sec 4.7 Limitation 3, reserves-to-baseline vs latency). But the specific artifact Maddison demands, the distribution of latency and cost-growth computed against the ORIGINAL versus the RE-SET baseline for the re-baseline subset, and an explicit by-era statement of which baseline the panel uses, is NOT in the dissertation. No re-baseline subset distribution is reported; the rule for choosing original vs reset baseline by era is not fixed in the document. The candidate's own concession that the early baseline is an informal practitioner target and the late one a JCL/GAO-anchored commitment is precisely Maddison's 1960-GDP-guess-vs-2020-national-accounts objection, and it exposes that era fixed effects cannot net out a distortion that is correlated with the regressor (harder/longer-latency phases are also the ones more likely to be re-baselined). I cannot produce the two distributions or the by-era baseline rule from retrieval; they require the re-baseline subset from GAO/NASA records to be assembled and tabulated, which the design-stage panel has not executed.
    - PHD-08 dissertation.md Sec 4.6 (changing documentary baseline), Sec 4.7 Limitations 1 & 3, Sec 1.8 (cost-growth/schedule-slip denominators), Ch6 (no executed estimates) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - maddison thinker brain dossier (constant comparable real units across years as precondition of cross-era comparison) | local:D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/maddison (hos-maddison dossier) | grade C
- **[identification]** Maddison's periodization standard is binding here and the candidate must meet it: phases must be defined by breaks in the measured series (growth and productivity rates) rather than by political events or a reform calendar. Drawing era cuts at administrative-reform dates makes the era dummies co-determined with the authorization-regime variation, so the fixed effects absorb the treatment; the correct procedure is to derive boundaries from a documented structural break in a measured series and then demonstrate the latency coefficient is invariant across at least two defensible periodizations, because the long horizon shows apparent trend breaks frequently revert and must be tested, not assumed. The candidate has not published this break-detection rule, so the era construction is currently a gap; the standard that must govern it is documented.
    - Maddison, A. (1982), Phases of Capitalist Development, Oxford University Press (ISBN 9780198284505) -- the work in which Maddison defined phase boundaries by measured breaks in growth and productivity rates, not political events | https://www.worldcat.org/isbn/9780198284505 | grade C
    - Bolt, J., & van Zanden, J. L. (2017), 'The Maddison Project: Historical GDP Estimates Worldwide', Journal of World-Historical Information / Maddison Project Database | https://doi.org/10.5195/jwhi.2017.46 | grade C
- **[mechanism]** Maddison's growth-accounting discipline requires exactly this decomposition: reading a single deflated cost-growth scalar as evidence that latency 'drives' overrun is forbidden until growth is split into accumulation (the mechanical standing-cost component that is close to an identity with elapsed time) versus the total-factor-productivity residual (the substantive claim about whether faster decisions are a real lever). The candidate must compute the residual; if they cannot, they have not shown the cost growth is genuine productivity loss rather than merely extensive accumulation that elapsed schedule predicts by construction. The space literature has not yet performed a formal sector TFP decomposition, which is itself the finding Maddison would flag, so the decomposition the candidate owes is both required and currently absent.
    - Maddison Project Database 2020 (Bolt, J. & van Zanden, J. L.) -- the living continuation of Maddison growth-accounting framework covering GDP, capital, and productivity by country from antiquity to 2018, embodying the factor-decomposition discipline Maddison developed in Dynamic Forces in Capitalist Development (Oxford 1991) | https://www.rug.nl/ggdc/historicaldevelopment/maddison/releases/maddison-project-database-2020 | grade C
    - Feenstra, R. C., Inklaar, R., & Timmer, M. P. (2015), 'The Next Generation of the Penn World Table', American Economic Review 105(10):3150-3182 | https://doi.org/10.1257/aer.20130954 | grade C
    - Paravano, A., Rosseau, B., Locatelli, G., Weinzierl, M., & Trucco, P. (2024), 'Toward the LEO economy: A value assessment of commercial space stations', Acta Astronautica 228 | https://doi.org/10.1016/j.actaastro.2024.11.060 | grade A
- **[measurement]** Maddison's rule for an unbalanced panel that is dense for the modern segment and sparse for 1958-1980 is the conjectural-but-explicit estimate: do not leave the early record blank, do not let the well-measured modern window masquerade as the long run, but publish the early figures with documented uncertainty bands and prove the headline finding is not an artifact of the dense period. This obliges the candidate to (a) report per-decade observation counts and the point-identified-versus-bounded share with explicit uncertainty, and (b) re-estimate on the pre-1980 segment alone; honesty about how little is actually measured in the thin window must replace a false-precision long-run claim. The candidate has not supplied this coverage accounting or the pre-1980 re-estimate, so it stands as the gap the standard demands be filled.
    - Maddison, A. (2007), Contours of the World Economy 1-2030 AD: Essays in Macro-Economic History, Oxford University Press (ISBN 9780199227211) -- the work in which Maddison applied the conjectural-but-explicit estimate standard to reconstruct GDP from 1 AD to 2030 with documented uncertainty | https://www.worldcat.org/isbn/9780199227211 | grade C
    - Eastwood, J. P., Biffis, E., Hapgood, M. A., Green, L., Bisi, M. M., et al. (2017), 'The Economic Impact of Space Weather: Where Do We Stand?', Risk Analysis 37(2):206-218 | https://doi.org/10.1111/risa.12765 | grade A
- **[measurement]** The construct-validity threat in Q1 is real and grounded in theory: realized strategy is the joint product of deliberate and emergent streams, and patterns can be realized in the absence of, or despite, prior apex intention, so a recorded authorization event may ratify a commitment that crystallized earlier rather than constitute the decision itself. This concession is supportable. However, the empirical object Q1 actually demands, a measured divergence distribution (in months) between the reconstructed commitment-crystallization date and the recorded authorization date across a stratified sample of NASA program-phases, was NOT retrieved from any corpus this turn (Mintzberg thinker-brain, AMOS, ACTA, Space Economy, OpenAlex, NTRS). The candidate cannot, on retrieved evidence, assert that the two dates diverge by any specific number of months, nor that measured latency tracks the commitment-to-paperwork gap rather than the trigger-to-choice gap. The theoretical critique stands; the quantitative answer is absent.
    - Mintzberg & Waters, Of strategies, deliberate and emergent, Strategic Management Journal (1985) | https://doi.org/10.1002/smj.4250060306 | grade A
- **[mechanism]** Q2's premise is grounded: structured observation of practicing chief executives shows managerial work is fragmented, fast-paced, interruption-driven, and verbally mediated, which is precisely the decision activity that leaves no dated artifact, so two programs with identical documentary entries can have radically different real deliberation. Separately, standardized review machinery has been documented to record compliance rather than learning in NASA, reinforcing that the dated review trail can be a paperwork surface rather than the substantive process. These two points are supportable as a threat to the construct. But the test Q2 demands, an actual measured correlation between documentary latency and an independent process-derived deliberation-duration measure built from oral histories or the Apollo management-control record, was NOT retrieved this turn. No source supplies a process-derived deliberation-duration series for NASA program-phases, so the candidate cannot demonstrate that the two co-move (or fail to). On retrieved evidence the variable's claim to index decision speed rather than records-keeping density is unestablished.
    - Mintzberg, Managerial Work: Analysis from Observation, Management Science (1971) | https://doi.org/10.1287/mnsc.18.2.b97 | grade A
    - Madsen & Dillon, Near-Miss Evaluation Bias as an Obstacle to Organizational Learning: Lessons from NASA, NASA NTRS (2006) | https://ntrs.nasa.gov/citations/20060020000 | grade C
- **[mechanism]** The critique is theoretically grounded and demands a configuration-stratified estimate. Mintzberg's Structure in Fives establishes five configurations, each with a distinct dominant coordinating mechanism: machine bureaucracy (standardization of work processes, technostructure-dominant), professional bureaucracy (standardization of skills, operating-core-dominant), and adhocracy (mutual adjustment, fused project teams). NASA is explicitly a hybrid: a machine-bureaucratic procurement/safety/review apparatus grafted onto a professional-bureaucratic engineering core, with adhocratic mission teams. Because review-time means conformance under standardization but obstructs mutual adjustment under adhocracy, a single pooled latency slope can average effects of opposite sign and opposite economic meaning. A grounded answer therefore concedes the pooled beta is not interpretable until stratified by the dominant configuration of each program-phase (routine production vs. novel first-of-kind development).
    - Mintzberg, Structure in Fives: Designing Effective Organizations (1983) | https://doi.org/10.2307/2393181 | grade A
    - Hall of Shoulders Mintzberg dossier (collegium), anchoring Mintzberg 1983 | https://doi.org/10.2307/2393181 | grade A
    - Madsen & Dillon, Near-Miss Evaluation Bias as an Obstacle to Organizational Learning: Lessons from NASA (NTRS 20060047554, 2006) | https://ntrs.nasa.gov/citations/20060047554 | grade A
- **[measurement]** The construct-validity attack is well-founded and the candidate cannot rebut it from the retrieved record alone. Mintzberg's structured-observation evidence (Managerial Work, 1971) shows managerial work is fragmented, fast-paced, interruption-driven, and verbally mediated, contrary to the plan-organize-coordinate-control folklore; substantive decisions are reached orally and the documentary record captures only the later ratification. This directly predicts that a documentary trigger and a documentary authorization timestamp are ratification artifacts that bracket a process which substantively opened and closed earlier. Worse, the safety-culture corpus shows standardized review machinery records compliance/ceremony rather than substance, so the ratification lag plausibly scales with administrative ceremony, which itself scales with phase difficulty, biasing beta. The honest response is to concede the threat and that validating it requires a hand-coded oral-history/correspondence sample the dissertation does not present; nothing retrieved supplies that validation.
    - Mintzberg, Managerial Work: Analysis from Observation (Management Science, 1971) | https://doi.org/10.1287/mnsc.18.2.b97 | grade A
    - Madsen & Dillon, Near-Miss Evaluation Bias as an Obstacle to Organizational Learning: Lessons from NASA (NTRS 20060047554, 2006) | https://ntrs.nasa.gov/citations/20060047554 | grade A
- **[identification]** The identification objection is theoretically correct: era fixed effects that absorb reorganizations soak up precisely the regime variation that would discriminate cause from symptom. Mintzberg's diagnosis of configuration mismatch, forcing a machine-bureaucratic logic onto innovative (adhocratic) work, makes long latency and high cost co-products of one ill-fitting structure rather than latency causing cost. NASA institutional history (Marshall Space Flight Center, 1960-1990) documents recurring shifts between centralized program control and decentralized engineering expertise, i.e., real reorganizations that move the center between configurations, which is exactly the regime variation era-FE discards. So the candidate must concede the era-FE strategy is conservative against, not supportive of, the causal claim.
    - Hall of Shoulders Mintzberg dossier (collegium), anchoring Mintzberg 1983 | https://doi.org/10.2307/2393181 | grade A
    - Dunar & Waring, Power to Explore: A History of the Marshall Space Flight Center 1960-1990 (NASA SP-4313, NTRS 20000031366, 1999) | https://ntrs.nasa.gov/citations/20000031366 | grade B
- **[identification]** North's objection is grounded and load-bearing: institutions (the rules of the game) must be distinguished from the players, and a fixed-effects design that absorbs the rule regime into era effects and the standing organization into program effects risks differencing out exactly the institutional variation it claims to measure, leaving residual variation that may be organizational or idiosyncratic. The empirical resolution the question demands, regressing residual latency on a coded register of rule-text changes and reporting R-squared, is a result internal to the candidate's assembled panel and is NOT present in any retrieved source; therefore the grounded expert can confirm the objection's force but cannot supply or invent the validation result.
    - North, D. C. (1990), Institutions, Institutional Change and Economic Performance (Hall of Shoulders North dossier, frameworks 1-2: institutions as rules of the game; institutions vs organizations / players vs rules) | https://doi.org/10.1017/CBO9780511808678 | grade A
- **[measurement]** The construct-validity objection is grounded in North's method and in the contemporary institutional-dataset literature: formal labels and operative rules routinely come apart, so a documentary (de jure) measure cannot be assumed to track the de facto rule-in-use without validation. Pic, Evoy & Morin (2023) show empirically across 1042 space arrangements that invoking 'global commons' language 'does not result in significantly different operational rules,' and Morin & Couette (2025) show a formally polycentric architecture 'fail[s] to promote sustainability norms,' both instances of the de jure / de facto gap North's program is built to expose. This confirms the objection is real and method-founded. The specific validation result the question demands, a correlation between the candidate's documentary latency and an independent de facto decision-timing measure on a program subset, is internal to the candidate's data and is NOT in any retrieved source; it cannot be supplied or invented.
    - Pic, P., Evoy, P., & Morin, J.-F. (2023), Outer Space as a Global Commons: An Empirical Study of Space Arrangements, International Journal of the Commons 17 (de jure label vs de facto operative-rule gap, 1042 arrangements) | https://doi.org/10.5334/ijc.1271 | grade A
    - Morin, J.-F., & Couette, C. (2025), The Missing Ingredients for a Polycentric Governance System of Orbital Debris, Global Environmental Politics 25 (formal architecture vs effective rules-in-use) | https://doi.org/10.1162/glep_a_00775 | grade A
- **[rival]** North's grounds for the exclusion-restriction challenge are sound: the rules of the game determine which actions are routed to an authorizing office (workload), so an instrument drawn from workload is endogenous to the very rule regime under test rather than an exogenous shock to it; and an institution with its own path-dependent effects (the appropriations cycle) can move cost and schedule through channels independent of latency. North's discipline requires identifying the actual transaction-cost mechanism and separating the rules from the players before a causal claim is licensed. The empirical demonstration the question demands, a test that workload has zero partial association with the outcomes conditional on latency, is a result internal to the candidate's panel and is NOT present in any retrieved source; the grounded expert can affirm the objection but cannot fabricate the exclusion-restriction test or its outcome.
    - North, D. C., & Weingast, B. R. (1989), Constitutions and Commitment, Journal of Economic History 49(4):803-832 (transaction costs, credible commitment; Hall of Shoulders North dossier framework 4) | https://doi.org/10.1017/S0022050700009451 | grade A
    - North, D. C. (1990), Institutions, Institutional Change and Economic Performance (path dependence / increasing returns to institutions; rules route the players' actions; Hall of Shoulders North dossier frameworks 2 and 5) | https://doi.org/10.1017/CBO9780511808678 | grade A
- **[governance]** On the governance/scope-condition logic of Q2 the framework anchor is citable but does NOT supply the empirical coefficient: transaction cost is a property of the rules of the game, not a portable constant, and institutional change is incremental and path-dependent, so a within-NASA estimate is bounded by NASA's enabling rules and is not entitled to portable-law status without a stated scope condition. This grounds the demand that the candidate name the enabling rule and bound the inference, but no retrieved source estimates, replicates, or partitions the candidate's latency-to-overrun coefficient.
    - North, D. C. (1990), Institutions, Institutional Change and Economic Performance, Cambridge University Press (north thinker dossier, Hall of Shoulders brain hos-north) | https://doi.org/10.1017/CBO9780511808678 | grade A
- **[identification]** Partially answerable from the candidate's own documentary record, then it stalls at the empirical test Pearl demands. The dissertation defines eras at the institutional-regime granularity (Apollo / Space Shuttle / Constellation / modern), holding apart agencies that authorized 'against an open-ended national commitment' vs 'a cost-recovery promise' vs 'an explicit affordability constraint'; this is a multi-decade regime partition, NOT year-resolution. Critically, the candidate's own text confirms the back-door Pearl draws: the appropriations-calendar instrument and the funding-instability index are BOTH constructed from the same NASA budget record's 'requested-versus-appropriated dynamics' (the index is built 'from year-over-year deviations between requested and appropriated funds'), and the candidate's stated defense is that 'the same funding-instability control absorbs the direct-funding channel.' On Pearl's DAG, conditioning on a control built from the same fiscal source the instrument exploits does not establish instrument independence; it is the conditioning-on-a-correlated-node move whose validity is exactly what must be tested. The dissertation is design-stage and explicitly reports no estimates from the assembled dataset, so the one conditional independence Pearl names (instrument residual indep of funding-instability index | era) is pre-specified as a probe but never executed and no within-era correlation number exists. Grounded conclusion: the partition is regime-coarse and the co-source risk is real and conceded; the independence test that would settle exclusion is absent.
    - PHD-08 dissertation.md (candidate's own record), Sec 1.3 periodization & Sec 5.3 identification | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md Sec 4.5 / 5.4 instrument validity | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
- **[measurement]** Pearl has correctly read the candidate's own concession, and the candidate's record confirms the definitional-overlap risk without supplying the disjointness proof that would dispel it. The dissertation states explicitly that schedule slip is predicted largest and most robust 'because authorization latency is itself a component of elapsed schedule and the path from latency to slip is the most direct of the three.' Latency is operationalized as 'the median elapsed time across the authorization events in a phase,' measured between trigger and resolution events drawn from the same lifecycle-review architecture (KDPs, PCRs, confirmation reviews, NPR 7120.5 reviews, Standing Review Board) that also bounds the phase intervals from which schedule slip is computed. The candidate provides a measurement TAXONOMY (a two-resolution scheme: coarse milestone-to-milestone for the full span, fine KDP for the modern subperiod) but does NOT provide a set-theoretic decomposition proving the authorization sub-interval is disjoint from the engineering-execution sub-interval of the same phase. By the candidate's own 'component of elapsed schedule' language, the latency interval is presented as a SUB-interval of elapsed phase time, which is the subset relation Pearl flags: if the schedule-slip numerator is elapsed phase time inclusive of the authorization months, the Latency -> Schedule-slip arrow is an identity, not an estimate. Grounded conclusion: the candidate concedes the partial-containment, the disjointness proof Pearl asks for is not in the record, and on the current operationalization the schedule-slip coefficient is at risk of being mechanically tautological exactly as charged.
    - PHD-08 dissertation.md Sec 5.3 / Ch 3 mechanism statement (candidate's own concession) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md Sec 4.3 measurement, Sec 4.4 outcomes | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Majerowicz & Shinn, schedule-delay vs cost-overrun analysis (candidate ref [5]); corroborated by NASA instrument/mission schedule-growth literature | https://doi.org/10.1109/aero.2014.6836219 | grade B
- **[identification]** The candidate names exactly this reverse-causation threat and answers it with the precise temporal-ordering move Pearl interrogates, but supplies no dated event trail to license it. Section 5.5.1 states: 'a program already overrunning may generate more authorization events and longer latency as a consequence of trouble,' and the design's responses are 'to measure latency early in each phase, before most of the phase's cost growth has accrued, and to lean on the instruments, which shift latency for reasons that cannot be a downstream consequence of the phase's own overrun.' This defends against trouble that EMERGES after phase entry, but Pearl's sharper variant is trouble FORESEEN at phase entry (immature TRL, contested baseline) which the candidate's own control vector already flags as relevant (TRL at commitment is a named technical control). The candidate's program fixed effects are explicitly program-level and time-invariant; they cannot absorb a phase-varying anticipated-difficulty signal, which is the residual confounder Pearl names. The dissertation pre-commits the IV layer as the backstop precisely because latency is 'plausibly endogenous to program difficulty,' but it produces no executed test of temporal precedence: nowhere does it exhibit a dated sequence, for any program-phase, establishing that the latency-trigger event precedes the first risk-elevation / review-board finding / requirements-change request. Grounded conclusion: the defense is the temporal-precedence assumption Pearl describes, it is asserted not demonstrated, and the phase-varying anticipated-difficulty channel is not closed by the program FE the candidate relies on.
    - PHD-08 dissertation.md Sec 5.5.1 internal validity (candidate's own defense) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - PHD-08 dissertation.md Sec 4.5 controls, Sec 1.5 / 5.3 identification | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
- **[identification]** Answerable at the criterion level from the candidate's own record plus the causal-inference literature, and it resolves AGAINST the candidate on availability while CONCEDING the route was never drawn. The dissertation states the mediated chain verbatim: 'each pending action accrues elapsed time, the authorization latency, during which standing program costs continue to be incurred and during which the technical baseline drifts as requirements are revisited,' i.e. latency -> {standing-cost accrual, requirements drift} -> cost growth, plus latency -> schedule slip -> cost via the Majerowicz-Shinn standing-cost channel (ref [5]). Pearl's front-door criterion requires three conditions: (i) the mediator set intercepts all directed paths from treatment to outcome; (ii) there is no unblocked back-door from treatment to the mediator; (iii) every back-door from the mediator to the outcome is blocked by the treatment. The candidate's own design simultaneously asserts that program/era fixed effects 'remove the time-invariant program difficulty' AND concedes latency is 'plausibly endogenous to program difficulty.' But standing-cost accrual, requirements churn, and workforce idle time are themselves driven by technical difficulty (a hard program carries a larger standing workforce and churns requirements more), so difficulty arrows DIRECTLY into the mediator. That violates front-door condition (iii): difficulty is an unblocked common cause of mediator and outcome that conditioning on latency does not close. Grounded conclusion: with difficulty arrowing into the mediator, the front-door criterion is NOT satisfied, so the front-door route does NOT identify the effect without the IV; combined with the back-door being blocked only by the contested IV, BOTH doors are closed on the current graph. The candidate never drew this mediated DAG or attempted front-door identification, which is the omission Pearl charges.
    - PHD-08 dissertation.md Ch3 mechanism statement & Sec 5/Ch8 (candidate's own record) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
    - Bellemare, Bloem & Wexler, 'The Paper of How: Estimating Treatment Effects Using the Front-Door Criterion,' Oxford Bulletin of Economics and Statistics (2024) | https://doi.org/10.1111/obes.12598 | grade B
    - Pearl, 'The Causal Foundations of Structural Equation Modeling' (2012) | https://doi.org/10.21236/ada557445 | grade B
- **[measurement]** Pearl's structural reading is correct and the candidate's own record CONCEDES the non-classical ingredient while supplying no error-structure analysis. The dissertation builds latency at two resolutions BECAUSE documentary density is era-correlated: it states the early decades are 'sparser and coarser... where baselines were defined less formally,' and that 'documentary detail differs across the agency's history.' That is precisely a measurement error whose magnitude is a function of the era node and is co-determined with looser baselines (hence larger measured overruns), which is the non-classical, outcome-correlated structure Pearl describes, NOT mean-zero classical noise. The candidate treats the two resolutions only as a robustness/sensitivity check ('every headline estimate is re-estimated at both... a result that holds only at one resolution is reported as resolution-dependent'), never as an errors-in-variables problem, and never states whether the coarse-era error is classical or non-classical. Pearl's further point holds on the candidate's graph: because the coarse measurement error and the outcome are both functions of era documentary density / baseline looseness, conditioning on era FE is conditioning on a node that sits downstream of both the mismeasured latency and the outcome, which is collider-adjacent and does not purge non-classical error; the candidate's own warrant for era FE ('rule regimes are real features... not noise') defends era as a confounder control but does not address its collider role w.r.t. measurement error. Grounded conclusion: the candidate's record establishes the error is era-correlated (non-classical), and the era FE defense does not neutralize it; the charge stands on the candidate's own admissions.
    - PHD-08 dissertation.md Ch3/Ch4 data construction & Sec 5.9 robustness battery (candidate's own record) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
    - Pearl, 'On the Validity of Covariate Adjustment for Estimating Causal Effects' (2012) | https://doi.org/10.48550/arxiv.1203.3515 | grade B
- **[empirics]** The methodological premise is grounded and the candidate's record shows the specific graph-falsifying test is ABSENT from the pre-registered battery. Pearl: a DAG's exclusion/exogeneity assumptions carry testable conditional-independence (d-separation) implications, and the exclusion restriction 'instrument affects outcome only through treatment' implies a vanishing partial correlation between instrument and outcome residual given treatment and controls (an overidentification-style independence test when instruments outnumber the endogenous regressor). The candidate's pre-registration (Sec 5.9) and validity battery enumerate: two latency resolutions, multiple cadence definitions, alternative fixed-effects structures, heterogeneity-robust estimators, Goodman-Bacon decomposition, few-clusters/wild-bootstrap inference, and the IV with 'first-stage strength' plus 'imperfect-instrument bounds.' NONE of these is the conditional-independence / overidentification test of the exclusion restriction Pearl names; the IV defense rests on first-stage F (relevance, not exclusion) and on imperfect-IV sensitivity BOUNDS (which weaken under assumed violation, they do not test-and-falsify the graph). With two instruments (authorizing-office workload, appropriations-calendar proximity) for one endogenous regressor the design is overidentified and the vanishing-partial-correlation test is computable in principle, yet it is neither pre-registered nor committed to as graph-falsifying. Grounded conclusion: the testable implication Pearl asks for is real and available, the candidate's pre-registered battery omits it, and on the current design the causal (vs associational) claim rests on an exclusion restriction that is bounded-against but not falsification-tested.
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
    - Hall of Shoulders pearl dossier (review lens), brain hos-pearl | file:///D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/pearl/dossier.md | grade C
    - PHD-08 dissertation.md Sec 5.9 pre-registration & validity battery table (candidate's own record) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/PHD-08/dissertation.md | grade C
- **[measurement]** The challenge is well-founded and the candidate must measure alpha before estimation, not after. Megaproject and infrastructure cost overruns are empirically fat-tailed and dominated by a handful of catastrophic realizations (Flyvbjerg's 'Big is Fragile' and the cost-overrun synthesis), which is the same Extremistan regime in which Taleb shows the sample mean and variance are dominated by a few extreme observations, the historical record undersamples the tail, and Gaussian point estimates and standard-error machinery understate true exposure. Tail-index estimation via the Hill estimator on upper order statistics is the established diagnostic for this regime. If alpha <= 2 the second moment does not exist, so program-clustered Gaussian intervals are computed around a moment that is not defined; if alpha is also at or below 1 the first moment fails and beta targets a non-convergent mean. The correct posture is a pre-estimation tail test, and if alpha is in the infinite-variance range the mean-regression apparatus (point estimate and clustered CIs in 5.4) must be replaced rather than reported.
    - Flyvbjerg, 'Big is Fragile: An Attempt at Theorizing Scale' (2016) | https://doi.org/10.48550/arxiv.1603.01416 | grade B
    - Flyvbjerg, 'Five things you should know about cost overrun', Transportation Research Part A (2018) | https://doi.org/10.1016/j.tra.2018.07.013 | grade A
    - Hall of Shoulders dossier: Nassim Nicholas Taleb (hos-taleb), statistical-engine section | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/hos-taleb | grade C
    - Beirlant et al., 'Tail Index Estimation and an Exponential Regression Model', Extremes (1999) | https://doi.org/10.1023/a:1009975020370 | grade A
- **[empirics]** The leave-extreme-out jackknife is the correct falsification design and should be pre-registered. Under fat tails the aggregate is dominated by a few realizations, so the policy-relevant object is the tail program, not the average program; a coefficient that is stable only because the tail programs are retained in a mean estimator is reporting the influence of those few points, not a population effect. This is the empirical analogue of Taleb's track-record critique: confidence resting on an observed sample is fragile precisely when a few deleted observations move the estimate, because the sample undersamples the tail. The candidate should commit ex ante to a threshold (the dossier's natural cut is movement exceeding the coefficient's own standard error, or any sign flip / loss of significance, upon deleting the top-k extreme programs) at which the mean frame is conceded to have failed and the estimand is re-posed on the tail. Refusing to pre-register converts a falsification test into a post-hoc robustness narrative.
    - Hall of Shoulders dossier: Nassim Nicholas Taleb (hos-taleb), 'Fat tails and the track record' review-lens item | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/hos-taleb | grade C
    - Flyvbjerg, 'Big is Fragile: An Attempt at Theorizing Scale' (2016) | https://doi.org/10.48550/arxiv.1603.01416 | grade B
- **[identification]** The conditional-mean estimand is mis-targeted for a ruin-bearing fat-tailed outcome and must be re-specified on the tail. The decisive distinction in Taleb's frame is between repeated risks a system survives and absorbing risks that end the game; where the outcome is ruin (a Kessler cascade is an effectively irreversible absorbing barrier on operational timescales), the relevant quantity is the upper-tail probability of an extreme realization, not the conditional mean. Quantile regression directly estimates covariate effects on conditional quantiles and is the standard tool for asking whether a regressor moves the upper tail differently from the center, so latency's effect at the 90th/95th percentile of cost growth is identifiable on the candidate's own panel and can diverge in sign from its mean effect. Therefore a null mean effect with a positive tail effect is NOT a confirmation of H0: under a ruin-bearing process the tail coefficient is the only one a program manager should act on, the pre-registered mean-based 5.2 decision rule can mechanically retain H0 while the policy-relevant tail relationship is alive, and the decision rule must be re-anchored on the tail estimand or it will systematically miss the failure mode it claims to address.
    - Hall of Shoulders dossier: Nassim Nicholas Taleb (hos-taleb), 'Ruin, ergodicity, and the precautionary principle' review-lens item | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/hos-taleb | grade C
    - Lewis, 'Understanding long-term orbital debris population dynamics', J. Space Safety Engineering (2020) | https://doi.org/10.1016/j.jsse.2020.06.006 | grade A
    - Koenker & Hallock, 'Quantile Regression', Journal of Economic Perspectives (2001) | https://doi.org/10.1257/jep.15.4.143 | grade A
- **[rival]** The policy claim 'compress latency to reduce overrun' is, as specified, a within-survivor association that is silent on the worst latency outcome. Cancellation is an absorbing barrier: program death ends the game and is the realization where latency's cost is effectively infinite, a ruin outcome categorically distinct from a survivable cost or schedule increment. Because all three dependent variables (cost growth, schedule slip, cadence) are defined only against a baseline that ruined programs never set, the ruin outcome is censored out of every regression, and the historical survivor record undersamples exactly the extreme realizations that dominate the process. A coefficient fit on survivors therefore measures the cost/schedule response among programs that lived; it does not, and cannot, speak to latency's effect on program-killing ruin, and may run opposite to it. The contribution must accordingly be restated with an explicit scope restriction to surviving programs, or re-specified with a ruin-inclusive outcome (e.g., cancellation hazard), before any agency-history-spanning or causal policy claim is licensed.
    - Taleb dossier (Hall of Shoulders), ruin/ergodicity and tail-undersampling lens: 'in Extremistan the variance and the mean of a process are dominated by a few extreme realizations, so the historical record undersamples the tail, sample means are unreliable'; absorbing-barrier framing 'the decisive distinction is between repeated risks you can survive and absorbing risks that end the game ... when a system faces ruin (an absorbing barrier ...)' | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/taleb/ | grade C
    - Taleb, N.N., et al., 'The Precautionary Principle (with Application to the Genetic Modification of Organisms)' (ruin vs. repeated-risk distinction, absence-of-evidence under absorbing-barrier exposure) | https://doi.org/10.48550/arxiv.1410.5787 | grade A

## Gaps

- **[measurement]** No retrieved source (AMOS, ACTA, Space Economy, or the Allison brain) reports an empirical NASA-budget/NTRS/GAO study that hand-codes authorization intervals into Model I/II/III process types and re-estimates latency with process-type interaction terms. The decision-theory warrant for the decomposition exists; the verdict on whether a single latency coefficient survives disaggregation is unsupported and is asserted nowhere. (raised by allison)
- **[identification]** No retrieved source provides the candidate's panel cases, a seat-by-seat documented exclusion argument for the workload and appropriations-calendar instruments, or an overidentification/placebo test (e.g., no-year-funded programs) distinguishing a true instrument from a mechanism channel. Whether the exclusion restriction holds is unresolved by this retrieval. (raised by allison)
- **[rival]** No retrieved source constructs an independent per-program-phase proxy for bureaucratic-bargaining intensity (reprogramming-request counts, competing-center-stakeholder counts, congressional-mark reversals) or demonstrates the latency coefficient is robust to it. Whether latency and overrun are causally linked or are co-produced resultants of one coalition fight cannot be adjudicated from the retrieved evidence. (raised by allison)
- **[rival]** No retrieved source (AMOS, ACTA, Space Economy, Allison brain, OpenAlex/Crossref gap-fill) contains a NASA authorization-interval dataset partitioned by owning seat, nor any within-seat vs between-seat decomposition of the latency-to-cost/schedule beta. The empirical claim that long latency concentrates in one seat (appropriations) and short latency in another (center sign-off), making beta a composition artifact, CANNOT be confirmed or refuted from retrieval and is asserted as a gap. (raised by allison)
- **[identification]** No retrieved source provides a documentary measure of NASA program standing (champion / district interest / flagship) joined to authorizing-office workload, nor any test of whether workload predicts cost/schedule differently for winners vs losers. Whether the workload->outcome path runs through deprioritized losers (violating exclusion) is unresolved by retrieval. The OpenAlex gap-fill surfaced adjacent but non-settling work (NBER w24201 'Bureaucratic Competence and Procurement Outcomes') that does not address NASA workload-as-instrument; treated as a gap. (raised by allison)
- **[governance]** No retrieved source contains a coded inventory of discrete NASA re-seating reforms (clearance eliminated, authority delegated, review office stood up/disbanded) nor an event-study showing whether latency and cost/schedule outcomes co-move when the seat changes but the program is held fixed. Whether latency is a causal lever or a passive regime marker is unresolved by retrieval; asserted as a gap. (raised by allison)
- **[identification]** Whether PHD-08's ATT(latency) curve is actually flat across the dose support, or whether its slope is driven by which programs select into high latency, is an empirical claim about the candidate's own data. No retrieved source contains the candidate's ATT(latency) estimates across dose, so the empirical half of Q1 (is the curve flat vs selection-driven?) cannot be settled from retrieval. The literature fixes the required assumption and the correct fallback estimand; it cannot certify the candidate's data satisfy it. (raised by callaway_santanna)
- **[identification]** Whether PHD-08's pre-period placebo ATT(g,t) on cost growth and schedule slip are zero or non-zero for cohorts entering each regime, and the documentary distribution of the foreseeability-to-realization gap, are empirical results about the candidate's data and documentary record. No retrieved source contains these estimates, so whether no-anticipation actually holds (and the design is salvageable) cannot be decided from retrieval. The literature settles HOW to test it and how to bound the inference; it cannot report the candidate's test result. (raised by callaway_santanna)
- **[empirics]** Whether PHD-08's instrumented (2SLS) latency coefficient is a convex-weighted average or places negative weight on specific program-time cells (e.g. late-authorizing offices used as implicit controls for early ones), and whether the exclusion restriction survives dynamic portfolio-difficulty spillovers, are claims about the candidate's specific instruments, data, and identifying argument. No retrieved source contains a Goodman-Bacon-style decomposition of the candidate's INSTRUMENTED estimand or evidence on the instruments' exogeneity. Retrieval settles the general 2SLS-under-heterogeneity weighting result and the design standard; it cannot certify the candidate's specific decomposition or exclusion restriction. (raised by callaway_santanna)
- **[empirics]** No realized FE-IV headline beta exists, so the implied-weight (Goodman-Bacon / effective-sample) decomposition -- each program's and each era's share of identifying variation, ranked, and the flagship-vs-science-mission split -- cannot be produced or refuted from the current artifact. Settleable only after the panel is assembled and the coefficient fitted; until then the design names the diagnostic but delivers no weights. (raised by callaway_santanna)
- **[rival]** Era-by-era ATT(latency) cannot be reported: the era-regime partition (Table C.1) and coverage bands (Table B.1) are defined but unpopulated, and no pooled or era-specific estimate has been computed. Whether the latency-cost association survives within the modern regime, or is carried by extinct Apollo/Cold-War and Shuttle cohorts, is unsettleable from the design-stage document. (raised by callaway_santanna)
- **[identification]** Transportability of the aggregated estimand to the in-flight/future target population cannot be demonstrated: cancelled/right-censored cohorts (Constellation, SLS upper stage, second mobile launcher) are documented as a named survivorship/attrition threat and partially bounded by planned safeguards, but matching the estimand's implicit weights to managerial relevance requires a fitted estimand whose weights are computable, which does not yet exist. (raised by callaway_santanna)
- **[identification]** Retrieval returned no evidence that PHD-08's panel was tested for cross-lagged Granger precedence between phase latency and within-program cost/schedule trouble, nor any result on whether that cross-lag is asymmetric. Whether the reported FE+IV beta is a one-way cause or a mislabeled loop coefficient cannot be settled from any source retrieved this turn. UNANSWERED: the candidate must run the within-program cross-lag on their own data; the empirical direction is not in the corpus. (raised by forrester)
- **[governance]** No retrieved source establishes, for PHD-08's panel, whether post-regime (reorganization / procurement-reform) cost-growth and schedule-slip trajectories actually bent in the direction the static beta predicts, or whether latency was compensated elsewhere so total elapsed program time did not fall. The policy-resistance test (an event study around the regime switches) was not found to have been run. UNANSWERED on the data; only the mechanism and the requirement are grounded. (raised by forrester)
- **[measurement]** Retrieval returned no measurement, for PHD-08's panel, of the latency-to-outcome delay relative to the phase window, and no impulse-response / distributed-lag estimate showing the cost and schedule consequences of a phase's latency land inside that phase versus being carried into later phases of the same program. Whether the phase-level unit commits a stock-flow timing error cannot be settled from the corpus. UNANSWERED on the data. (raised by forrester)
- **[mechanism]** Retrieval supports the conceptual critique (the loop is real and an open-path FE model cannot represent it) but contains no KDP-dated within-program authorization-event count series for this candidate's panel, and nothing establishing whether that count rises endogenously with accumulated cost-growth or whether beta is stable once conditioned on the loop state. The empirical endogeneity test the question demands cannot be settled from sources retrieved this turn. (raised by forrester)
- **[measurement]** Retrieval grounds the bathtub/integration-artifact risk (Sterman; Forrester stock-flow decomposition) but contains no dated within-phase cost/schedule re-baseline accrual series for this candidate's programs. Whether latency predicts the instantaneous accrual rate versus only the end-of-phase integrated stock cannot be determined from sources retrieved this turn. (raised by forrester)
- **[empirics]** Retrieval grounds the aggregation-discipline principle and the documented congestion/coordination-burden scaling, but contains no authorizing-office-by-year aggregate panel for this candidate. Whether agency latency empirically exhibits queue-congestion dynamics and whether summed program-level betas reproduce the observed cadence series (vs a fallacy of composition) cannot be determined from sources retrieved this turn. (raised by forrester)
- **[empirics]** DEMOTED: evidence rested entirely on internal pointers (file:// dissertation self-ref and local: Space-Economy corpus). No resolvable external citation found for the OCEA GDP-deflator claim. The dual-deflator / NASA New Start Index question is recorded in open_questions.jsonl (status: partial) pending location of a verifiable external DOI or URL for the OCEA source (Bryce/SIA Forecasting the Space Economy, key SD08). (raised by maddison)
- **[measurement]** UNANSWERABLE FROM RETRIEVAL THIS TURN. Overlap-window splice validation of the two latency resolutions (Pearson correlation of coarse vs fine on shared programs; coarse-minus-fine level gap in months; constancy-vs-drift of that gap) does not exist in the design-stage dissertation and cannot be sourced; it must be produced by executing the panel on the GAO/NTRS overlap programs. No number can be asserted without fabrication. (raised by maddison)
- **[empirics]** UNANSWERABLE FROM RETRIEVAL THIS TURN. The dual-deflator sensitivity (New Start Index vs economy-wide GDP deflator) and the resulting movement in beta cannot be reported: the dissertation estimates no beta and concedes the New Start Index is not yet a documented, versioned corpus source. Reporting a coefficient or a delta would be confabulation. (raised by maddison)
- **[measurement]** UNANSWERABLE FROM RETRIEVAL THIS TURN. The original-vs-reset baseline distributions of latency and cost-growth for the re-baseline subset, and the by-era rule for which baseline the panel uses, are not in the dissertation and cannot be retrieved; they require assembling the re-baseline subset from GAO/NASA records and tabulating it, which the design-stage panel has not done. (raised by maddison)
- **[measurement]** Q1 EMPIRICAL CORE REFUSED. No source retrieved this turn provides a divergence distribution between reconstructed commitment-crystallization dates and recorded authorization dates for NASA program-phases, nor any number of months by which they diverge, nor evidence on whether measured latency tracks the real-commitment-to-paperwork gap. The candidate must either (a) execute the documentary reconstruction on a stratified program-phase sample and report the divergence distribution as a construct-validation study, or (b) concede that, absent that study, the identifying assumption authorization-event = realized-decision is asserted, not validated. The emergent/deliberate distinction (10.1002/smj.4250060306) supplies the threat but not the measurement. (raised by mintzberg)
- **[mechanism]** Q2 EMPIRICAL CORE REFUSED. The managerial-work-realism threat is grounded (10.1287/mnsc.18.2.b97), but no retrieved source supplies an independent, process-derived measure of how long substantive deliberation took on NASA program-phases against which documentary latency could be validated. Without that external criterion the candidate cannot show convergent validity, so the claim that latency measures decision speed rather than records-keeping density is untested. Required: a process-evidence subsample (oral histories / NASA institutional histories / Apollo management-control record) coded for deliberation duration and correlated against the documentary latency measure. (raised by mintzberg)
- **[identification]** Q3 FULLY REFUSED. No source retrieved this turn provides a per-era estimate of the fraction of substantive decision time captured by the documentary trigger-to-authorization interval, nor any evidence that this capture fraction is stable across eras. The identification threat, that era fixed effects cannot neutralize an era-dependent capture fraction because the variable then means something different in 1965 than in 2020, is logically coherent and consistent with the documented era-dependence of NASA's review formalization, but the candidate has no retrieved empirical basis to estimate the capture fraction per era or to demonstrate its stability. Until a capture-fraction estimate exists, the within-era latency coefficient cannot be shown to compare like with like. No assertion is made on this question. (raised by mintzberg)
- **[measurement]** No retrieved source supplies the validation Q2 demands: a hand-coded sample where oral histories or contemporaneous correspondence are placed alongside the formal record to show the documentary trigger/authorization timestamps are NOT late ratifying bookends and that the ratification lag is uncorrelated with phase difficulty. The managerial-work-realism literature establishes the threat but cannot, by itself, certify the candidate's clock survives it. The construct-validity rescue is unproven on the evidence in hand. (raised by mintzberg)
- **[identification]** No retrieved source supplies the affirmative falsification Q3 requires: on documented latency-compressing reorganizations, post-regime cost-and-schedule improvement that EXCEEDS the mechanical removal of the compressed waiting time, the only test that rules out the latency remedy being a downstream symptom of configuration mismatch rather than the cause. The theory frames the test; the empirical result that would pass or fail it was not retrieved this turn. (raised by mintzberg)
- **[identification]** No retrieved source contains the candidate's panel or any regression of residual (within-program-within-era) latency on a coded register of rule-text changes, nor any R-squared establishing that residual latency tracks documented rule changes rather than idiosyncratic case-handling noise. The empirical claim the question demands is absent from retrieval and is refused under the no-confabulation contract. (raised by north)
- **[measurement]** No retrieved source contains the candidate's documentary latency series or any correlation of it against an independent de facto decision-timing measure (internal correspondence timestamps, oral-history dates, working-group minutes) on a program subset. The validation result the question demands is absent from retrieval and is refused; the grounded expert cannot certify whether the measure captures the transaction cost of deciding or only the cadence of paperwork. (raised by north)
- **[rival]** No retrieved source contains the candidate's panel or any empirical test of the exclusion restriction (zero partial association of authorizing-office workload with cost growth, schedule slip, or cadence conditional on latency, ruling out a direct congestion or rationing channel). The empirical demonstration the question demands is absent from retrieval and is refused; the grounded expert cannot certify the instruments separate the latency mechanism from the difficulty rival. (raised by north)
- **[identification]** No retrieved source provides a cross-institutional replication or anchor of the candidate's NASA latency-to-cost/schedule coefficient against DoD SAR programs, ESA, or commercial fixed-price firm-commitment contracts. AMOS, ACTA, and Space-Economy brains returned zero hits on acquisition cost-overrun/latency; the nearest out-of-NASA empirical item retrieved (Explaining the cost of European space and military projects, 1999, doi:10.1145/302405.302645) is a cost-driver study, not a replication of a latency coefficient. The cross-regime sign/magnitude survival cannot be asserted from retrieval; refused. (raised by north)
- **[empirics]** No retrieved source partitions a within-NASA latency coefficient by documented internal rule-regime (formal reorganizations, procurement-reform statutes) or demonstrates beta stability vs. regime-dependence across NASA's appropriations/review/reserve eras. The thinker brain supplies only the path-dependence rationale (rules drift, pooled estimates average over distinct rules-sets) but no per-regime estimate exists in retrieval; the empirical stability claim is refused. (raised by north)
- **[identification]** The decisive empirical object Pearl demands does not exist in any retrieved corpus: an executed test, from the NASA appropriations time-series, of whether within-era residual variation in the appropriations-calendar instrument is statistically independent of the funding-instability index. The dissertation is design-stage and pre-registers this as a probe but reports no estimate; AMOS/ACTA/Space-Economy returned zero hits on NASA CR-vs-cost-growth; OpenAlex/NTRS/Crossref gap-fill surfaced general NASA cost-growth and defense-CR literature but no instrument/funding-instability co-movement test for this dataset. Whether the exclusion restriction survives the fiscal-regime back-door is therefore unsettled by retrieval and cannot be asserted either way. (raised by pearl)
- **[measurement]** The settling artifact Pearl asks for is absent from retrieval: an explicit set-theoretic decomposition, from the NASA/NTRS milestone-interval taxonomy, of a phase's elapsed time into authorization intervals versus engineering-execution intervals with a proof of disjointness (i.e., that no authorization month is double-counted in the schedule-slip numerator). The dissertation supplies a measurement taxonomy and concedes latency is a 'component of elapsed schedule' but never proves disjointness; NTRS queries returned no milestone-interval taxonomy document, and AMOS/ACTA returned nothing. Without that decomposition it cannot be settled from retrieval whether the schedule-slip coefficient is an identity or a genuine estimate, so the charge stands unrefuted but not corpus-proven. (raised by pearl)
- **[identification]** The empirical fact that decides the arrow direction is not in any retrieved source: a dated event sequence, from GAO and NTRS program records, for a sample of program-phases, showing whether the latency-clock-start (trigger) event precedes or follows the first documented risk-elevation, review-board finding, or requirements-change request in that phase. The dissertation asserts early measurement as the defense but exhibits no such dated trail; NTRS queries for milestone/risk/baseline sequences returned nothing usable, AMOS/ACTA returned zero. Whether risk signals systematically predate the latency-trigger (collapsing the defense, difficulty -> latency) or follow it (preserving latency -> cost) is therefore unsettled by retrieval and cannot be asserted. (raised by pearl)
- **[identification]** The constructive half of the question is unsettled by retrieval: the candidate's three named datasets (NASA budget/program records, NTRS documentation, GAO major-project assessments) are never operationalized into measured mediator variables (per-phase standing-cost burn rate, count/magnitude of requirements changes, idle-workforce months), and no front-door estimate or mediator-node measurement exists anywhere in the design-stage record. AMOS/ACTA/Space-Economy returned zero hits on front-door/mediation; OpenAlex surfaced the general front-door method (Bellemare 2024) but no NASA-specific mediator construction. So WHICH budget-record line items would populate the mediator node, and whether they are measurable at program-phase resolution across 1958-2026, cannot be asserted from any retrieved source. (raised by pearl)
- **[measurement]** The decisive empirical object Pearl demands does not exist in any retrieved corpus: the joint distribution of (coarse latency, fine latency, outcome) on the documented overlap window where both the milestone-to-milestone and the key-decision-point rule can be computed for the same program-phases. The dissertation is design-stage and reports NO estimates from the assembled panel; it pre-registers re-estimation at both resolutions but never tabulates the within-overlap joint distribution that would reveal whether (coarse minus fine) latency is mean-zero and outcome-independent or co-moves with cost growth. AMOS/ACTA/Space-Economy returned zero hits; no NTRS/OpenAlex source supplies this program-specific overlap tabulation. Therefore whether the coarse-era error is innocuous attenuation or outcome-correlated bias cannot be asserted from retrieval; only that the candidate's record makes the non-classical structure plausible and leaves it unmeasured. (raised by pearl)
- **[empirics]** The forward commitment itself is not settleable by retrieval: whether the candidate WILL pre-register the instrument-vs-outcome-residual conditional-independence test and accept it as graph-falsifying is a design decision the candidate has not made (the design-stage record neither includes nor refuses it), and no corpus source can supply that commitment or its result. Moreover the test's outcome on the assembled panel does not yet exist (the dissertation reports no estimates), so whether the exclusion restriction actually survives a vanishing-partial-correlation test is unknown. AMOS/ACTA/Space-Economy returned zero hits; pearl brain and OpenAlex supply the method but not this candidate's executed test. Therefore only the methodological point (the test exists, is available under overidentification, and is currently absent) is grounded; the commitment and the verdict are gaps. (raised by pearl)
- **[measurement]** No retrieved source this turn enumerates the censored 'graveyard' population of NASA programs that died during latency or never reached authorization 1958-2026, nor supplies the missing-fraction denominator. AMOS, ACTA, and Space-Economy corpora returned zero hits on program cancellation/survivorship; NTRS and OpenAlex gap-fill returned nothing usable. The count and the missing-fraction cannot be asserted without confabulation; the survivorship principle is grounded (Taleb dossier: historical record undersamples the tail) but the NASA-specific census is absent. (raised by taleb)
- **[identification]** No retrieved source supplies a Heckman selection model or Manski bounds estimate for this specific survivor panel. The collider/selection logic is grounded conceptually (survival downstream of latency conditions on a collider; tail undersampling biases sample means), but the requested sign-and-magnitude of the survivorship bias on the latency coefficient, and the width of the bound once dead programs re-enter, requires the enumerated cancellation set from g1, which was not retrieved. The numeric bound cannot be produced without confabulation. (raised by taleb)