# Interrogation mind-map: JPL_INSTRUMENTS_NAV_05

Nodes: 122 | questions: 46 | grounded claims: 42 | gaps: 34

## Questions

- **[identification]** Concentration on the carrying windows is computed on a realized log that already encodes requesters' hedging (strategic padding, early submission). Does the top-decile loss share survive when re-derived from a two-adapting-wills model of schedulers and requesters, and is it stable across epochs in which the priority discipline or padding behavior demonstrably changed, or does the concentration move with the discipline (proving the windows are an artifact of the current game, not a structural feature)? (raised by beaufre)
- **[empirics]** The load-response curve is offered as the externally valid object, but it is fitted on today's mission-phase mix. What in the design lets a planner separate, within the carrying windows, those that are geometry-durable (declination, viewing overlap, fixed-longitude complex placement - set by celestial mechanics) from those contingent on a mission-phase mix that will not exist in five years, so he knows which localized windows to harden and which will dissolve when the configuration changes? (raised by beaufre)
- **[mechanism]** The remedy menu's scheduling-discipline lever protects the deferred subclass, but that changes the payoff to padding and early submission for every other mission. Quantify, even ordinally, the second-order recoil: when re-run on logs from a regime that protected a deferred class, does the relief persist, or does the relieved window reappear elsewhere because requesters re-optimize padding against the new discipline, leaving aggregate lost downlink unchanged? (raised by beaufre)
- **[mechanism]** Can multi-cycle DSN scheduling logs exhibit cross-cycle migration of the top-decile carrying windows, demonstrating that the measured concentration is a stable structural invariant rather than a moving equilibrium that a single-epoch fixed-effects design freezes by construction (the iterated two-wills test)? (raised by beaufre)
- **[identification]** State the log observable that would falsify the classification of DSN contention as indirect attrition (deferral, padding, priority discipline): what logged pattern would instead show a direct seizure of a decisive window, and if none can distinguish the two modes the mechanism is description not falsifiable claim. (raised by beaufre)
- **[governance]** For each of the three remedies (added aperture, scheduling-discipline change, demand shaping), quantify even ordinally from the logs how much it expands served missions' freedom of action versus how much it transfers maneuver room to the most padding-aggressive missions who re-optimize against the new rule; if a discipline change hands the heaviest padders a larger share of relieved windows, on what evidence is it a recovery of science return rather than a redistribution? (raised by beaufre)
- **[measurement]** Section 4.5 concedes bit-loss is only a proxy for lost science return, yet the sole remedy is re-weighting bits by mission-declared criticality drawn from the same service-agreement records that generated the bits, a re-scaling that never leaves the measurement frame in doubt. Which qualitative data source (mission scientists' or operators' retrospective accounts of what a specific lost window cost scientifically) will you collect, by what sampling rule (e.g., purposively sampling the top-decile carrying windows the concentration test localizes), to establish that bit-loss tracks 'lost science return' rather than merely asserting the mapping? Candidate post-pass anomaly reports, mission-scientist debriefs, or DSN scheduling-conflict escalation records. (raised by creswell)
- **[measurement]** From the post-estimation set of identified carrying windows, what is the purposive sampling rule for which lost-pass events get a qualitative case (extreme-tail only vs also off-tail confirming/disconfirming), what record constitutes the qualitative datum (anomaly reports, scheduling correspondence, operator phase narratives), and what stopping rule (thematic saturation vs fixed n per stratum) bounds the QUAL strand so a replicator can reconstruct it? (raised by creswell)
- **[empirics]** Is mission-declared criticality a measured quantitative field in the downlink-requirement records or a qualitative judgment that must be coded? Produce the integration as a joint display whose rows are the top-decile carrying windows and whose columns pair (a) bit-volume loss, (b) criticality-weighted loss, and (c) the qualitative theme of why that window's loss mattered. Which records populate column (c), and at how many windows do the volume and meaning rankings diverge? (raised by creswell)
- **[governance]** State the paradigm: pragmatist (question-driven, value-neutral on which mission loses) or transformative (foregrounding which mission classes, small/low-priority/non-NASA-partner, are systematically deferred by the priority discipline modeled)? Which field in the logs records the requesting mission's priority class or institutional standing, and would a strata-by-priority-class breakdown of where the loss concentrates change which intervention the analysis recommends? (raised by creswell)
- **[identification]** Will you re-solve a held-out subset of the realized request stream to (or near) MILP optimality and show the top-decile loss concentration survives, separating a heavy tail intrinsic to the demand-vs-supply geometry from the fingerprint of the incumbent scheduler's suboptimality? (raised by dantzig)
- **[measurement]** Can you produce, from the logs, the dual/shadow price on each binding complex-band-epoch capacity constraint at an optimized allocation, and show the 'carrying windows' are exactly the high-shadow-price (genuinely capacity-binding) windows rather than windows the incumbent discipline merely deprioritized? (raised by dantzig)
- **[empirics]** Can you exhibit, from the logs, a de-padded 'true need' series (e.g. executed-vs-requested ratios, or a window where padding was administratively capped) and show the heavy tail and top-decile concentration persist on de-padded demand, settling whether the tail is a primitive of true science demand or an artifact of strategic, scheduler-induced request padding? (raised by dantzig)
- **[identification]** State precisely which of your three estimands' loss is attributable to binding capacity (Ax<=b binds at the physical limit) versus to suboptimal scheduling (a feasible reallocation would have served the dropped pass), and show from a re-solve on a held-out window that your top-decile carrying windows are still binding at the MILP optimum and not just at the heuristic's output. (raised by dantzig)
- **[mechanism]** Can you produce, from the logs, the gap between the penalty measured on padded requests and the penalty a decomposition with a published shadow-price (congestion charge) on each complex-band-epoch would induce, to show how much of your heavy tail is genuine physical contention versus an artifact of un-priced padding that a coordinating dual would eliminate? (raised by dantzig)
- **[empirics]** Demonstrate, by fitting the convex load-response separately to the deployed-scheduler output and to a re-solved near-optimal allocation on the same demand, that the convexity is a property of the underlying feasible frontier and not merely an artifact of where the current discipline rations; otherwise the fragility reading is indistinguishable from ordinary marginal-cost behavior at a capacity limit. (raised by dantzig)
- **[mechanism]** Section 2.3 names stocks (backlog, unmet-downlink), flows, and delayed feedback, but the 4.2 estimators (Cox hazard, tail fit, cross-sectional logistic/GLM) carry no accumulating state variable. Where in the data is the backlog stock and unmet-downlink stock measured as a time-integrated level, and can you show its realized rise-and-drain trajectory across the multi-year log rather than inferring the loop from a snapshot of pass-loss against instantaneous concurrent load? (raised by forrester)
- **[identification]** H1's reinforcing loop has the operator response (earlier submission, duration padding, priority escalation) as its feedback edge. The design holds mission and epoch fixed and never observes that edge. Can the logs show that a mission's requested-duration padding and submission lead-time actually rise as that mission's recent backlog rises, closing the reinforcing edge, rather than asserting it as narration behind a convex curve that an exogenous bursty arrival process would equally produce? (raised by forrester)
- **[empirics]** The claimed dominant delay (lag between backlog rising and relieving feedback) is asserted to exceed the window. The concentration test is a Gini / top-decile share over discretized geometry-phase cells with no lag structure. What is the dominant request-to-allocation loop delay measured in the data, and does the design distinguish concentration produced by that delay outrunning the window from concentration that is simply fixed celestial-mechanics geometry of overlapping viewing windows, which carries no feedback at all? (raised by forrester)
- **[identification]** If requested duration and concurrent-mission load are a closed-loop hedge (missions watch backlog stock, then pad/submit-earlier/escalate), they are OUTPUTS of the scheduler's deferral controller, not exogenous inputs, and the cross-sectional logistic has reversed the arrow. From the SPS schedule-archive revision history, what lag-1 and lag-2 coefficient and sign, in a mission-fixed-effects panel regressing this cycle's padding on prior-cycle realized pass-loss at the same geometry-phase window, would separate an exogenous geometry-demand collision from an endogenous deferral-hedge loop? (raised by forrester)
- **[rival]** A closed-loop scheduler and an open-loop scarcity story predict the same static top-decile concentration but different dynamics. Take a window the scheduler relieves (added aperture, array combination, logged policy change) and track over subsequent cycles whether requested load DECAYS toward the new capacity (open-loop: scarcity was real and is relieved) or REBOUNDS to saturation as missions re-pad against freed slots (closed-loop: the controller restores the backlog set-point). What rebound fraction and settling time, estimated from pre/post-intervention logs, confirm a loop gain near one (set-point restoration) and show the 'carrying windows' are a moving controller artifact, not a fixed physical property the aperture lever can durably relieve? (raised by forrester)
- **[mechanism]** The convex load-response in 5.3 is offered as a Taleb fragility signature, but a closed-loop deferral controller manufactures convexity for free: as backlog rises the hedge gain rises (more padding, more escalation), so measured lost-downlink accelerates against measured load even if physical service capacity is perfectly linear. Using the executed-versus-requested ratio over time as a proxy for instantaneous hedge gain, can the observed convex load-response be decomposed into (a) a physical service-saturation component and (b) an endogenous hedge-amplification component, and does the convexity survive once the hedge component is differenced out? If the convexity vanishes when the executed/requested ratio is held fixed, the fragility is in the loop, not in the antennas. (raised by forrester)
- **[measurement]** Completeness is measurable, not assumable (Sec 4.3 vs 3.6): from actual DSN service-catalog and SPS schedule-archive records, what fraction of submitted requests, executed passes, and dropped/expired requests is captured versus inferred from aggregate loading reports, and is that capture rate uniform across complexes, bands, and epochs or does it drop precisely in the high-contention windows where H1 lives? Show the coverage census before the Gini. (raised by mcdowell)
- **[measurement]** The pass-loss indicator rests on a 'requested duration' the candidate labels strategically inflated/hedged (Sec 3.6 lim.1) and lost-downlink uses 'assumed nominal data rates' (lim.3). Operationally, against the records actually held, how is genuine unmet demand separated from a hedge pass never intended to fly: is there an independent field (priority class, criticality flag, post-hoc mission acknowledgment) reconciling declared vs real demand, and what is the top-decile concentration after every sub-threshold request is reclassified as non-loss? (raised by mcdowell)
- **[identification]** Catalog discipline requires reconciling competing counts against an independent record, but demand and outcome both come from one arrangement-filtered request stream (Sec 3.2). Name a second independently-produced record of the same passes (mission-ops executed-pass logs, station downtime/maintenance logs, or published NTRS loading tabulations at matching granularity) and show the counted pass-loss events reconcile window-by-window; where they disagree, which record is missing the loss, and does the disagreement cluster on the geometry-phase windows H1 nominates as carriers? (raised by mcdowell)
- **[measurement]** The unit of analysis is 'one tracking-pass request' (Sec 3.3), but the DSN schedule is produced in stages (long-range forecast, mid-range negotiation, near-real-time conflict resolution; Sec 2.1, Johnston et al. 2014): a request in the forecast catalog, the negotiated mid-range schedule, and the real-time log are not the same object (durations re-stated, requests split/merged, provisional entries superseded). State the catalog of record from which 'request' is drawn, define which schedule stage each row comes from, demonstrate the request count is stable rather than double-counting superseded/re-negotiated entries, and say which of three stage durations is the 'requested duration' and the denominator. (raised by mcdowell)
- **[empirics]** Lost-downlink magnitude D is computed from 'requested duration, band, and the mission's nominal data rate' (Sec 3.4, 3.6 lim.3) entirely from the request side. The DSN keeps an independent executed-pass record of what each pass actually radiated (returned-bits, link margin, station monitor data). Will you reconcile computed lost-downlink against that independent executed-pass accounting and report the discrepancy rate, and does the heavy tail and top-decile concentration survive when loss is anchored to measured executed-pass accounting rather than a nominal rate inferred from a hedged request? (raised by mcdowell)
- **[rival]** The design needs a multi-year span across several cycles of high-contention windows (Sec 3.5) and treats the heavy tail as a stable structural property, but the DSN catalog is non-stationary across that span: antennas subtracted/added (70m and 34m availability, arraying), scheduling-floor and logging granularity changed, mission mix inflected. Will you fit the wait-time and loss tails epoch-by-epoch against the documented antenna-complement and logging-convention changes (NTRS loading reports + DSN configuration record) and show the heavy-tail exponent and top-decile concentration are stable WITHIN a fixed-configuration regime rather than a regime-mixture artifact of pooling across configuration breaks and a shifting logging floor? (raised by mcdowell)
- **[identification]** The covariate vector (geometry, band, concurrent load, with mission and epoch fixed effects) holds the rule-in-use, the priority-and-preference allocation discipline, constant by omission. In IAD terms only the biophysical conditions and the action arena are modeled. Can the allocation rule be entered as an explicit covariate, and can the logs show whether two requests identical in geometry, band, and load but facing different priority/deferral rules suffer different pass-loss, separating the rule's contribution from the geometry's? (raised by ostrom)
- **[mechanism]** Ostrom's robust empirical finding is that monitoring and graduated sanctions, not load, determine whether a shared resource is governed or collapses into open-access loss. The DSN mid-range process is such an institution (mutual monitoring of requests, graduated bargaining, critical-event escalation as sanction/priority). The candidate concedes requested time is inflated by strategic padding, a monitoring-and-enforcement failure in Ostrom's vocabulary. Do the logs let one measure padding and escalation directly (requested vs. executed duration, escalation frequency by mission), and can the concentration of lost downlink be tested as tracking the breakdown of bargaining discipline rather than geometry windows? (raised by ostrom)
- **[governance]** The stated payoff, where added aperture or policy change recovers the most science per dollar, treats carrying windows as fixed physical properties relieved by capacity. But a deferral rule is a collective-choice variable: the same geometry window shows a different tail under a different bargaining rule, so part of the concentrated loss is a curable governance artifact, not an aperture shortfall. The panacea critique warns against one universal lever. Can the design decompose localized loss in each high-concentration window into genuine physical oversubscription versus the deferral discipline that resolved the conflict, so NASA is told whether a window needs a new antenna or merely a different rule? (raised by ostrom)
- **[governance]** Partition the DSN logs into the distinct collective-choice stages (forecast / mid-range inter-mission negotiation / near-real-time de-conflict) and show from the SPS schedule-archive revision history whether the heavy-tail concentration is produced at the rules-on-paper stage (published priority discipline) or only emerges at the rules-in-use stage (realized negotiation and de-conflict trades), so the design can tell a fragile rule from a fragile institution. (raised by ostrom)
- **[rival]** Re-estimate the concentration (tail fit, Gini, top-decile share) separately by complex-band-phase stratum (S/X/Ka x complex x mission phase, e.g., X-band cruise vs Ka-band EDL) using service-catalog and band-eligibility records, and demonstrate the heavy tail and top-decile share are a shared structural property rather than a pooling artifact in which one Ka-band stratum's congestion drives an apparently network-wide fat tail. (raised by ostrom)
- **[governance]** Produce the user-community boundary the concentration result actually defends: identify which appropriators sit at the collective-choice table for the carrying windows (who can submit, escalate, and trade), and quantify whether the loss concentrates on appropriators excluded from or weakly represented in the mid-range negotiation, distinguishing a contention problem solvable by aperture from a representation problem solvable only by changing who governs the window. (raised by ostrom)
- **[identification]** The geometry instrument/placebo (Section 4.5) treats target viewing-window geometry as 'set by celestial mechanics' and therefore exogenous to contention. But the same encounter/occultation/conjunction geometry that compresses a window also makes the pass scientifically high-value. State the exclusion restriction explicitly and identify which DSN scheduling-log records and mission downlink-requirement / criticality records would falsify it, i.e. show geometry affecting pass-loss through the mission's own declared pass value rather than through contention. (raised by rubin)
- **[mechanism]** Contention is constructed (Section 3.4) as the count and aggregate requested antenna-time of OTHER missions, so the treatment one mission receives is constituted by other missions' treatments. That is interference by construction (the SUTVA no-interference limb), not a nuisance to be absorbed by mission fixed effects. Write the estimand respecting interference (a partial-interference / exposure-mapping ATT on realized concurrent-load exposure), justify that the exposure mapping is well-defined and free of hidden versions of treatment given that requested time is strategically padded (Section 3.6), and identify which SPS schedule-archive fields would separate hedged padding from genuine need so the exposure mapping is identified. (raised by rubin)
- **[empirics]** The model targets a conditional association (Section 4.1) via logistic regression with mission and epoch fixed effects, but never writes the causal estimand as a potential-outcomes contrast on a fixed tracking-pass-request unit: the policy-relevant contrast is Y_i(low-load) minus Y_i(high-load) on a fixed request i, not Y_i(served) vs Y_i(lost) (served/lost IS the outcome). Define that contrast and then demonstrate overlap/positivity: do comparable requests matched on declination, band, mission phase, and complex exist under both low- and high-load, or do the high-value critical-phase windows (encounter, EDL) never occur under low load, so there are no donor units and any effect there is extrapolation past support? Report covariate cells with no common support before fitting. (raised by rubin)
- **[identification]** Survivorship/selection rival: the concentration on carrying windows could be a property of the request log's entry-and-retention process (off-log triage by pre-coordination, withdrawal, descope, re-time) in exactly the high-geometry-overlap cells, not of physical contention. Produce, from an independent out-of-stream record (off-log negotiation/withdrawal trail, service-agreement amendments, pre-submission coordination logs), the entry/retention propensity for a request to APPEAR as a logged loss as a function of geometry window, and show it is flat across windows. Which dataset settles it? (raised by rubin)
- **[rival]** Omitted-variable limb: per-request anticipated criticality is unobserved and drives BOTH whether the mission fights-for-and-logs the pass AND the geometry/phase window it falls in (encounters/EDL/maneuvers are simultaneously highest-criticality and tightest-geometry by celestial mechanics). Mission and epoch fixed effects cannot absorb it because it varies request-to-request within a mission-epoch. State the request-level criticality covariate, measured independently of the schedule outcome (e.g. mission science-criticality declarations or instrument-cadence records, NOT requested duration which you label strategically inflated), that would enter the propensity/regression to break the geometry-criticality confound, and confirm it can be measured before the outcome is seen, per design discipline. (raised by rubin)
- **[empirics]** Overlap/positivity test of the geometry placebo: the instrument identifies only on the sub-population with common support across treatment arms, i.e. the off-diagonal cells of the geometry-by-criticality table (geometry-tight-but-NOT-high-criticality, and geometry-loose-but-high-criticality), populated under both served and lost outcomes. If tight geometry and high criticality are collinear in deep-space ops (encounters are both), those off-diagonal cells are near-empty, overlap fails, the instrument is not identified, and the placebo cannot rule out the rival. From the actual DSN logs crossed with mission-phase records, produce the geometry-by-criticality cross-tabulation with served/lost counts per cell and show the off-diagonal cells carry enough donor units under both arms. If empty, concede the rival is observationally equivalent. (raised by rubin)
- **[identification]** The conjunction-driven overlaps, critical-event coincidences, and peak-concurrency windows that H1 most needs to characterize recur only a handful of times per orbital/mission cycle, so a multi-year DSN log yields perhaps a few dozen tail exceedances above the peaks-over-threshold cutoff, not thousands. Specify, from the actual DSN log structure, the realized exceedance count N above the selected threshold, and demonstrate by parametric bootstrap that at that N the Clauset-Shalizi-Newman likelihood-ratio test can in fact discriminate a true power-law/GPD-heavy tail from a log-normal or a merely-truncated exponential at the pre-set significance level, rather than returning the indistinguishability small samples produce. If it cannot, the decision rule adjudicates sample size, not H0 vs H1. (raised by taleb)
- **[measurement]** The dependent variable is built only from requests that entered the log, but a deadline-driven queue with abandonment self-censors its worst realizations: a mission that knows a window is hopeless never submits, downgrades its requirement in advance, or has the request silently dropped before it becomes a lost-pass record, so the largest collisions of demand may leave the fewest surviving log entries. Using the schedule logs together with the mission downlink-requirement records and the NTRS loading reports, quantify the gap between requested-and-logged demand and true latent demand during the highest-concurrency windows, and show that the heavy-tail and concentration estimates are not biased toward the thin-tailed null precisely because the worst windows are the ones the log under-records. (raised by taleb)
- **[empirics]** The convex load-response curve is promoted as the fragility signature and licensed by the second-order-response argument, but fragility detection is meant to substitute for forecasting exactly where you cannot extrapolate, whereas the candidate intends to read convexity off observed load and use it to plan for a crewed-exploration concurrency beyond any recorded load. Within observed support a queue can show convex waits then saturate or invert past an instability point the sample never reached. From the DSN logs and the Abraham human-exploration traffic projections, bound where the realized concurrent-load support actually ends, and show by out-of-sample backtest (fit the convex curve on early cycles, predict loss in held-out high-load cycles) that the fitted second-order response holds outside the fitting range, rather than presenting in-sample curvature as if it forecast a regime the data do not contain. (raised by taleb)
- **[rival]** Rival hypothesis: the heavy tail and the convex load-response are an artifact of a handful of high-leverage observations, not a structural property. From the realized scheduling logs, run a leave-one-out / jackknife and a top-k-exceedance deletion on the observations that drive the generalized-Pareto fit; identify the smallest set of pass-loss records whose removal drops the fitted GPD shape parameter to insignificance (or below zero) and collapses the top-decile loss share from the illustrative one-half-to-two-thirds back toward the ten-percent uniformity benchmark. State that count as a fraction of total exceedances, and show the surviving tail still rejects the exponential after their removal. If a single-digit number of windows carries the result, the rival wins by default. (raised by taleb)
- **[measurement]** The dependent variable, lost-downlink magnitude, is computed in Sec 3.4 from an assumed nominal data rate, and Sec 3.6 lim.3 concedes this assumption carries measurement error, yet no error model on that nominal rate is stated. Demonstrate from mission link-budget and telemetry records the actual distribution of the requested-versus-realized data-rate ratio and show it is itself thin-tailed: if the per-pass nominal-rate error is multiplicative and even mildly fat-tailed or drifting across the multi-year span (calibration drift), it manufactures a heavy upper tail in lost-downlink mechanically, with no contention mechanism present. Until the tail of the measurement error is bounded below the tail being claimed, the Extremistan finding is indistinguishable from a fat-tailed-noise finding. (raised by taleb)
- **[empirics]** The Extremistan/fat-tail discipline is invoked as the warrant for the heavier-tailed model, but the Clauset-Shalizi-Newman procedure is preasymptotic-fragile: a log-normal from ordinary congestion, or a finite-support truncated distribution, routinely passes a power-law likelihood-ratio test against the exponential at the sample sizes available above a high threshold. Pre-register, and show by parametric bootstrap on a simulated light-tailed-plus-calibration-drift DSN log that reproduces the arrival burstiness and skewed service, that the decision rule does NOT spuriously accept H1 on data constructed to contain no heavy-tailed contention mechanism. Report the false-positive rate of the pre-registered test against that null. If it accepts the fat tail on noise-only synthetic data, the test cannot adjudicate mechanism versus artifact. (raised by taleb)

## Grounded claims

- **[identification]** GROUNDED-BUT-PARTIAL. The design concedes the premise: it treats the request stream as a hedged strategic artifact (Sec 4.6.1) and names the requester-vs-scheduler hedging feedback as the Forrester loop driving concentration (Sec 7.1.1 lever three; Sec 7.3.2). Crucially, the candidate already admits the cross-sectional design 'tests the consequence the Forrester lens predicts but does not directly observe the loop operating in time,' and names a 'dynamic, longitudinal model of the request-and-allocation process' as future work (Sec 7.3.2). The concentration-stability check actually built into the design is mission fixed effects plus leave-one-mission-out (Sec 7.4 rival one) and multi-cycle coverage to test whether the SAME windows recur across cycles (Sec 7.1 rebuttal; Sec 4.5.1) - NOT a test across epochs of changed scheduling discipline. Beaufre's dialectic is therefore conceded as a real limit, partially insured by the behavior-independent geometry instrument (Sec 4.4.4) but NOT settled: the design cannot currently re-derive concentration from two adapting wills, and has no epoch-of-changed-discipline stability test. This is the falsifiable test the candidate must add, not one the design already passes.
    - JPL_INSTRUMENTS_NAV_05 dissertation, ch4_data_and_measurement.md Sec 4.6.1 and 4.4.4 | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch4_data_and_measurement.md | grade C
    - JPL_INSTRUMENTS_NAV_05 dissertation, ch7_discussion.md Sec 7.3.2, 7.4, 7.1 rebuttal | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch7_discussion.md | grade C
    - Andre Beaufre, An Introduction to Strategy (1965), via hall-of-shoulders beaufre brain dossier + Crossref record | https://doi.org/10.2307/2147679 | grade A
- **[empirics]** GROUNDED-BUT-PARTIAL. The raw materials for the discriminant exist but the discriminant itself is not produced. The design cleanly partitions covariates into behavior-independent, celestial-mechanics-set orbital geometry (target declination, viewing-window overlap, elevation profile, conjunction/occultation compression - ephemeris-derived, 'set by celestial mechanics and not chosen by the mission,' Sec 4.4.4) versus mission-phase-window covariates that depend on the roster and critical-event calendar (Sec 4.4.7). The concentration unit is the cell {geometry x mission phase x complex x band} (Sec 4.4.8, Table 4.1), so each carrying window is already indexed by separable geometry and mix coordinates - the covariates needed to classify a window as geometry-durable vs mix-contingent are present. AND the candidate concedes in the Sec 7.5 rebuttal that if the exploration-era mix changes the SHAPE (not just the position) of the load-response, the extrapolation is invalid, and post-era re-estimation is the named confirmation. What is MISSING is the discriminant as an output: the design does not decompose T10/G by the geometry-coefficient share versus the mission-phase-coefficient share, nor label each carrying window geometry-durable or mix-contingent. The planner cannot, from the current design, be told which windows to harden and which will dissolve - only that a curve exists and may change shape. So the decision basis is, as Beaufre charges, a snapshot whose mix-dependence is acknowledged but not partitioned per window.
    - JPL_INSTRUMENTS_NAV_05 dissertation, ch4_data_and_measurement.md Sec 4.4.4, 4.4.7, 4.4.8 (Table 4.1) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch4_data_and_measurement.md | grade C
    - JPL_INSTRUMENTS_NAV_05 dissertation, ch7_discussion.md Sec 7.5 (Qualifier and Rebuttal) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch7_discussion.md | grade C
    - Abraham, MacNeal, Heckman, 'Traffic Modeling for Deep Space Network in the Human Exploration Era,' SpaceOps 2016 (cited as design ref [17]); corroborated by ACTA DSN/Artemis-I support paper | https://doi.org/10.2514/6.2016-2523 | grade B
- **[empirics]** Corroborating space-domain anchor: NASA DSN support is real, fielded, and mission-mix-dependent, confirmed by a peer-reviewed account of DSN support for the Artemis I crewed-exploration-era mission - grounding both that the mix is shifting and that geometry/complex placement is a fixed structural feature of the network the planner could in principle harden.
    - Harmon et al., 'Pre-launch lessons learned from NASA's deep space network support for the Artemis I mission to the moon,' Acta Astronautica Vol 210 (2023), retrieved via ACTA_Papers brain | https://doi.org/10.1016/j.actaastro.2023.05.016 | grade A
- **[mechanism]** GROUNDED-BUT-PARTIAL, and the recoil is conceded in principle but not quantified. The candidate already names the first-order recoil: 'protecting the deferred subclass redistributes wait onto the previously favored subclass, so the lever must be evaluated against the full distribution of waits, not against the tail alone' (Sec 7.1.1 lever two). The design's evaluation mechanism for any lever is to re-run the estimation on post-intervention logs and check whether the tail flattens, T10 falls, and load-response convexity drops (Sec 7.1.1 synthesis) - exactly the displacement test Beaufre asks for, but only as a post-hoc re-run, not as an ex-ante model. Decisively, the SECOND-order recoil Beaufre specifies - requesters re-optimizing padding against the new discipline so the relieved window reappears elsewhere - is NOT modeled. The candidate's own demand-shaping lever (lever three, Sec 7.1.1) acknowledges the bullwhip/order-stability mechanism by which hedging re-amplifies (citing Croson et al. order-stability [63]), which is the very mechanism that would re-game a protected-class discipline; yet the discipline lever is evaluated without simulating that re-gaming. So the design has the conceptual ingredients (full-distribution evaluation + named hedging feedback + post-intervention re-runnability) but produces NO simulation or re-run on a protected-deferred-class regime, and therefore cannot say whether relief persists or merely displaces. Beaufre's freedom-of-action accounting (the recoil of the indirect instrument must be checked against its intended effect) is unmet at the quantitative level.
    - JPL_INSTRUMENTS_NAV_05 dissertation, ch7_discussion.md Sec 7.1.1 (lever two and synthesis) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch7_discussion.md | grade C
    - Croson, Donohue, Katok, Sterman, 'Order Stability in Supply Chains,' Production and Operations Management (2013), design ref [63] | https://doi.org/10.1111/j.1937-5956.2012.01422.x | grade A
    - Andre Beaufre, An Introduction to Strategy (1965) and Deterrence and Strategy (1965), via hall-of-shoulders beaufre brain dossier + Crossref records | https://doi.org/10.2307/40199592 | grade A
- **[mechanism]** Beaufre's iterated-dialectic frame is real and citable: strategy is the reciprocal adaptation of two thinking wills, so a concentration result that holds for one cycle proves planning against a passive opponent, not strategy against an adapting one. This frame correctly motivates Q1, but it is strategic theory only and supplies no DSN empirical content; the candidate must settle Q1 from the logs, and no retrieved source provides multi-cycle top-decile window-identity tracking or any invariant-versus-moving-equilibrium test for DSN scheduling.
    - Andre Beaufre, An Introduction to Strategy (1965), Hall-of-Shoulders beaufre dossier and primary-text record | https://doi.org/10.2307/2147679 | grade A
    - Johnston et al., Integrated Planning and Scheduling for NASA's Deep Space Network (AIAA 2018-2728); Claps et al., Deep Space Network Scheduling via Mixed-Integer Linear Programming (IEEE Access 2021) | https://doi.org/10.2514/6.2018-2728 | grade B
- **[measurement]** The construct-validity gap the question names is the canonical 'connecting through sampling' (building) integration point in mixed methods: a quantitative result is used to purposively select cases for a qualitative strand whose purpose is to interpret what the numbers mean. Fetters, Curry & Creswell locate integration at the methods level via four approaches, of which connecting (through sampling) and building (one phase shaping the next) are exactly the operations that would let the concentration test's top-rank carrying windows drive a purposive qualitative sample of mission-scientist accounts of consequence. Re-weighting bits by criticality records is a within-strand quantitative transformation and performs none of these integration operations, so on the Creswell/Fetters framework it cannot establish the proxy-to-construct mapping; only a qualitative strand connected at the sampling point can. The corrective is procedural: name the design, justify the mix, and show the integration, rather than supplying a quantitative appendix that never merges.
    - Fetters, Curry & Creswell, 'Achieving Integration in Mixed Methods Designs: Principles and Practices,' Health Services Research (2013) | https://doi.org/10.1111/1475-6773.12117 | grade A
    - Creswell dossier (Hall of Shoulders, hos-creswell brain), synthesizing Creswell & Plano Clark and Fetters et al. | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/creswell/ | grade C
- **[identification]** The design the construct claim requires must be named and placed in the typology, and the choice between explanatory-sequential (QUAN concentration result first, then interview affected mission teams to explain the loss) and exploratory-sequential (interview first to surface dimensions of scientific cost no bit-count captures, then build the criticality-weighting instrument from those dimensions) determines the integration point. Each design is fixed by four decisions (level of interaction, priority, timing, and where/how to mix). The decisive auditing test is the deletion test: if the qualitative strand appears only as illustrative quotes after the statistics, ask what would change in the conclusions if it were deleted; if nothing changes the strand is decorative, and if a conclusion in Section 7 would change then the current single-strand design structurally cannot reach the stated socio-technical 'science-throughput penalty' contribution. The integration would be evidenced by a joint display (a statistics-by-themes table pairing the top-rank carrying windows with mission-team accounts of consequence).
    - Creswell & Plano Clark, 'Designing and Conducting Mixed Methods Research,' 3rd ed., SAGE | https://us.sagepub.com/en-us/nam/designing-and-conducting-mixed-methods-research/book241842 | grade A
    - Guetterman, Fetters & Creswell, 'Integrating Quantitative and Qualitative Results in Health Science Mixed Methods Research Through Joint Displays,' Annals of Family Medicine (2015) | https://doi.org/10.1370/afm.1865 | grade A
- **[rival]** The fit question (whether strands confirm, expand, or contradict) is a mandatory reporting obligation, and a credible mixed-methods design pre-commits to reporting the discordant case rather than smoothing it. The hypothesized inversion (heaviest bit-loss windows scientifically cheap because of redundant recoverable cruise telemetry; low bit-loss windows catastrophic because of one-time non-recurring encounter geometry) is exactly a divergence between the bit-loss concentration measure and the scientific-value concentration, which would invert the aperture-aiming policy conclusion. The evidence that could adjudicate it is not in the scheduling logs at all: it is qualitative criticality narratives plus mission-event recurrence records (whether a window's geometry recurs), gathered as a strand connected by sampling to the concentration test's localized windows and merged in a joint display. The Younas/Inayat/Sundus synthesis-review variant shows the same joint-display logic scales to integrate and adjudicate where strands disagree, which is the auditable mechanism for detecting and reporting the discordance a single-strand design cannot see.
    - Fetters, Curry & Creswell (2013), and Creswell dossier review-lens (Hall of Shoulders, hos-creswell) | https://doi.org/10.1111/1475-6773.12117 | grade A
    - Younas, Inayat & Sundus, 'Joint displays for synthesis of mixed methods/qualitative findings,' Journal of Mixed Methods / methods literature (2021) | https://doi.org/10.1177/2632084320984374 | grade A
- **[measurement]** The methodological standard the question invokes is settled and citable: in an explanatory-sequential design the QUAL strand connects to the QUAN strand AT SAMPLING (the 'connecting' integration point), so the candidate must state a purposive follow-up rule that selects cases from the quantitative results, name the qualitative datum, and report the stopping rule, otherwise the design label is unauditable. Ivankova/Creswell/Stick specify that sequential-explanatory designs require explicit procedural decisions on priority, the point of connecting the two phases, and how qualitative cases are selected to explain the quantitative results; Fetters/Curry/Creswell formalize 'connecting through sampling' as one of four methods-level integration approaches; Fetters et al. require that a replicator be able to reconstruct the mixing logic (sample-mismatch and weak-inference 'legitimation' threats) from the reporting. HOWEVER, the candidate's ACTUAL sampling rule, datum type, and stopping rule for JPL_INSTRUMENTS_NAV_05 are not present in any retrieved source; the standard is grounded, the candidate's specification is not.
    - Ivankova, Creswell & Stick, 'Using Mixed-Methods Sequential Explanatory Design: From Theory to Practice', Field Methods | 10.1177/1525822X05282260 | grade A
    - Fetters, Curry & Creswell, 'Achieving Integration in Mixed Methods Designs: Principles and Practices', Health Services Research | 10.1111/1475-6773.12117 | grade A
    - Fetters et al., 'A Comprehensive Taxonomy of Research Designs ... Achieving Design Naming Conventions in Mixed Methods Research', J. Mixed Methods Research | 10.1177/15586898221131238 | grade A
- **[empirics]** The integration operation the question demands is a named, citable artifact: a JOINT DISPLAY, defined by Guetterman, Fetters & Creswell as a visual table arraying QUAN results (here the volume ranking and criticality-weighted ranking) against QUAL findings (the coded theme per window) to surface inferences neither strand reveals alone, and to make fit/confirmation/expansion/discordance between the rankings visible. A second quantitative scalar (criticality-weighted loss) re-weighting the first is integration WITHIN the quantitative strand, not cross-strand integration; only column (c), drawn from a genuinely qualitative record and analysis, makes Section 4.5 an integration rather than a re-weight. The candidate must therefore state whether criticality is a measured field or a coded judgment, name the records behind column (c), and report the divergence count. HOWEVER, whether mission-declared criticality is a measured field, which records populate column (c), and the number of windows where the rankings diverge are NOT present in any retrieved source.
    - Guetterman, Fetters & Creswell, 'Integrating Quantitative and Qualitative Results in Health Science Mixed Methods Research Through Joint Displays', Annals of Family Medicine | 10.1370/afm.1865 | grade A
    - Fetters, Curry & Creswell, 'Achieving Integration in Mixed Methods Designs: Principles and Practices', Health Services Research | 10.1111/1475-6773.12117 | grade A
- **[governance]** The paradigm distinction the question forces is settled and citable. Pragmatism, the worldview most associated with mixed methods, rejects the positivism/constructivism dichotomy, treats the research question as the driver of method choice, and judges knowledge by workability; a distinct TRANSFORMATIVE worldview foregrounds justice and the priorities of marginalized/low-power groups and can be layered onto any design, including a transformative explanatory-sequential design (Maleku et al. demonstrate exactly this archetype in a cross-cultural resettlement context). Fetters et al.'s taxonomy treats paradigm as an essential, declarable dimension of a design. Because the candidate's own payoff (recovering science-return-per-dollar and choosing whose deferred passes are protected) is explicitly distributional, the methodological literature holds that a study carrying distributional/justice stakes owes an explicit worldview declaration and that the worldview constrains sampling and stance toward the studied population; a strata-by-priority-class breakdown is the transformative-lens move. HOWEVER, the candidate's actual paradigm declaration, whether any log field records requesting-mission priority class or institutional standing, and whether a priority-class stratification would change the recommended intervention are NOT present in any retrieved source.
    - Maleku, Kim, Kagotho & Lim, 'Expanding the Transformative Explanatory Sequential Mixed Methods Design Archetype in a Cross-Cultural Context', J. Mixed Methods Research | 10.1177/1558689820936378 | grade A
    - Fetters et al., 'A Comprehensive Taxonomy of Research Designs ... Achieving Design Naming Conventions in Mixed Methods Research', J. Mixed Methods Research | 10.1177/15586898221131238 | grade A
- **[identification]** Dantzig's review lens makes the demand legitimate and the methodology exists: every primal LP/MILP has a dual whose optimal objective coincides at optimality, and a scheduling result without a certified optimality gap is an assertion, not a benchmark; the DSN antenna-satellite problem is explicitly an oversubscribed MILP (Sabol/Claudet/Johnston line), and SDA sensor-scheduling work (Heimdall) already produces certified optimality gaps proving how far a schedule sits from the true optimum, demonstrating a held-out re-solve-to-bound is feasible. So the counterfactual dantzig asks for is well-posed and constructible. BUT it CANNOT be asserted to succeed or fail here: the dissertation is, by its own repeated statement, a design-stage analysis plan with no empirical results executed and expected magnitudes labeled illustrative, so no re-solve has been run and survival of concentration under re-solve is unestablished in the candidate's record.
    - Dantzig dossier (Hall of Shoulders), citing Dantzig, Linear Programming and Extensions (1963, Ch.6) and the duality/shadow-price review lens | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/dantzig | grade A
    - Claudet, Alimo, Goh, Johnston, Delta-MILP: Deep Space Network Scheduling via Mixed-Integer Linear Programming, IEEE Access (2022) | https://doi.org/10.1109/access.2022.3164213 | grade A
    - Space Domain Awareness Sensor Scheduling with Optimality Certificates, Proc. AMOS Conference 2023 (via Dantzig brain source entry) | https://amostech.com/TechnicalPapers/2023/ | grade B
- **[measurement]** Dantzig's framework supplies exactly the benchmark dantzig is asking for and confirms the candidate's stated benchmark is inadequate: the shadow price on a binding capacity constraint is the marginal value of that resource at the optimum (e.g. the shadow price on antenna capacity tells an operator what a marginal additional ground station is worth), so a high shadow price identifies a physically oversubscribed window, distinguishing it from one that is administratively deprioritized. The candidate, however, measures concentration only against a 'ten-percent uniformity benchmark' and a Gini coefficient, never against a dual-based achievable optimum and never computes shadow prices. The dissertation contains no shadow-price computation. Therefore the benchmark dantzig demands is correct and absent; whether the candidate CAN extract binding-constraint shadow prices from the logs is unestablished because no optimized allocation has been solved.
    - Dantzig dossier (Hall of Shoulders), duality/shadow-price review lens over the Spangelo-style antenna-scheduling MILP | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/dantzig | grade A
    - JPL_INSTRUMENTS_NAV_05 dissertation, hypothesis and design-of-measurement sections (concentration clause) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[empirics]** The candidate concedes the exact endogeneity dantzig names but does not resolve it: the dissertation states that requested antenna time is plausibly endogenous to expected contention, that a request log read naively as demand would import the missions' strategic padding (early submission, duration padding, priority escalation) directly into both the dependent and independent variables and 'manufacture apparent contention out of hedging behavior,' and it grounds this in the demand-access/system-dynamics literature on locally optimizing missions amplifying concentration. It treats careful operationalization with explicit bias handling as a precondition. BUT it exhibits no de-padded true-need series (no executed-vs-requested construction, no padding-capped window) and does not pre-commit to one as the falsification test for padding-vs-demand. So the dissertation acknowledges the threat dantzig raises without producing the empirical instrument that would settle it.
    - JPL_INSTRUMENTS_NAV_05 dissertation, endogeneity / operationalization discussion (Ch.3 and design-of-bias-handling sections) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
    - JPL_INSTRUMENTS_NAV_05 dissertation, framing statement (design-stage, no executed results) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[identification]** Dantzig's lens makes the decomposition demand legitimate and the candidate's own text shows the decomposition cannot be performed as designed. By LP duality the irreducible-vs-avoidable split IS the gap between the optimal objective and the realized objective: the optimal allocation is the only baseline at which a binding constraint (positive shadow price) certifies that loss is capacity-irreducible rather than discipline-induced; a result reported only against a heuristic's realized output, with no duality bound or optimality certificate, is an assertion not a benchmark. The candidate concedes its three estimands are measured against the realized log, not an MILP optimum: estimand two is 'a feature of the realized loss record aggregated to windows' and 'makes no claim about why the loss concentrates, only whether it does' (5.1), and the descriptive estimands are 'properties of the realized log... their validity rests on the completeness and accuracy of that record' (5.3.1). The DSN antenna-scheduling problem is an explicit MILP solvable to optimality (Spangelo-style), so the held-out re-solve dantzig asks for is well-posed and constructible. BUT it CANNOT be asserted to succeed: the work is by its own repeated statement a design-stage plan with no logs analyzed and no re-solve run (5.10, 6.0), so the candidate's design as written cannot distinguish capacity-bound loss from scheduling-suboptimality loss for ANY of the three estimands, and the binding-at-the-MILP-optimum survival of the carrying windows is unestablished in the record.
    - Dantzig dossier (Hall of Shoulders), duality/shadow-price and optimality-certificate review lens over the Spangelo-style DSN antenna-scheduling MILP; Dantzig, Linear Programming and Extensions (Princeton, 1963, Ch.6) | https://doi.org/10.1109/TAES.2023.3326422 | grade A
    - JPL_INSTRUMENTS_NAV_05 dissertation, Sec 5.1 Estimands and Sec 5.3.1 | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
    - JPL_INSTRUMENTS_NAV_05 dissertation, Sec 5.10 and Sec 6.0 | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[mechanism]** The candidate's stated mechanism IS a Dantzig-Wolfe coordination failure, and the candidate concedes the artifact while supplying no instrument to size it. Dantzig-Wolfe decomposition coordinates many independent subproblems coupled only by a few shared linking resources, iterating via pricing signals (duals) until the global optimum; missions that pad durations, submit early, and escalate priority because each is locally optimal under contention are precisely semi-autonomous subproblems coupled by the shared scarce asset (the complex-band-epoch) and pricing it wrong because no dual signal disciplines their requests. The candidate names this exact failure: 'requested time systematically overstates true need' and the request stream is 'a hedged quantity rather than a clean demand signal' (4.x, limitation 1), and warns 'hedging manufactures apparent contention out of strategic asking, so a naive analysis could find concentration where there is only coordinated padding' (4.6). So the heavy tail dantzig worries about could be an artifact of un-priced padding that a coordinating congestion charge (shadow price) would eliminate, exactly the Dantzig-Wolfe reading. BUT the candidate produces no shadow-price/congestion-charge counterfactual and no de-padded penalty series; its only padding defenses are mission fixed effects mu_m (which 'absorb... characteristic strategic-padding behavior', 5.3.2) and a behavior-free geometry-compression lever (5.3.3), neither of which prices the linking constraint or computes the dual a coordinating mechanism would publish. Therefore the gap dantzig asks for (penalty on padded requests minus penalty under a shadow-priced decomposition) is well-posed and conceded as material, but is unquantified in the record and, design-stage, uncomputed.
    - Dantzig dossier (Hall of Shoulders), Dantzig-Wolfe decomposition and shadow-price-as-congestion-charge review lens; Dantzig & Wolfe, 'Decomposition principle for linear programs,' Operations Research 8(1), 1960 | https://doi.org/10.1287/opre.8.1.101 | grade A
    - JPL_INSTRUMENTS_NAV_05 dissertation, Sec 4 (central measurement claim), Sec 4.6 (limitation 1), Sec 5.3.2-5.3.3 | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[empirics]** Dantzig's apparatus confirms the convexity is observationally confounded with ordinary marginal-cost-at-a-binding-frontier, and the candidate's design cannot separate the two. The optimal value of a resource-allocation program is a convex (for a minimization/loss reading) function of the binding capacity, and the shadow price is its slope, rising as the constraint tightens; so a convex lost-downlink-versus-load curve is exactly what a near-optimal allocator also produces as it approaches its binding capacity frontier, independent of any fat-tail mechanism. The candidate reads its convex load-response as Taleb's fragility signature: step five tests 'whether lost downlink rises faster than linearly as load increases,' grounding the elevation of curvature to a primary estimand in the fragility framework where 'a convex response of a system's loss to the dose of a stressor is the formal signature of fragility' and the Jensen-gap characterization makes the second-order response the object (6.2.5), and under a positive result 'the load-response curve would be convex... This convexity is the Taleb fragility signature' (6.6.3). The candidate's only stated robustness for the convexity is dropping the most extreme load decile (6.2.5, 6.6.3), which guards against a few-points artifact but does NOT separate frontier-convexity from discipline-convexity. Because the curve is fit only to the deployed-scheduler's realized output and never to a re-solved near-optimal allocation on the same demand (5.1, 5.3.1 firewall the estimands to the realized log), the fragility reading is, in dantzig's terms, indistinguishable from ordinary increasing marginal cost at a capacity limit. The candidate even concedes the convex load-response 'is the closest the cross-sectional design comes to a causal signature' and is 'consistent with explanations other than feedback-driven piling' (6.6.x), conceding the under-identification dantzig names. The re-solve-and-fit-separately demonstration dantzig demands is well-posed and, design-stage (6.0, no estimation run), unperformed.
    - Dantzig dossier (Hall of Shoulders), LP duality and shadow-price-as-marginal-value-of-a-binding-constraint review lens; Dantzig, Linear Programming and Extensions (Princeton, 1963, Ch.6) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/dantzig | grade A
    - JPL_INSTRUMENTS_NAV_05 dissertation, Sec 6.2.5 (step five, load-response) and Sec 6.6.3 / 6.6.x (expected driver and load-response results, rival-explanation boundary) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[mechanism]** GROUNDED CONCESSION. The design does not measure the backlog or unmet-downlink stock as a time-integrated level and shows no rise-and-drain trajectory. The measurement codebook (Table 4.1, sec 4.4) contains no accumulating state variable: 'concurrent-mission load' (4.4.6) is an instantaneous collision-intensity covariate (count and aggregate requested antenna-time overlapping a window), admitted to the Cox model as a time-VARYING input, not a stock integrated over time. Section 2.3.4 itself concedes 'The design is cross-sectional over the realized logs; it can detect the signature the loop predicts ... but it cannot trace the feedback edges themselves.' This is exactly the Forrester/Sterman bathtub failure: a stock's behavior cannot be read off a snapshot of its flows; the accumulation must be modeled as a level over time, which this design does not do. Correct answer: the design tests levels and a load-response signature, not the stock-and-flow structure it invokes.
    - Forrester, Industrial Dynamics / Dynamic Models of Economic Systems and Industrial Organizations (System Dynamics Review reprint) | https://doi.org/10.1002/sdr.284 | grade A
    - Sterman, Bathtub Dynamics: initial results of a systems thinking inventory (System Dynamics Review) | https://doi.org/10.1002/sdr.198 | grade A
    - Candidate dissertation JPL_INSTRUMENTS_NAV_05, sec 2.3.4 and 4.4.6 + Table 4.1 (measurement codebook) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[identification]** GROUNDED CONCESSION. The reinforcing feedback edge is NOT observed. The design fixes mission and epoch and records padding/lead-time only as a static hedging BIAS on the request fields (sec 4.2.1, 4.4.6), never as a response regressed on that mission's recent backlog. No lagged operator-behavior estimator exists in Ch4-5; padding inflates the requested-time and load covariates but is never measured as rising-with-backlog. The dissertation states plainly the cross-sectional design 'tests the signature of the mechanism (a convex load-response) but does not, by itself, prove the Forrester feedback loop generated it' (2.3.2) and cannot 'rule out that some other generative process produced the same signature' (2.3.4). Forrester's endogeneity test is precisely this: reproduce the trouble from internal feedback with no exogenous shock. The candidate cannot, because an exogenous bursty/geometry-driven arrival process produces the same convex curve. The lagged padding-vs-backlog regression that would close the edge is feasible on the same logs but is not in the design.
    - Forrester thinker dossier (hall_of_shoulders), Endogeneity test + bullwhip-amplification-under-delay review lens; anchored to Forrester, Counterintuitive Behavior of Social Systems | https://doi.org/10.1007/bf00148991 | grade A
    - D'Ambrosio et al., A Dynamical Systems Analysis of the Effects of the Launch Rate Distribution (exogenous inflow shape/burstiness alone drives concentration/runaway with no operator feedback) | https://www.semanticscholar.org/search?q=Dynamical%20Systems%20Analysis%20Launch%20Rate%20Distribution | grade B
    - Candidate dissertation JPL_INSTRUMENTS_NAV_05, sec 2.3.2 and 2.3.4 (signature-not-loop concession; cross-sectional, fixed mission/epoch) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[empirics]** GROUNDED CONCESSION. The dominant request-to-allocation loop delay is asserted ('a delay longer than the window itself', 2.3.2) but never MEASURED. The concentration estimators are a Gini coefficient and top-decile window share computed over discretized geometry-phase cells (4.3.2), with explicitly NO lag structure. The design therefore cannot separate delay-generated concentration (the Forrester loop) from concentration that is the fixed celestial-mechanics geometry of overlapping viewing windows, which the dissertation itself states is 'set by celestial mechanics and is not chosen by the mission' (4.4.4) and 'carries no feedback at all.' Indeed sec 4.4.4 promotes that geometry to a behavior-INDEPENDENT robustness instrument, conceding the concentration could be pure deterministic geometry. The two discriminating tests Forrester would demand, (a) an estimate of the loop delay relative to window length, and (b) a geometry-only placebo that holds operator behavior fixed, are absent. Correct answer: the measured concentration is, on this design, equally consistent with static deterministic geometry (no loop) as with delayed feedback; the delay measurement and geometry placebo that would settle it are not performed.
    - Forrester thinker dossier (hall_of_shoulders): 'if the delay exceeds the horizon over which you claim success, your equilibrium is overshoot' + leverage/loop-structure lens (anchored to Meadows, Leverage Points; Forrester Industrial Dynamics) | https://doi.org/10.1002/sdr.284 | grade A
    - Sterman, Bathtub Dynamics (stock trajectory and its delay cannot be inferred from a no-lag snapshot) | https://doi.org/10.1002/sdr.198 | grade A
    - Candidate dissertation JPL_INSTRUMENTS_NAV_05, sec 4.3.2 (Gini/top-decile over geometry-phase cells, no lag) and 4.4.4 (geometry set by celestial mechanics, behavior-independent) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[identification]** The rival is identifiable with the predicted SIGN of the lagged coefficient, and this is the correct framing because system behavior is dominated by closed loops of causation in which actors perceive local short-term cause-and-effect but not the delayed loop structure (Forrester). In a within-mission fixed-effects panel, regress this cycle's padding/early-submission on PRIOR-cycle realized pass-loss at the same geometry-phase window. EXOGENOUS-demand null: lag-1 coefficient is zero or NEGATIVE and statistically indistinguishable from zero once mission and window fixed effects absorb the geometry, because last cycle's loss carries no information about a mission's own request behavior when the collision is a physical property of the orbit. ENDOGENOUS deferral-hedge: lag-1 is POSITIVE and significant (a window that deferred you last cycle raises your padding this cycle) and lag-2 is positive but attenuated (geometric decay of the hedge memory), the ordering-policy signature of a balancing controller whose information feedback teaches missions to re-pad against exactly the windows H1 then flags. The decisive discriminant is therefore the sign-and-decay pattern of the lag structure, NOT the cross-sectional concentration, which both hypotheses reproduce; a positive, decaying lag profile is the closed-loop diagnosis and the static GLM has reversed the arrow. The candidate must additionally demonstrate the standard endogeneity test: reproduce the heavy-tail concentration using only the internal feedback structure with no exogenous geometry shock; if the concentration requires an external driver to appear, the missing loop is identified, and if it does not, the concentration is a control artifact.
    - Forrester, J.W., Industrial Dynamics / Dynamic models of economic systems and industrial organizations (System Dynamics Review) | https://doi.org/10.1002/sdr.284 | grade A
    - Forrester, J.W., Counterintuitive Behavior of Social Systems (1971) | https://doi.org/10.1007/bf00148991 | grade A
    - D'Ambrosio et al., Novel Source-Sink Model for Space Environment Evolution with Orbit Capacity Assessment (2023) | https://doi.org/10.2514/1.A35579 | grade A
- **[rival]** The dynamic falsification is well-posed and the predicted measurements follow directly from balancing-loop theory. Define loop gain as the fraction of relieved capacity that re-fills with newly padded/escalated requests over the post-intervention cycles. OPEN-LOOP (genuine scarcity relieved): rebound fraction near 0, requested load at the window monotonically decays toward and stabilizes below the new capacity, and the carrying window does not migrate, because the relief addressed a fixed physical shortfall. CLOSED-LOOP set-point restoration: rebound fraction approaches 1 (loop gain near unity), requested load returns to saturation within a settling time of order the controller's information delay (a few request cycles), and the 'carrying window' reappears or MIGRATES to the next-binding geometry-phase slot, because the controller defends a backlog set-point and policy resistance counteracts the intervention. This is the diagnostic Forrester named policy resistance: a complex system tends to respond to a well-intentioned policy by counteracting it, and a low-leverage parameter intervention (more aperture) can deepen the very saturation it targeted because it acts on a number rather than on loop structure or the strength/gain of the balancing loop (Meadows leverage hierarchy: parameters #12 are weak, balancing-loop strength #8 and reinforcing-loop gain #7 are strong). The payoff claim that targeted aperture durably recovers science survives ONLY if the measured rebound fraction is near zero and the window does not migrate; a rebound fraction near one with bounded settling time refutes durable relief and reclassifies carrying windows as a moving set-point artifact.
    - Forrester, J.W., Counterintuitive Behavior of Social Systems (1971) | https://doi.org/10.1007/bf00148991 | grade A
    - Meadows, D.H., Leverage Points: Places to Intervene in a System (1999) | https://doi.org/10.4324/9781849773386-15 | grade A
    - Forrester, J.W., Urban Dynamics | https://doi.org/10.2307/214050 | grade A
- **[mechanism]** The decomposition is the correct test and a static GLM cannot perform it, because convexity in a measured load-response is exactly what a rising-gain balancing loop produces when load feeds an information channel that amplifies requests faster than physical service degrades. The executed/requested ratio is a defensible proxy for instantaneous hedge gain: as backlog rises the ratio falls (more requested padding goes unexecuted), so its time path tracks the controller's amplification. Decompose by conditioning the load-response on the executed/requested ratio (e.g., interact load with the ratio, or stratify/match on it). PHYSICAL saturation: the convex curvature of lost-downlink-on-load PERSISTS at a fixed executed/requested ratio, because antenna service degrades nonlinearly with true offered load independent of hedging. ENDOGENOUS hedge amplification: the curvature VANISHES or collapses toward linear once the ratio is held fixed, because the convexity lived in the rising hedge gain (the controller's gain curve) and not in the physical service function. This is the same identification discipline as the other facets: reproduce the convex signature from the internal feedback structure alone, with the physical capacity held linear, and if it reproduces, the fragility is a loop artifact; the curvature must be shown to survive differencing-out of the hedge component before it can be claimed as a physical Taleb fragility curve. Because information delays and rising loop gain generically produce overshoot, oscillation, and accelerating response, the burden is on the candidate to separate the controller's gain curve from the antennas' service curve, which the static cross-sectional GLM in 5.3 structurally cannot do.
    - Forrester, J.W., Industrial Dynamics / Dynamic models of economic systems and industrial organizations (System Dynamics Review) | https://doi.org/10.1002/sdr.284 | grade A
    - Meadows, D.H., Leverage Points: Places to Intervene in a System (1999) | https://doi.org/10.4324/9781849773386-15 | grade A
    - Forrester, J.W., Counterintuitive Behavior of Social Systems (1971) | https://doi.org/10.1007/bf00148991 | grade A
- **[measurement]** The question is well-posed and the records needed to answer it exist in principle: the DSN schedule of record is produced by the Service Scheduling Software (S3, the modern mid-range successor system) through an explicit peer-to-peer negotiation among all DSN users, with a central scheduling database; this means submitted requests, conflicts, and the negotiated outcome are first-class objects in the archive, so a capture census (requests/executed/dropped captured vs inferred from aggregate loading) is a measurable property the candidate can and must report, not an assumption. McDowell's standing methodological commitment (dossier facet F6) treats 'how full is this record, and by what measure?' as a first-order question and demands completeness be quantified before any derived statistic. RETRIEVAL DOES NOT CONTAIN the candidate's actual capture fractions or their cross-complex/band/epoch distribution; those numbers are facts about the candidate's own dataset, absent from every corpus queried, so no capture-rate value can be asserted here.
    - Automating Mid- and Long-Range Scheduling for the NASA Deep Space Network (NASA NTRS 20130009147) | https://ntrs.nasa.gov/citations/20130009147 | grade C
    - Hall of Shoulders, Jonathan McDowell Reviewer-Brain Dossier (collegium hos-mcdowell) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/mcdowell/ | grade C
- **[identification]** The reconciliation the question demands is the canonical McDowell move and the DSN literature confirms an independent second record exists in principle: the DSN scheduling problem is formally a request/requirement-language problem with explicit conflict resolution, and rescheduling systems (e.g. the Demand Access Network Scheduler) operate on a baseline schedule that is revised for equipment outages and weather, meaning executed-pass and station-downtime records are produced independently of the demand-request stream. McDowell's dossier makes two-source reconciliation ('how do your counts reconcile with operator-declared figures and the independently tracked population') the discipline's hygiene, so a single arrangement-filtered stream serving as both demand and outcome is exactly the unverifiable configuration he flags. RETRIEVAL DOES NOT CONTAIN any window-by-window reconciliation of this candidate's counted pass losses against an independent mission-ops or station-downtime log, nor evidence on whether disagreements cluster on the H1 geometry-phase windows; those are facts about the candidate's analysis, absent from all queried corpora, so no reconciliation result, hedge-reclassified top-decile share, or capture-rate-by-window finding can be asserted.
    - The Deep Space Network scheduling problem; Automating Deep Space Network scheduling and conflict resolution (NASA NTRS 20060042808; 20060042856) | https://ntrs.nasa.gov/citations/20060042808 | grade C
    - Resource Scheduling for a Network of Communications Antennas (Demand Access Network Scheduler, NASA NTRS 20210003845) | https://ntrs.nasa.gov/citations/20210003845 | grade C
    - Δ-MILP: Deep Space Network Scheduling via Mixed-Integer Linear Programming, IEEE Access (2022) | https://doi.org/10.1109/access.2022.3164213 | grade A
- **[measurement]** The candidate's own Sec 3.7/4 names a catalog of record: the DSN service catalog and Service Preparation Subsystem (SPS) schedule archives are the PRIMARY event source, NTRS loading/forecasting reports are demoted to a CROSS-CHECK role, and mission service agreements supply downlink requirements; the unit is fixed in Sec 1.7 as 'one requested tracking activity, for one mission, on one antenna class, over one viewing window' (the analytical-panel row). The multi-stage process McDowell invokes is real and externally documented: Johnston et al. 2014 (AI Magazine) describe the DSN Scheduling Engine as interpreting user scheduling requirements, ELABORATING them into tracking passes, and resolving conflicts within a peer-to-peer schedule-negotiation process, with the system being extended to long-range planning. So the structural premise (one physical pass can appear as forecast, negotiated, and realtime rows) is correct and the candidate has named a primary catalog, satisfying McDowell's F1 (catalog as ground truth) and F6 (definitional rigor / reconciliation) at the level of WHICH archive is canonical.
    - Johnston, Tran, Arroyo, Sorensen, Tay et al., 'Automated Scheduling for NASA's Deep Space Network', AI Magazine 35(4), 2014 (Crossref abstract: DSE interprets requirements, elaborates them into tracking passes, resolves conflicts, peer-to-peer negotiation, extended to long-range planning) | https://doi.org/10.1609/aimag.v35i4.2552 | grade A
    - Johnston, Tran, Arroyo, Page, 'Automating Mid- and Long-Range Scheduling for NASA's Deep Space Network', SpaceOps/AIAA 2012 (multi-stage mid- and long-range scheduling) | https://doi.org/10.2514/6.2012-1296235 | grade B
    - JPL_INSTRUMENTS_NAV_05 dissertation Sec 1.7 (unit-of-analysis definition) and Sec 3.7/4.7 (SPS schedule archives as primary catalog of record; NTRS demoted to cross-check) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[rival]** The McDowell 'trend, not snapshot' and definitional-stability objection is a verified standing lens in his dossier: his interrogation protocol explicitly demands 'Show your claim against the launch/activity trend' and treats reconciliation of competing counts and reproducible definitions as the discipline's hygiene (F4 activity/trend accounting; F6 definitional rigor). This GROUNDS the legitimacy and exact form of the objection (a pooled non-stationary catalog can manufacture an apparent power law from a mixture of regimes) and obliges the candidate to fit the tail within fixed-configuration epochs. The dissertation already carries 'epoch fixed effects' in the regression (Sec abstract/methods), which addresses level shifts but does NOT establish that the tail EXPONENT and the top-decile share are invariant within a fixed antenna-complement regime; epoch dummies in a mean-regression do not test stability of a Clauset-Shalizi-Newman / generalized-Pareto tail estimate.
    - mcdowell dossier (Hall of Shoulders), interrogation lens 'Trend, not snapshot' and frameworks F4 (activity-statistics/trend accounting) + F6 (definitional rigor and reconciliation as the discipline's hygiene) | file:///D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/mcdowell/ | grade C
    - McDowell 2020, 'The Low Earth Orbit Satellite Population and Impacts of the SpaceX Starlink Constellation', ApJL 892 L36 (exemplar of catalog-anchored trend accounting cited in dossier F4/F5) | https://doi.org/10.3847/2041-8213/ab8016 | grade A
- **[identification]** The methodological demand is well-founded: in Ostrom's Institutional Analysis and Development (IAD) framework a governance situation is generated jointly by three exogenous classes, biophysical/material conditions, attributes of the community, AND the rules-in-use, which together structure the action arena and outcomes. Modeling only geometry/band/load (biophysical) and fixed effects while leaving the allocation rule unmodeled mis-specifies the action situation: the rule-in-use is a first-class explanatory variable, not background, so attributing a heavy tail to physics while the priority/deferral rule is omitted is an identification error in IAD terms. DSN time is a genuinely demand-oversubscribed scheduled resource (mission requests exceed antenna availability and are resolved by a negotiated priority/preference process), so an allocation rule does exist to be modeled.
    - Ostrom dossier (Hall of Shoulders, brain=ostrom), citing E. Ostrom, Understanding Institutional Diversity, Princeton Univ. Press, 2005; and 'A general framework for analyzing sustainability of social-ecological systems' | https://doi.org/10.1126/science.1172133 | grade A
    - Claps et al., 'Xi-MILP: Deep Space Network Scheduling via Mixed-Integer Linear Programming', IEEE Access (2022) | https://doi.org/10.1109/access.2022.3164213 | grade B
- **[mechanism]** The mechanism the question proposes is theoretically licensed: Ostrom's most counter-intuitive finding is that cheap, mutual, accountable monitoring (design principle 4) is the keystone that makes graduated sanctions and trust self-reinforcing, and that without accountable monitoring and an enforceable graduated sanction a common-pool institution collapses into the open-access dynamic regardless of load. Strategic padding of a requested quantity is precisely a monitoring-and-enforcement failure in this vocabulary (an appropriator over-claims a subtractable resource because the discipline does not bind the request to true need). Therefore 'concentration of loss tracks breakdown of bargaining discipline' is a distinct, named generative mechanism that competes with the geometry-window hypothesis and must be tested, not assumed away. A request-vs-execution gap is the natural observable of that failure.
    - Ostrom dossier (Hall of Shoulders, brain=ostrom), citing E. Ostrom, Governing the Commons: The Evolution of Institutions for Collective Action, Cambridge Univ. Press, 1990 | https://doi.org/10.1017/CBO9780511807763 | grade A
    - Weeden & Chow, 'Taking a common-pool resources approach to space sustainability: A framework and potential policies', Space Policy 28(3) (2012) | https://doi.org/10.1016/j.spacepol.2012.06.004 | grade A
- **[governance]** The governance critique is sound and citable: Ostrom's panacea critique holds that no single institutional lever is a universal remedy and that prescriptions must follow a diagnostic decomposition of the specific situation rather than a blueprint; treating one lever (here, aperture) as the universal fix is exactly the panacea error she warned against. Because the deferral discipline is a collective-choice (rules-in-use) variable, not a biophysical constant, a fraction of concentrated loss is institutional and curable by re-ruling rather than by capital. It follows that, in principle, concentrated loss must be decomposed into a physical-oversubscription component and a deferral-rule-artifact component before any remedy is recommended; recommending aperture without that decomposition risks spending capital to fix a collective-choice problem.
    - Ostrom dossier (Hall of Shoulders, brain=ostrom), citing E. Ostrom, 'A diagnostic approach for going beyond panaceas', PNAS 104(39):15181-15187, 2007 | https://doi.org/10.1073/pnas.0702288104 | grade A
    - 'Moving beyond panaceas in fisheries governance', PNAS (2018) | https://doi.org/10.1073/pnas.1716545115 | grade A
- **[governance]** The partition Ostrom demands is operationally available rather than hypothetical: NASA's DSN schedule is built through a documented multi-stage, peer-to-peer negotiation among its user community via the Service Scheduling Software (SPS/S3), which maintains multiple versioned schedule 'workspaces' and a revision history, and is explicitly architected across distinct phases (mid- and long-range planning/forecasting and near-real-time scheduling) with separate search- and repair-based algorithms for conflict and constraint-violation resolution. This means the candidate's logs CAN in principle be cut into the rules-on-paper stage (published priority/forecast discipline) versus the rules-in-use stage (realized inter-mission de-conflict trades in the revision history). What is NOT settled by any retrieved source is the candidate's own finding: no retrieved evidence reports the stage at which the heavy-tail concentration is actually produced in this candidate's data. The Forrester-loop / fixed-effects treatment that absorbs inflated requests and early submission as exogenous noise is, in Ostrom's IAD terms, absorbing exactly the collective-choice action situation that should be the unit of analysis; the institutional-diagnosis question (fragile rule vs fragile institution) is unanswered until the stage-partition is run on the SPS revision history.
    - Johnston, Tran, et al., 'Automating Mid- and Long-Range Scheduling for NASA's Deep Space Network' (SpaceOps 2012) | https://doi.org/10.2514/6.2012-1296235 | grade B
    - Johnston et al., 'Automated Scheduling for NASA's Deep Space Network', AI Magazine 35(4), 2014 | https://doi.org/10.1609/aimag.v35i4.2552 | grade A
    - Ostrom dossier, Collegium Hall of Shoulders (hos-ostrom): rules-in-use vs rules-on-paper; IAD framework | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/ostrom/ | grade C
- **[governance]** The premise of the question is corroborated and the polycentric framing is the correct lens: the DSN is not a single central allocator whose rule is simply redesigned, but a distributed peer-to-peer negotiation among a defined, bounded user community (~37 projects including international partners and ground-based science/calibration users), each able to make changes across shared schedule workspaces. This is a polycentric arrangement of semi-autonomous mission planning centers under shared rules and shared schedule visibility, exactly Ostrom's polycentric case. Therefore the candidate's three Sec 6.1 levers (aperture, scheduling-discipline change, demand shaping) presuppose a single redesignable allocator that the operational architecture contradicts. The user-community boundary (Ostrom design principle 1: clearly defined boundaries and a recognized appropriator set) IS specifiable from the SPS participant roster and submit/escalate/trade permissions. What is NOT settled by any retrieved source is the candidate's empirical result: whether realized loss concentrates on appropriators excluded from or weakly represented at the mid-range negotiation table. That representation-vs-contention partition is unanswered in retrieval and is the binding diagnostic that decides whether aperture or a change in window governance is the operative remedy.
    - Johnston, Tran, et al., 'Automating Mid- and Long-Range Scheduling for NASA's Deep Space Network' (SpaceOps 2012) | https://doi.org/10.2514/6.2012-1296235 | grade B
    - Goh et al., 'Deep Space Network Scheduling via Mixed-Integer Linear Programming', IEEE Access, 2021 | https://doi.org/10.1109/access.2021.3064928 | grade A
    - Ostrom dossier, Collegium Hall of Shoulders (hos-ostrom): boundary specification (principle 1), polycentricity vs centralization; cf. Morin & Richard 2021 polycentric space-debris governance | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/ostrom/ | grade C
- **[identification]** Rubin's review discipline supplies the exact form of the demand: a causal estimate is licensed not by the outcome model but by the assignment mechanism, the process by which units came to receive treatment as a function of covariates and possibly potential outcomes, which in observational work is unknown and must be reconstructed and defended. The exclusion restriction the candidate must state is precisely that geometry enters pass-loss ONLY via concurrent contention and is conditionally independent of the mission's own potential outcome (the pass's declared value/criticality) given covariates. The dossier names this failure mode directly: an associational object is reported, a causal verb attached, and 'no assignment mechanism or design is supplied to license the counterfactual contrast.' If geometry co-determines both contention (treatment) and the pass's science value (a determinant of the outcome), it is a confounder, and a propensity/balancing adjustment removes bias only from OBSERVED covariates and 'does nothing for unmeasured ones'; mission fixed effects remove only the time-invariant level, not the within-mission within-epoch variation the candidate relies on. The candidate, not this brain, must exhibit the specific log/criticality fields that test the restriction; the brain settles only that the restriction must be written and defended, not asserted by labeling geometry an instrument.
    - rubin reviewer-brain dossier (Hall of Shoulders), Frameworks sec. 2, citing Rubin, 'Bayesian Inference for Causal Effects: The Role of Randomization,' Annals of Statistics (1978) | https://doi.org/10.1214/aos/1176344064 | grade A
    - rubin dossier Frameworks sec. 3, citing Rosenbaum & Rubin, 'The central role of the propensity score in observational studies for causal effects,' Biometrika (1983) | https://doi.org/10.1093/biomet/70.1.41 | grade A
    - rubin dossier Review-lens Q2 ('Defend the assignment mechanism') and Synthesis sec. 3 | https://doi.org/10.1017/cbo9781139025751 | grade A
- **[mechanism]** SUTVA requires (a) no interference, one unit's treatment does not change another unit's potential outcomes, and (b) no hidden versions of treatment, a single well-defined intervention; violations make the estimand itself incoherent BEFORE any estimation question arises. By the candidate's own variable construction the treatment IS others' treatments, so the no-interference limb fails by construction, exactly the structure the dossier flags for shared-resource space settings ('one operator's treatment changes another's potential outcome, so SUTVA's no-interference limb fails; restate your estimand to respect interference'). Mission fixed effects remove a unit's time-invariant level and do nothing about a contemporaneous spillover, so a unit-level estimand is not merely biased but ill-posed; the coherent target is an interference-aware (exposure-mapping) estimand. The strategic-padding limitation additionally threatens the no-hidden-versions limb: if a 'requested hour' is sometimes genuine need and sometimes hedge, the exposure (others' requested load) is not a single well-defined treatment and the mapping is contaminated. Whether the SPS schedule archive contains fields that distinguish padded from genuine requested hours is a property of the candidate's data that no queried corpus establishes; the brain settles that the estimand must be rewritten to be interference-aware and that an ill-defined exposure breaks identification, and it places the burden on the candidate to show the exposure mapping is well-defined.
    - rubin dossier Frameworks sec. 4 (SUTVA), citing Rubin (1978) and Imbens & Rubin, 'Causal Inference for Statistics, Social, and Biomedical Sciences' (2015) | https://doi.org/10.1017/cbo9781139025751 | grade A
    - rubin dossier Review-lens Q4 ('SUTVA and overlap') and Synthesis sec. 3 (Challenge 2) | https://doi.org/10.1214/aos/1176344064 | grade A
- **[empirics]** The unit-level causal effect is the contrast Y_i(1) minus Y_i(0) on a defined set of units, and a causal question is not yet stated until both potential outcomes and the units are named; the dossier's first falsifiable demand is exactly 'write the estimand as a contrast of potential outcomes Y_i(1) - Y_i(0) on a defined set of units.' This validates Rubin's reframing: the contrast must be over the load condition on a fixed request, not over the served/lost outcome itself. On overlap, a treatment effect is estimable for a unit only if units with similar covariates exist under both arms (common support / positivity); where overlap fails, no amount of modeling recovers the effect and the honest move is to redefine the estimand to the sub-population where overlap holds, not to extrapolate. The dossier applies this to precisely the high-value-regime worry: when there is 'essentially no overlap' between a high-value regime and the data on which an effect was estimated, 'transporting an effect there is extrapolation past the support, exactly the move Rubin's overlap doctrine forbids,' and it requires the analyst to 'concede where overlap simply does not exist, in which case the honest estimand is narrower.' This settles that the candidate must (i) restate the estimand as a fixed-unit load contrast and (ii) report the no-common-support covariate cells before fitting; the actual cross-tabulation of the realized DSN log (which declination x band x phase x complex cells contain both low- and high-load realizations) is a fact about the candidate's data that no queried corpus contains and that the candidate must produce.
    - rubin dossier Frameworks sec. 1 and Review-lens Q1 ('Name the two potential outcomes and the units'), citing Rubin, 'Estimating causal effects of treatments in randomized and nonrandomized studies,' J. Educ. Psychol. (1974) | https://doi.org/10.1037/h0037350 | grade A
    - rubin dossier Frameworks sec. 7 (Overlap, positivity, and the limits of estimability), citing Imbens & Rubin (2015) and Rubin, 'For objective causal inference, design trumps analysis,' Ann. Appl. Stat. (2008) | https://doi.org/10.1214/08-aoas187 | grade A
    - rubin dossier Synthesis sec. 3 (Challenge 5, cislunar no-overlap example) and Review-lens Q4 | https://doi.org/10.1017/cbo9781139025751 | grade A
- **[identification]** GROUNDED CONCESSION, NO SETTLING DATASET IN HAND. In Rubin's framework the logged-loss indicator is itself an assignment-into-the-sample whose mechanism must be reconstructed and defended, not assumed; the relevant object is the propensity for a request to enter and be retained as a logged loss conditional on the geometry covariate, and concentration is observationally equivalent to survivorship unless that propensity is shown flat across windows (Rosenbaum & Rubin 1983; Imbens & Rubin 2015; rubin dossier on assignment-mechanism reconstruction). The candidate's own design CONCEDES the point structurally: the concentration clause 'stands or falls on the completeness of the log, not on an identification argument' and is deliberately placed 'outside the identification strategy and inside the descriptive layer' (Sec 5.3.1 firewall; Sec 6.2.3). Critically, the out-of-stream record Rubin demands does not exist in the candidate's panel: requested time is treated 'throughout as a strategic artifact', the only independent cross-check named is the NTRS AGGREGATE load (an interval-level total, not a per-request withdrawal/entry trail), and the mission service-agreement records are flagged as 'the least-sourced dataset' with 'no public citation fully documenting per-mission' windows (Sec 4.6.1, Sec 4.6.5). The honest answer: NO dataset in the assembled panel settles the entry/retention propensity by geometry window; the off-log triage trail is not collected, so the survivorship rival is NOT broken and the concentration result must be read as descriptive-of-the-realized-log only, exactly as the firewall already states.
    - Rosenbaum & Rubin, 'The central role of the propensity score in observational studies for causal effects', Biometrika (1983) | https://doi.org/10.1093/biomet/70.1.41 | grade A
    - Imbens & Rubin, 'Causal Inference for Statistics, Social, and Biomedical Sciences', Cambridge UP (2015); rubin dossier (Hall of Shoulders) | https://doi.org/10.1017/cbo9781139025751 | grade A
    - JPL_INSTRUMENTS_NAV_05 dissertation, Ch6 Sec 6.2.3 and Ch5 Sec 5.3.1, Ch4 Sec 4.6.1 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch6_analysis_plan.md | grade C
    - JPL_INSTRUMENTS_NAV_05 dissertation, Ch4 Sec 4.6.5 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch4_data_and_measurement.md | grade C
- **[rival]** GROUNDED PARTIAL ANSWER + CONCESSION. The correct covariate to name is a request-level, outcome-independent criticality measure built from mission-declared critical-event windows / mission-phase declarations (and, where available, instrument-cadence or science-plan records), NOT requested duration: the candidate already defines exactly this object as the 'criticality weight = mission-declared criticality of the phase' and the criticality-weighted construct of D (Ch4 measurement table; Sec 5.5.3), and it is in principle measurable from mission service-agreement / downlink-requirement records that pre-date the schedule outcome, satisfying Rubin's design-before-outcomes discipline (rubin dossier: 'design ... while blind to the outcome data'; Imbens & Rubin 2015). This is the right move because fixed effects only absorb between-mission and between-epoch variation, so a within-mission-epoch confounder requires an explicitly measured covariate to enter the propensity/regression (Rosenbaum & Rubin 1983). BUT the candidate concedes this covariate is precisely the one it cannot reliably obtain: the mission-record and criticality input is 'the least-sourced dataset', criticality weighting is 'feasible only for missions whose service agreements are accessible' and is carried as a partial-coverage robustness construct, deliberately kept out of the headline so the core claim does not depend on it (Sec 4.6.5, Sec 5.5.3, Sec 5.6 check 8). The candidate also already names this exact confound as its 'central internal-validity threat': within-mission timing of high-value requests into geometry-forced crowded windows survives the fixed effects (Sec 5.3.2, Sec 5.5.1). So: the covariate is correctly identifiable (criticality declarations, NOT duration) and outcome-independent in principle, but is NOT reliably measured across the portfolio, so the geometry-criticality confound is acknowledged-but-unresolved, not broken.
    - JPL_INSTRUMENTS_NAV_05 dissertation, Ch5 Sec 5.3.2 and Sec 5.5.1 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch5_research_design.md | grade C
    - Rosenbaum & Rubin, Biometrika (1983) | https://doi.org/10.1093/biomet/70.1.41 | grade A
    - Imbens & Rubin (2015); rubin dossier 'design trumps analysis; design without outcomes' | https://doi.org/10.1017/cbo9781139025751 | grade A
    - JPL_INSTRUMENTS_NAV_05 dissertation, Ch4 measurement table and Sec 4.6.5; Ch5 Sec 5.5.3 and robustness check 8 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/chapters/ch4_data_and_measurement.md | grade C
- **[identification]** GROUNDED PARTIAL: The dissertation correctly imports the Clauset-Shalizi-Newman ML-plus-goodness-of-fit-plus-likelihood-ratio procedure precisely because log-log least-squares over-detects power laws, and it cites the scarce-tail-data cautions (Nettasinghe-Krishnamurthy friendship-paradox sampling; Chattopadhyay modified-Lomax) acknowledging that the rarest worst windows are by definition undersampled. Taleb's own program supplies the warrant that in Extremistan the historical record systematically undersamples the tail, so a high observed success/coverage rate is weak evidence about tail exposure. REFUSED ON THE EMPIRICAL DEMAND: the candidate cannot supply the realized exceedance count N or a parametric-bootstrap discrimination-power result, because by its own statement the design-stage power analysis numbers are 'illustrative placeholders, not computed estimates, because no log has been analyzed and the realized sample sizes are unknown at design stage' (dissertation Sec 5.7). No retrieved source establishes the realized N above a DSN peaks-over-threshold cutoff or the LR-test power at that N, so Taleb's core charge that the test at the available N may adjudicate sample size rather than tail family stands unrebutted at the design stage.
    - Hall-of-Shoulders taleb dossier (citation-grounded), anchored to Taleb et al. non-naive precautionary principle (arXiv:1410.5787, 2014) | https://arxiv.org/abs/1410.5787 | grade B
    - JPL_INSTRUMENTS_NAV_05 dissertation, Sec 5.7 power analysis and Sec 2.4.3 / Ch 6 tail-fit specification | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
    - Clauset, Shalizi & Newman, Power-law distributions in empirical data (cited as ref [4] in the dissertation literature review) | https://doi.org/10.1137/070710111 | grade A
- **[measurement]** GROUNDED PARTIAL: The candidate explicitly recognizes the self-censoring/survivorship problem and names it as the central measurement claim: 'the request log is a hedged, strategic artifact, not a clean demand signal,' with strategic early submission and hedged requests treated as known biases, and the design built to measure the predicted tail and concentration rather than the artifacts of how missions ask for time. It also frames the absent-worst-record concern through the competing-risks treatment (expiry as a competing event, not ordinary censoring) and through reporting the load-response curve rather than the mean. This is the right direction of concern and matches Taleb's absent-evidence point that the worst event is undersampled by construction. REFUSED ON THE EMPIRICAL DEMAND: the candidate does NOT quantify the gap between logged demand and latent demand during the highest-concurrency windows, and no retrieved source supplies that quantified gap from the SPS schedule logs, the mission downlink-requirement records, or the NTRS loading reports. Critically, the bias direction Taleb names is conceded but not bounded: if the worst windows under-record, the observed distribution is truncated toward H0, which would bias the heavy-tail verdict toward false-negative (favoring the null), and nothing retrieved demonstrates this bias is small or correctable.
    - JPL_INSTRUMENTS_NAV_05 dissertation, Ch 4 measurement chapter (Sec 4 summary and construct definitions) and Sec 5.5 internal-validity threats | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
    - Andersen et al. competing-risks framework (cited as ref [64] in the dissertation); dissertation Sec 5.6 | https://doi.org/10.1093/ije/dyr213 | grade A
    - Hall-of-Shoulders taleb dossier, anchored to Taleb non-naive precautionary principle (arXiv:1410.5787, 2014) | https://arxiv.org/abs/1410.5787 | grade B
- **[empirics]** GROUNDED PARTIAL: The candidate explicitly concedes the out-of-regime boundary: the crewed-exploration era is projected (Abraham et al.) to shift the mission mix toward high-rate time-critical traffic and raise loading substantially, which the design cannot observe in advance; claims are conditioned on the studied mix; the load-response curve, not the satisfaction rate, is reported as the externally valid object; and 'extrapolation beyond the observed load range is flagged as extrapolation and is not asserted as established.' Proposition 3 (the convex load-response) is self-rated the lowest-confidence clause and tested only as a signature, not as a causal/forecast object. This concession is consistent with Taleb's own instruction to detect fragility from the second-order response rather than forecast the event. REFUSED ON THE EMPIRICAL DEMAND: the candidate does NOT bound where realized concurrent-load support ends, and does NOT run the held-out high-load backtest Taleb specifies; no retrieved source supplies either the empirical load-support ceiling from the DSN logs or a backtest of the fitted convex curve against held-out high-load cycles. Taleb's specific charge stands: a within-support convex fit says nothing about behavior past an unreached saturation/instability point, and using it to price crewed-era exposure is precisely the extrapolation the fragility framework was meant to avoid. The candidate's defense is honest scoping, not evidence that the second-order response holds out of range.
    - JPL_INSTRUMENTS_NAV_05 dissertation, Sec 5.8 external-validity / Sec 2.4.4 and Sec 1.7.4 boundary statements; Abraham et al. cited as refs [17] and [31] | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
    - Hall-of-Shoulders taleb dossier, anchored to Taleb & Douady fragility/mathematical-definition work and Antifragile | https://doi.org/10.1080/14697688.2013.829244 | grade B
    - Flyvbjerg et al., regression to the tail / megaproject cost convexity (cited as ref [61] in the dissertation) | https://doi.org/10.1016/j.envsci.2021.06.013 | grade B
- **[rival]** GROUNDED PARTIAL, THEN REFUSED ON THE EMPIRICAL DEMAND. The methodological warrant for the rival is real and retrieved: GPD/power-law tail estimates are well documented to be dominated by the few most extreme order statistics, so a leave-one-out/jackknife and top-k-exceedance deletion is the correct diagnostic to separate a structural fat tail from a leverage artifact, and the candidate's own gating statistic (a CSN likelihood-ratio test plus a GPD shape xi>0) is exactly the quantity such deletion would move. Taleb's Extremistan program supplies the prior that a tail estimate resting on a handful of realizations is fragile and that the sample mean/share is driven by a few extreme points by construction. BUT the question demands a result computed from the candidate's realized scheduling logs: the smallest deletion set, its fraction of total exceedances, the post-deletion shape parameter, and the post-deletion exponential-rejection test. No such computation is retrievable; the candidate's own design-stage figures (the one-half-to-two-thirds top-decile share, the exceedance count) are conceded in round 1 to be illustrative placeholders because no log has yet been analyzed. With no realized jackknife/deletion result in any retrieved source, the rival's charge that a single-digit set of windows could carry the entire heavy-tail verdict stands unrebutted; the candidate cannot show robustness it has not computed.
    - Clauset, Shalizi & Newman, Power-Law Distributions in Empirical Data, SIAM Review (2009) | https://doi.org/10.1137/070710111 | grade A
    - Stumpf & Porter, Critical Truths About Power Laws, Science (2012) | https://doi.org/10.1126/science.1216142 | grade A
    - Hall-of-Shoulders taleb dossier (citation-grounded), anchored to Taleb et al. non-naive precautionary principle (arXiv:1410.5787, 2014) | https://arxiv.org/abs/1410.5787 | grade B
    - JPL_INSTRUMENTS_NAV_05 dissertation, Sec 5.7 power analysis and Ch 6 tail-fit specification (as established in taleb round-1 interrogation) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_INSTRUMENTS_NAV_05/dissertation.md | grade C
- **[empirics]** GROUNDED PARTIAL, THEN REFUSED ON THE EMPIRICAL DEMAND. The preasymptotic-fragility critique is fully grounded and the prescribed remedy is the correct one. The retrieved literature confirms that (i) the CSN procedure itself recommends a likelihood-ratio test against named alternatives precisely because the power-law form is frequently NOT the best fit, (ii) log-normal and truncated distributions routinely masquerade as power laws over limited ranges so few claimed power laws survive scrutiny, and (iii) incorrect likelihood-based inference has produced spurious heavy-tail/scaling conclusions in a concrete published case, establishing the false-positive hazard is real and not hypothetical. The discipline the candidate invokes (Extremistan) does not license accepting the heavier tail; it demands the opposite, a calibrated guard against accepting a fat tail on the strength of undersampled or noisy data. A pre-registered parametric-bootstrap false-positive check against a synthetic light-tailed-plus-drift null that mimics the DSN arrival burstiness and skewed service is the appropriate calibration, and Taleb's program is the warrant for requiring it. BUT no retrieved source supplies the candidate's actual pre-registered null specification or the reported false-positive rate of its decision rule against that null. Without that number, the test's ability to discriminate mechanism from preasymptotic/measurement artifact is asserted, not demonstrated, so Taleb's charge stands: a CSN rule that has not been calibrated against a DSN-mimicking light-tailed-plus-drift null cannot be shown to adjudicate H0 versus H1 rather than to over-accept H1 on noise.
    - Clauset, Shalizi & Newman, Power-Law Distributions in Empirical Data, SIAM Review (2009) | https://doi.org/10.1137/070710111 | grade A
    - Stumpf & Porter, Critical Truths About Power Laws, Science (2012) | https://doi.org/10.1126/science.1216142 | grade A
    - Edwards et al., Incorrect Likelihood Methods Were Used to Infer Scaling Laws of Marine Predator Search Behaviour, PLoS ONE (2012) | https://doi.org/10.1371/journal.pone.0045174 | grade A
    - Hall-of-Shoulders taleb dossier (citation-grounded), anchored to Taleb et al. non-naive precautionary principle (arXiv:1410.5787, 2014) | https://arxiv.org/abs/1410.5787 | grade B

## Gaps

- **[identification]** No falsifiable epoch-stability test exists in the design. The candidate cannot currently show whether top-decile concentration (T10/G) holds constant across SPS epochs in which the priority discipline or padding behavior demonstrably changed, nor re-derive concentration from a structural two-player (scheduler vs requester) adaptation model rather than one realized equilibrium log. Mission FE + leave-one-mission-out controls for roster composition, not for discipline-induced re-gaming. Retrieval (AMOS, ACTA, Space Economy, OpenAlex) returned no source that settles whether observed DSN concentration is discipline-invariant; the air-traffic congestion-externality literature is adjacent but does not test DSN epoch-stability. Required: a panel split on documented discipline-regime changes with T10/G re-estimated per regime, or a structural request-and-allocation model the candidate already defers to future work. (raised by beaufre)
- **[empirics]** The geometry-durable vs mix-contingent per-window discriminant is not an output of the design. The covariates to build it exist (separable ephemeris-geometry vs mission-phase coordinates on each {geometry x phase x complex x band} cell), but the design produces no decomposition of T10/G into a geometry-driven share versus a mission-phase-driven share, and no per-window durability label. The planner therefore cannot be told which carrying windows to harden (celestial-mechanics-durable) and which will dissolve under a changed mission-phase mix. Required: a variance/coefficient decomposition or counterfactual re-weighting that holds geometry fixed while perturbing the mission-phase mix, classifying each carrying window. This is constructible from the stated covariate set but is not in the analysis plan. (raised by beaufre)
- **[mechanism]** The second-order recoil of the discipline lever is not quantified. The design concedes first-order wait redistribution and offers post-intervention re-running as the test, but contains no simulation or empirical re-run on a protected-deferred-class regime, and no model of requesters re-optimizing padding/early-submission against the new discipline. It therefore cannot answer, even ordinally, whether protected-subclass relief persists or the relieved window simply reappears elsewhere with aggregate lost downlink unchanged. Retrieval found no DSN-specific source settling this; the candidate's own bullwhip/order-stability citation supplies the mechanism but is invoked only for a different lever. Required: a counterfactual re-game simulation (requester padding best-response to the new discipline) or an empirical T10/G re-estimation across discipline regimes, comparing aggregate lost downlink pre- and post-protection. (raised by beaufre)
- **[mechanism]** REFUSED for the empirical core. No retrieved source (AMOS, ACTA, Space-Economy brains all empty on DSN scheduling; Beaufre brain is theory-only; OpenAlex/NTRS gap-fill returns DSN planning/MILP papers but none that track window identity across cycles) demonstrates cross-cycle migration of top-decile carrying windows or tests whether the concentration is a structural invariant versus a moving equilibrium. The candidate must supply a multi-cycle window-identity panel and a Markov/transition or persistence statistic on top-decile windows; until then Q1 cannot be settled and no number can be asserted. (raised by beaufre)
- **[identification]** REFUSED. No retrieved source operationalizes a falsifying log observable separating indirect attrition from direct decisive-window seizure in DSN logs (e.g. wholesale surrender of low-priority passes to guarantee one critical window, or a complex held in reserve as a credible threat). Beaufre's dossier asserts the direct/indirect distinction and the falsifiability demand ('if no observation could falsify your classification, your category is decorative') but provides no schedule-log operationalization, and the DSN scheduling literature retrieved does not classify contention by resolution mode. The candidate must define, in log terms, the seizure signature versus the attrition signature; the same Gini number cannot distinguish the two generative stories. (raised by beaufre)
- **[governance]** REFUSED. No retrieved source provides an ordinal freedom-of-action ledger across the three remedies or a back-tested padding-response on post-rule DSN logs. Beaufre's review-lens demands exactly this accounting (a lever that expands the adversary's room as much as your own is not advantageous), but supplies no data; OpenAlex/NTRS return no study simulating or back-testing mission padding re-optimization against a changed DSN scheduling rule, nor any distributive measure of who gains maneuver room. The candidate must run a counterfactual back-test of the heaviest padders' response on post-rule logs and report the redistribution-versus-recovery split; absent that, the recovery-of-science-return claim is unsupported. (raised by beaufre)
- **[measurement]** Retrieval cannot supply the candidate's own purposive sampling rule (tail-only vs off-tail confirming/disconfirming), the record type that constitutes the qualitative datum (anomaly reports vs scheduling correspondence vs operator narratives), or the stopping rule (saturation vs fixed n per window stratum). These are facts about JPL_INSTRUMENTS_NAV_05's design that no on-topic corpus contains; the candidate must specify them. The grounded standard is recorded as creswell_r2_c1; the candidate-specific procedure remains unanswered. (raised by creswell)
- **[empirics]** Retrieval cannot settle whether mission-declared criticality is a measured quantitative field or a qualitative judgment requiring coding, cannot identify which specific records populate the qualitative theme column, and cannot supply the count of windows where the volume ranking and the meaning ranking diverge. These are empirical properties of the candidate's own data and analysis, absent from every on-topic corpus. The grounded definition of the required joint-display integration is recorded as creswell_r2_c2; the candidate-specific empirics remain unanswered. (raised by creswell)
- **[governance]** Retrieval cannot supply the candidate's paradigm declaration, cannot confirm whether any field in the downlink/scheduling logs records the requesting mission's priority class or institutional standing, and cannot determine whether a priority-class stratification of where loss concentrates would change the recommended intervention. These are facts about JPL_INSTRUMENTS_NAV_05's design, data schema, and findings, absent from every on-topic corpus. The grounded paradigm distinction and the transformative-archetype precedent are recorded as creswell_r2_c3; the candidate-specific declaration and stratification remain unanswered. (raised by creswell)
- **[identification]** Whether the top-decile concentration survives a near-optimal re-solve of the realized request stream is UNANSWERABLE from the candidate's record: the dissertation is a design-stage plan with no logs analyzed and no re-solve performed, and it does not commit to the re-solve-to-optimality counterfactual dantzig demands. As written, H1 has not been shown to measure the demand/supply geometry rather than the incumbent heuristic's suboptimality. The candidate must add a re-solve-survival test to the analysis plan and execute it before the structural-vs-artifact claim can be settled. (raised by dantzig)
- **[measurement]** Whether the windows the candidate labels 'carrying windows' are exactly the high-shadow-price, genuinely capacity-binding windows is UNANSWERABLE from the record: the dissertation never computes dual/shadow prices and benchmarks concentration only against a uniform baseline, which cannot separate a physically oversubscribed window from an administratively deprioritized one. No optimized allocation has been solved, so no shadow prices exist to retrieve. The candidate must add a dual-extraction step at an optimized allocation and show carrying windows coincide with binding shadow prices. (raised by dantzig)
- **[empirics]** Whether the heavy tail and top-decile concentration persist on de-padded 'true need' is UNANSWERABLE from the record: the candidate concedes the padding/endogeneity threat but exhibits no de-padded series (executed-vs-requested or padding-capped window) and does not pre-register de-padding survival as the test that separates a demand-vs-supply primitive from operator gaming of the scheduler. Until that series is constructed and the tail re-fit on it, H1 cannot be distinguished from a measurement of strategic request behavior. The candidate must add an executed-vs-requested de-padding analysis as a pre-registered robustness test. (raised by dantzig)
- **[identification]** The decomposition of each estimand's loss into capacity-binding (irreducible) versus scheduling-suboptimal (avoidable) components is UNANSWERABLE from the candidate's record. By the candidate's own 5.1 and 5.3.1, all three estimands are functionals of the realized log with no optimal-allocation counterfactual, so none of the three carries a binding-vs-avoidable attribution; and because no held-out MILP re-solve has been run (5.10, 6.0), whether the top-decile carrying windows are still binding (positive shadow price) at the MILP optimum, rather than merely where the deployed heuristic rationed, is undemonstrated. The candidate must add an MILP re-solve-to-optimality on a held-out window, report the optimal-vs-realized objective gap as the avoidable-loss estimate, and show the carrying windows coincide with positive-shadow-price binding constraints before the irreducible-vs-suboptimal split can be claimed. (raised by dantzig)
- **[mechanism]** The split of the heavy tail into genuine physical contention versus an artifact of un-priced padding is UNANSWERABLE from the record. The candidate concedes (4.6) that coordinated padding can manufacture apparent concentration, which is exactly the Dantzig-Wolfe coordination failure (decentralized missions coupled by a shared complex-band-epoch, each mispricing it because no dual disciplines requests), but it computes no shadow-price/congestion-charge decomposition and constructs no de-padded penalty series, so the penalty-on-padded-requests minus penalty-under-shadow-priced-coordination gap dantzig demands does not exist in the design. Mission fixed effects and the geometry lever relieve but do not price the linking constraint. The candidate must add a decomposition that publishes a per-complex-band-epoch dual (congestion price) and re-measure the penalty on the resulting de-padded request stream before the heavy tail can be attributed to physical scarcity rather than missing-shadow-price coordination. (raised by dantzig)
- **[empirics]** Whether the convex load-response is a property of the feasible frontier or an artifact of where the deployed discipline rations is UNANSWERABLE from the record. The convexity of an optimal value function in the binding capacity means any near-optimal allocator shows increasing marginal cost at its frontier, so the candidate's convex curve is observationally confounded with ordinary capacity-limit behavior and is not, by itself, a fat-tail/fragility signature; the candidate's only robustness (drop the extreme load decile) does not address this confound, and the curve is fit only to the deployed scheduler's realized output (5.1/5.3.1 firewall), never to a re-solved near-optimal allocation on the same demand, and no estimation has been run (6.0). The candidate must re-solve a near-optimal allocation on the identical demand, fit the convex load-response separately to that allocation and to the deployed output, and show the convexity is present in the frontier (attributable to the feasible set) and not merely in the discipline, before the Taleb fragility reading can be distinguished from marginal-cost behavior at a binding capacity limit. (raised by dantzig)
- **[measurement]** No retrieved source reports this candidate's actual capture-rate census, the fraction of submitted requests, executed passes, and dropped/expired requests captured in raw logs versus inferred from aggregate loading reports, nor whether that capture rate is uniform across complexes, bands, and epochs or drops in high-contention windows. This is a property of the candidate's own dataset and cannot be settled from any external corpus; it must be produced by the candidate as a coverage census stratified by contention level before any Gini/heavy-tail statistic is trusted. Refused for lack of source-grounded numbers. (raised by mcdowell)
- **[measurement]** No retrieved source establishes whether the candidate's records contain an independent reconciled-demand field (priority class, criticality flag, post-hoc mission acknowledgment of a loss) that separates genuine unmet demand from strategically inflated hedge requests, nor what the top-decile concentration becomes after sub-threshold requests are reclassified as non-loss. The dependent variable is candidate-internal; whether the concentration survives hedge-stripping is unknown from any corpus and must be demonstrated by the candidate via a reclassification-and-recompute sensitivity analysis. Refused for lack of source-grounded numbers. (raised by mcdowell)
- **[identification]** No retrieved source provides a window-by-window reconciliation of this candidate's counted pass-loss events against a second independently-produced record, nor which record is missing the loss where the two disagree, nor whether the disagreement clusters on the geometry-phase windows H1 nominates as carriers. The two-source verification is the candidate's obligation; absent it, a concentration claim resting on one arrangement-filtered stream is unverifiable. Refused for lack of source-grounded reconciliation evidence. (raised by mcdowell)
- **[measurement]** The decisive sub-question is NOT settled by any retrieved source, including the dissertation: which single SPS stage row (forecast vs negotiated vs realtime) is the canonical 'request', the explicit de-duplication / supersession rule that proves the count is stable under split/merge/re-negotiation, and which of the three re-stated durations is the denominator. The dissertation names SPS as primary but does not specify a canonical stage or a supersession-collapse procedure, and no external source supplies a stable cross-stage request-identity key. By the candidate's own no-confabulation contract this must be answered from the SPS record structure, which was not retrieved this turn; until that demonstration exists the request count, every wait-time, and the top-decile concentration are exposed to double-counting of superseded entries. (raised by mcdowell)
- **[empirics]** No source retrieved this turn establishes (a) an accessible, independent DSN executed-pass returned-data-volume / tracking-data accounting record at pass granularity that the candidate could reconcile D against, or (b) a measured discrepancy rate between request-side nominal-rate estimates and measured returns, or (c) whether the heavy tail survives re-anchoring to measured returns. The candidate's design derives D from the SAME request-side stream that produced the demand signal (request-side nominal rate x requested duration), which is precisely the circularity McDowell's reconciliation discipline (F6: reconcile operator-declared counts against independently tracked records) targets. The vault sweep (OpenAlex/Crossref/NTRS) returned the Johnston scheduling papers and unrelated comms-survey hits but nothing documenting an independent executed-pass bit-accounting record usable for reconciliation. REFUSED: cannot assert the tail survives, cannot assert a discrepancy rate, and cannot assert that such a reconciliation is even feasible on the available archives, because no retrieved source settles it. (raised by mcdowell)
- **[rival]** No source retrieved this turn supplies the operative facts the candidate would need to DISCHARGE the objection: a documented DSN antenna-complement timeline (70m/34m subtractions/additions, arraying epochs) and logging-convention/floor-change dates at the resolution needed to define fixed-configuration regimes, nor any epoch-by-epoch tail-exponent / top-decile-share estimate demonstrating within-regime stability. The OpenAlex/NTRS queries on the antenna-complement history returned only unrelated comms-survey hits. So while the OBJECTION is grounded (claim c3), the candidate's AFFIRMATIVE demonstration that the heavy tail is a structural property surviving regime segmentation is unsupported by retrieval and remains owed. REFUSED on the affirmative: cannot assert the exponent is stable within regimes, and the existing 'epoch fixed effects' do not substitute for an epoch-segmented tail fit. (raised by mcdowell)
- **[identification]** Whether THIS candidate's DSN scheduling logs actually record priority class, escalation flags, and negotiated downgrades at the per-request level, and whether two requests matched on geometry/band/load but differing in priority/deferral rule can be constructed from them to estimate the rule's separate contribution, is not settled by any corpus retrieved this turn. The candidate's covariate vector, dataset schema, and data-limitations section were not in retrieval. The feasibility of adding the rule-in-use as an estimable covariate on the candidate's own logs is an open empirical claim that cannot be asserted without confabulation. (raised by ostrom)
- **[mechanism]** Whether the candidate's DSN logs in fact pair each request with its executed pass so that requested-minus-executed duration (padding) and per-mission escalation frequency are directly measurable, and whether the loss concentration empirically tracks that bargaining-discipline gap rather than geometry, is not established by anything retrieved this turn. The padding admission, the data-limitations section, and the log schema are the candidate's own artifacts and were not in any queried corpus. The claim that the request-vs-execution gap is observable in and decisive on this dataset cannot be asserted without confabulation. (raised by ostrom)
- **[governance]** Whether the named SPS schedule archives / this candidate's DSN log corpus actually support a per-window decomposition of concentrated loss into a physical-oversubscription part and a deferral-rule-artifact part (e.g., a counterfactual or matched-rule contrast within a fixed geometry window) is not settled by any source retrieved this turn. The specific archive contents, the realized-vs-counterfactual-rule identifiability, and whether the candidate's design already attempts this are the candidate's own artifacts, absent from retrieval. The feasibility of the decomposition on the candidate's data cannot be asserted without confabulation; the Ostromian requirement that it be done before prescribing aperture is grounded, but the empirical answer is open. (raised by ostrom)
- **[rival]** No retrieved source provides a stratified re-estimation of DSN allocation concentration by complex-band-phase resource unit, nor the candidate's per-stratum tail/Gini/top-decile numbers. The Ostrom-grounded conceptual warrant for demanding the stratification is solid (heterogeneous resource units with different subtractability/excludability profiles must not be pooled; the panacea critique forbids treating distinct CPRs as one), and the band-eligibility/service-catalog substrate plausibly exists in the SPS service definitions. But whether the candidate's pooled heavy tail survives stratification, or dissolves into a single congested Ka-band stratum, is an empirical claim that NO retrieved evidence settles. Refused as a finding; the panacea-test demand stands as a methodological requirement on the candidate, not an answer the expert can supply. (raised by ostrom)
- **[governance]** No retrieved source quantifies whether DSN allocation loss concentrates on appropriators excluded from or weakly represented in the mid-range negotiation. The expert can ground (a) that the arena is polycentric and bounded, and (b) that oversubscription forces loss onto a subset, but CANNOT supply the candidate's representation-weighted loss estimate. The contention-vs-representation distinction that decides aperture vs window-governance remedy is unanswered in retrieval. (raised by ostrom)
- **[empirics]** No queried corpus (AMOS returned 0; ACTA, Space-Economy, and the rubin brain hold only methodological content) contains the candidate's own DSN scheduling logs, SPS schedule-archive field structure, or mission downlink-requirement/criticality records. Therefore the specific data-level sub-asks cannot be answered from retrieval: (a) which exact log/criticality fields falsify the geometry exclusion restriction by showing geometry -> pass value -> pass-loss; (b) which SPS fields separate hedged/padded requested-hours from genuine need to identify the exposure mapping; (c) the realized cross-tabulation of low- vs high-load donor cells by declination x band x phase x complex. These are demands the candidate must satisfy from primary data; this grounded-expert brain asserts nothing about the contents of those records. (raised by rubin)
- **[empirics]** NO SETTLING EVIDENCE: the geometry-by-criticality cross-tabulation with served/lost cell counts cannot be produced from the retrieved record, and the design itself implies the off-diagonal will be sparse. The geometry-by-criticality cross-tab requires (a) the request-level criticality covariate that the candidate concedes is 'the least-sourced dataset' with 'no public citation fully documenting per-mission critical-event windows' (Sec 4.6.5) and (b) actual fitted DSN-log counts, but Chapter 6 states plainly 'the DSN scheduling logs have not been analyzed, no distribution has been fitted' and every result table is 'specified-but-unpopulated by design' (Sec 6.6). Rubin's overlap/positivity requirement is the correct test: an instrument identifies only where it has common support across arms (Imbens & Rubin 2015; rubin dossier 'SUTVA and overlap'), and the candidate's own argument that encounter/EDL geometry is simultaneously tightest AND highest-criticality (Sec 5.3.2) predicts collinearity that would empty the off-diagonal cells and break identification. The candidate also already concedes the geometry lever 'is not randomly assigned ... it relieves rather than eliminates the endogeneity' and is 'the more fragile of the three' estimands (Sec 5.3.3). Therefore the off-diagonal-overlap question is UNANSWERED on present evidence: no cross-tab exists, the criticality axis is not reliably measured, and the design concedes the lever does not eliminate the confound, so on the retrieved record the contention claim is NOT identified net of criticality and the rival is not ruled out. A refusal is correct: the only honest settling artifact (a populated cross-tab from fitted logs) is absent. (raised by rubin)
- **[identification]** No retrieved source supplies (a) the realized exceedance count N above the candidate's chosen DSN peaks-over-threshold cutoff, or (b) a parametric-bootstrap power result showing the Clauset-Shalizi-Newman likelihood-ratio test can separate power-law/GPD from log-normal or truncated-exponential at that N and significance level. The candidate's own power analysis is admitted to be illustrative placeholders with unknown realized sample sizes. The discrimination-power demand is therefore unanswered: the decision rule's ability to adjudicate H0 vs H1 (rather than sample size) at the rare-window N is undemonstrated. (raised by taleb)
- **[measurement]** No retrieved source quantifies the gap between requested-and-logged demand and true latent demand during the highest-concurrency DSN windows, and the candidate provides no such cross-check of SPS schedule logs against mission downlink-requirement records and NTRS loading reports. The candidate concedes the log is a self-censored, hedged artifact but does not bound the resulting bias. Because under-recording of the worst windows truncates the observed distribution toward the light-tailed null, the heavy-tail and concentration estimates could be biased toward false-negative, and that this is not the case is undemonstrated. The measurement objection is unanswered. (raised by taleb)
- **[empirics]** No retrieved source bounds where realized DSN concurrent-load support ends, and the candidate runs no out-of-sample backtest (fit on early/low-load cycles, predict held-out high-load cycles) to show the fitted convex load-response holds outside the fitting range. The candidate concedes that crewed-era load lies beyond recorded support and flags such extrapolation as not established, but a queue can be convex in-sample then saturate or invert past an unreached instability point. Whether the second-order response forecasts crewed-era exposure is therefore undemonstrated; honest scoping is offered in place of the backtest, so the empirics objection is unanswered. (raised by taleb)
- **[rival]** No retrieved source supplies the realized jackknife / top-k-exceedance-deletion result on the candidate's DSN pass-loss records: not the smallest set of records whose removal drives the GPD shape parameter to insignificance or below zero, not that set as a fraction of total exceedances, not the collapse of the top-decile loss share toward the ten-percent uniformity benchmark, and not the post-deletion exponential-rejection test on the surviving tail. The candidate's own tail figures are conceded illustrative placeholders with no analyzed log. Whether a single-digit number of high-leverage windows carries the heavy-tail and convex-load-response result, versus a structural fat tail robust to deletion, is therefore undemonstrated, and the rival/leverage objection is unanswered. (raised by taleb)
- **[measurement]** REFUSED: no retrieved source supplies an error model for the candidate's assumed nominal data rate, nor the empirical distribution of the requested-versus-realized data-rate ratio from mission link-budget and telemetry records, nor any bound showing that distribution is thin-tailed or non-drifting across the multi-year span. The candidate concedes in Sec 3.6 lim.3 that the nominal-rate assumption carries measurement error but states no error model for it, so the dissertation itself does not bound the measurement-error tail; and AMOS, ACTA, Space-Economy, the Taleb brain, and OpenAlex/Crossref return nothing that establishes the data-rate-ratio error distribution for this mission set. Because a multiplicative, mildly fat-tailed, or calibration-drifting nominal-rate error would mechanically manufacture a heavy upper tail in lost-downlink magnitude with no contention mechanism, and the measurement-error tail is not bounded below the claimed signal tail, the Extremistan finding cannot be distinguished from a fat-tailed-noise finding on the available evidence. The measurement objection is unanswered. (raised by taleb)
- **[empirics]** No retrieved source supplies the candidate's pre-registered null specification or the reported false-positive rate of its CSN-based decision rule against a parametric-bootstrap, light-tailed-plus-calibration-drift synthetic DSN log that reproduces the arrival burstiness and skewed service. The literature establishes that the test is preasymptotically fragile and that such a calibration is the correct remedy, but the candidate has not demonstrated, and nothing retrieved demonstrates, that its rule holds its nominal significance level (does not spuriously accept H1) on noise-only synthetic data. Whether the decision rule adjudicates mechanism versus artifact at the available threshold sample sizes is therefore undemonstrated, and the empirics objection is unanswered. (raised by taleb)