# Interrogation mind-map: JPL_AUTONOMY_EDL_06

Nodes: 121 | questions: 48 | grounded claims: 45 | gaps: 28

## Questions

- **[mechanism]** Will the NTRS Karlgaard-family reconstructions (MSL 20130010087, InSight 20200002910, MEDLI2) show the MSL ellipse reduction equals an isolated guided-entry effect plus an isolated range-trigger effect, or that the joint effect departs from the sum enough to overturn the sign/ordering the H1 decision rule treats as signature? (raised by ackoff)
- **[identification]** Does the assembled 9-to-11-vehicle design matrix contain even one mission flying guided entry without the range trigger (or vice versa), and absent that, on what evidence outside the candidate's own regression would three separable technology coefficients be justified over collapsing them to the coupled subsystem the data can identify? (raised by ackoff)
- **[measurement]** From the Mars 2020 LVS flight-performance and Safe-Target-Selection reconstructions, can the divert-enabled reachable-site gain be expressed as an ellipse-area reduction at all, or does the realized capability live in a site-reachability metric the area-based dependent variable cannot represent, so the curve attributes to TRN a number that misstates what TRN bought the mission? (raised by ackoff)
- **[measurement]** Is there a constructible NTRS/PDS measure of the reachable-and-safe site set per mission (count or fraction of candidate sites cleared by the EDL system) that would test whether the attributed levers act through ellipse area at all, or whether area is a node-level proxy that severs the site-difficulty-against-achievable-precision interaction that makes the capability matter? (raised by ackoff)
- **[mechanism]** Is there evidence in the Karlgaard-family reconstructions (per-phase hypersonic/parachute/powered error-budget decompositions) that GuidedEntry, RangeTrigger and TRN act additively in log-area rather than multiplicatively or with cross-phase coupling, and if the reconstructed budgets show interaction, on what basis can summed gamma coefficients mean anything? (raised by ackoff)
- **[governance]** Is there a documented MSR or human-Mars trade-study record showing landing-accuracy requirements are set jointly with mass, hazard-avoidance and surface-mobility budgets, such that a per-lever ellipse-reduction rate read off the curve could be checked against whether shrinking the ellipse is even the binding constraint on the containing architecture rather than asserted as a stand-alone investment input? (raised by ackoff)
- **[measurement]** Replace each binary 'first flew' dummy with a continuous capability-stock proxy (cumulative JPL reconstruction/EDL-simulation reports, guidance-algorithm version lineage, count of relevant-environment flight tests). Does the ellipse contraction load on the binary 'inserted' dummy or on the continuous capability stock already rising before and between insertions? (raised by christensen_c)
- **[identification]** Encode landing-site job-difficulty (site hazard density, terrain ruggedness, science-demanded targeting tolerance) as a covariate from the site-selection record. Does the technology block stay jointly significant once demanded targeting difficulty is controlled, or is 'TRN' partly a proxy for the mission choosing to be hired for a harder targeting job? (raised by christensen_c)
- **[rival]** TRN is set to one for exactly one mission (Mars 2020), so its coefficient is identified by a single observation and is statistically indistinguishable from any other one-off feature of that mission. State pre-estimation what observation could separate 'TRN collapsed the position-knowledge error' from 'Mars 2020 differs in some other one-off way' (heavier divert authority, denser onboard reference map, more capable flight computer); if none can, the attribution is asserted, not tested. (raised by christensen_c)
- **[rival]** Is the fitted technology gamma in-class for the future job? The historical job is science-target tolerance (deliver inside a science ellipse); MSR-retrieval and Human-Mars are fixed-point/hazard-bounded jobs where 'good enough' is asset/hazard geometry, not science tolerance. Construct the test from the PDS achieved-miss-distance series and the Mars 2020 STS reconstruction (Dutta NTRS 20230005646) deciding whether the lever that shrinks a science-tolerance ellipse is the same lever, with the same proportional effect, that shrinks a fixed-point-rendezvous error; name the observable that reveals the coefficient is hired for a non-recurring job. (raised by christensen_c)
- **[mechanism]** Using TechPort program-of-record records and the NTRS Karlgaard-family reconstruction lineage, produce a measurable institutional-continuity proxy (sustained EDL engineering headcount, mission cadence, or calendar gap between successive technology insertions) and show whether the proportional contraction per technology generation is invariant to that proxy or co-moves with it. If the gamma co-moves with the institutional-continuity proxy, the learning curve is a closed value-network artifact, not a portable technology effect, and a new-entrant/new-mission-class planner in a different value network cannot lift the coefficient as a transferable price. (raised by christensen_c)
- **[governance]** State the explicit scope condition or provenance band attached to each gamma that stops an MSR-retrieval or Human-Mars planner from lifting it out of its historical job-and-value-network. Does the prospectus contain an operational rule, expressed as a coefficient annotation, a stated job-class boundary, or a value-network-continuity precondition, that would make a planner reading exp(gamma)-1 for a fixed-point-rendezvous or hazard-bounded-site job stop and recompute rather than extrapolate? If no such rule exists, the out-of-sample-honesty (Kuznetsian) claim is asserted for calendar time but not for job class. (raised by christensen_c)
- **[identification]** Run the blind reconstruction the seed implies: hand forecasters the eleven ellipse values stripped of mission names, dates, and technology labels, ask them to recover step ordering and breakpoint locations, and report the inter-rater agreement against the assigned ordering. Does the blind reconstruction succeed (showing structure the data carry) or only succeed after labels are revealed (showing a narrative fitted backward from the known two-order-of-magnitude improvement, the WYSIATI/hindsight failure mode)? (raised by kahneman)
- **[empirics]** State the base rate before privileging InSight's above-trend residual: across the NTRS reconstruction record, what is the empirical residual distribution of a single Mars EDL ellipse around a fitted log-linear trend, how large is a purely chance residual for one mission given simulation-convention noise and Monte-Carlo dispersion, and does InSight's residual exceed that documented noise band? A single salient point is exactly what availability/law-of-small-numbers overweights. (raised by kahneman)
- **[measurement]** Quantify the garden of forking paths: enumerate the defensible pre-fit specifications (within-program netting of Viking/MER singly vs doubly, with-and-without-Viking, MSL guided-entry-plus-range-trigger apportioned three ways, log vs linear) and report across the full grid how often the technology block stays jointly significant and correctly signed. What fraction of the defensible specification grid supports H1? (raised by kahneman)
- **[identification]** The three technology indicators were coded by people who already knew each mission's ellipse outcome, so the labeling step is downstream of the dependent variable and exposed to hindsight bias. Specify a contamination test the record can settle: code the covariates from the pre-flight TechPort and as-designed manifests dated strictly BEFORE landing, blind to the realized ellipse, and show the macro-vs-incremental assignment is identical to the post-outcome coding. If any value shifts when the coder is blinded, the technology block is partly an artifact of outcome knowledge. (raised by kahneman)
- **[empirics]** The pre-registered expected-signs table (guided entry negative and large, range trigger negative and smallest, TRN negative and largest, approach control insignificant) reproduces the engineering field's own causal narrative almost exactly, which is what WYSIATI predicts a well-told inside-view story to produce. State an out-of-sample check: hand the eleven ellipse values, stripped of which technology flew when, to the independent physical error-budget decomposition (per-mission downrange/crossrange, parachute-deploy, position-knowledge magnitudes in the Karlgaard-family and MEDLI2 documents), and show the magnitude ORDERING of the three error sources falls out of the independent physics in the same rank pre-registered from the economic narrative. (raised by kahneman)
- **[identification]** You concede the technology-versus-time discrimination is carried by the position of one vivid observation, InSight, relative to the trend, with a decision rule of above-trend yields qualified support / on-trend yields an honest non-result. That is a single salient case doing the work of a base rate. From the published reconstruction uncertainty bands on the InSight ellipse and on the modern fitted trend, compute the probability that InSight's residual sign reverses within those bands, and state at what reversal probability the decision rule should refuse to discriminate. If the residual sign is close to a coin flip on the actual published uncertainty, the linchpin is a salience artifact and the design returns no discrimination. (raised by kahneman)
- **[measurement]** State the explicit cross-mission deflator that makes a 1976 Viking three-sigma ellipse area commensurable with a 2020 Mars 2020 ellipse area, harmonizing (a) the probability content / tail mass encoded by Viking-era vs modern 'three-sigma' under non-Gaussian dispersions and (b) the simulation generation (1970s limited-trial vs MEDLI-informed Monte Carlo), then prove two independent analysts re-deriving each ln-area from the named primary records land within an order of magnitude. (raised by kuznets)
- **[identification]** Partition the total observed cross-mission contraction into (i) the component attributable to convention/simulation-fidelity drift in the ellipse-construction method and (ii) residual real contraction, and show H1 survives when (i) is differenced OUT of the raw ln-area series rather than absorbed into beta_1 or the gamma block; if the PDS achieved-miss-distance cross-check cannot isolate partial (not just gross) drift collinear with the guidance-generation indicators, the construct-validity test only detects gross divergence. (raised by kuznets)
- **[empirics]** Decompose the ~two-orders-of-magnitude contraction over ~9-11 bespoke vehicles in ~45 years and state what calendar window of comparable observations would be required to distinguish a sustained science-driven secular trend from a small number of discrete one-time level shifts (Viking->Pathfinder airbag heritage; MSL guided entry; Mars 2020 TRN); if a 3-4 step-function model fits the eleven points as well as or better than the smooth log-linear learning curve, on what evidence in the data is the constant-proportional-rate interpretation retained? (raised by kuznets)
- **[measurement]** Three named levers each attack a bounded error source, yet the prospectus states no asymptotic ellipse floor. Commit a number: what is the conjectured irreducible three-sigma ellipse area toward which the series asymptotes, decomposed into residual atmospheric-density/wind uncertainty at entry interface, IMU/FADS sensor-resolution limit, MRO/CTX onboard-map georeference accuracy, and divert-fuel reachability? An aggregate with an unstated lower bound is undefined. (raised by kuznets)
- **[empirics]** Gamma and exp(beta_1)-1 are slopes read off the interior of a series the candidate's own bounded mechanisms say must flatten toward a floor. State the within-eleven-point diagnostic that distinguishes 'still on the steep part of an S-curve' from 'approaching the asymptote' (e.g., monotone decline in successive log-decrements, the second difference of ln area across guidance generations; or curvature of residuals against sequence). If no such diagnostic separates the regimes, on what ground is the learning rate reported as a transportable price for the NEXT vehicle rather than a regime-local, level-conditional quantity?
- **[identification]** The InSight counterfactual runs the secular-vs-transient test backward (separating trend from one-time level shift inside the observed window); forward use (pricing MSR/Human-Mars off gamma) is the same test time-reversed and more fragile because the next vehicle sits beyond the data and possibly past the inflection. Specify, from the PDS achieved-miss-distance series and NTRS reconstruction provenance bands, the empirical test that would license forward extrapolation: do realized-capability decrements match the design-ellipse decrements at the last two generations, or has realized capability already flattened relative to the design ellipse? Absent recent-end agreement, why should any forward project read a price off the curve?
- **[identification]** GuidedEntry/RangeTrigger/TRN are binaries on the artifact while a generic Sequence index absorbs the directed Mars program that built, flight-tested, and trusted that hardware. Can the gamma coefficients distinguish 'flying the hardware shrinks the ellipse' from 'the directed public program capable of producing the hardware shrinks the ellipse'? Name the census observation that lets the data separate the lever from the institution. (raised by mazzucato)
- **[measurement]** The stated decision-relevance is that the gamma-vs-delta contrast lets a future architecture 'value a candidate guidance investment against an approach-navigation investment on a common quantitative basis.' What in the prospectus (the coefficient, a scope condition, or a provenance band) prevents an MSR or Human-Mars planner from reading gamma as the price of buying TRN in isolation, when the historical record shows that number was achieved only inside a program that had already paid for the knowledge base, flight-test infrastructure, orbital reference imagery, and in-house reconstruction capability? (raised by mazzucato)
- **[rival]** Mokyr's macro-vs-incremental distinction is inseparable from the institutional base that converts propositional into prescriptive knowledge, yet the covariates encode only the artifact's first-flight (TechPort dates), not the directed-investment trajectory behind it, so H1 and H0 are both incompletely specified absent a program-direction term. Is there a measurable proxy in NTRS/TechPort/budget records (sustained program funding, mission cadence, retained in-house EDL headcount) you could enter as a fourth structural term to test whether the directed institution, not the isolated lever, is the binding explanatory variable, and if not, on what basis do you exclude it? (raised by mazzucato)
- **[measurement]** The three binary technology dummies (GuidedEntry, RangeTrigger, TRN) encode discrete artifacts (a market-fixing ontology), but each artifact is the visible output of an accumulating, retained in-house JPL EDL/GNC capability. An artifact-coded model cannot separate 'the lever shrank the ellipse' from 'the standing public capability that produced the lever shrank the ellipse.' Construct a capability-stock measure independent of the artifacts (retained JPL EDL/GNC headcount, recurrence of named reconstruction authors across the Karlgaard NTRS lineage, sustained mission-to-mission EDL funding) and regress ln(EllipseArea) on it alongside the binary block: does the contraction load on the artifacts or on the stock that produced them? (raised by mazzucato)
- **[governance]** The gamma-vs-delta learning-rate contrast is sold as a tool to value a guidance investment against an approach-navigation investment for MSR and human Mars, but a learning rate read off nine bespoke vehicles reports rate, not direction, and is self-confirming because extrapolating yesterday's lever set forecloses tomorrow's. State the observable in TechPort technology-insertion and program-structure records that distinguishes a directional, capability-building, spillover-generating mission lever from a re-badged procurement line item, and show whether the three levers generated cross-program spillover (reuse of guided-entry or TRN heritage outside the originating mission) or stayed locked to the vehicle that bought them. (raised by mazzucato)
- **[economics]** Mokyr's prescriptive-knowledge frame, invoked to call the levers 'extensible' and 'self-correcting,' presumes the codified gain is a retained, reusable asset, but extensibility is a distributional question the design never asks: who captured the codified capability? Using NTRS author affiliations across the reconstruction lineage and the Mars 2020 LVS reference-map pipeline, plus TechPort technology-transfer/infusion records, classify each lever as a retained public asset (reusable across architectures and centers) or a capability concentrated in a single prime and not transferable without re-procurement. Does the prescriptive-knowledge premise behind H1's 'extensible' claim survive that provenance audit for all three levers, or only the ones whose codification stayed inside the public agency? (raised by mazzucato)
- **[identification]** The macro/micro ranking of EDL levers (guided entry and TRN as macro-inventions, range trigger as incremental) is read off the very ellipse series the regression is built to explain. Name a source of propositional-knowledge evidence logically prior to and independent of the ellipse series, e.g. the documented breadth and maturity of the theory each lever rests on. If no independent metric of the knowledge base can be specified, on what basis is the macro/micro labeling a finding rather than a covariate-design assumption? (raised by mokyr)
- **[measurement]** The design measures only the one-time level shift in the ellipse at each lever insertion. A large but isolated drop is observationally identical to a non-extensible trial-and-error artifact. What measurable indicator, available across the missions or their follow-ons, would distinguish an extensible lever whose accuracy gain compounds and transfers (TRN's reuse from the 2015 flight test through Mars 2020 to lunar/MSR) from a lever that delivered one drop and plateaued? Without it, is the learning curve measuring cumulative knowledge deepening or just a staircase of disconnected one-time gains? (raised by mokyr)
- **[rival]** Is the contraction attributed to three named technologies partly the spillover of an accumulating, codified reconstruction-and-modeling capability concentrated in specific JPL and Langley teams (Karlgaard reconstruction, Lander Vision System pipeline, entry-guidance algorithms), a confound the InSight counterfactual does not break because InSight was reconstructed by the same apparatus? What dated evidence, e.g. when each reconstruction method became codified and reusable versus when each guidance lever first flew, would separate a technology-insertion effect from a knowledge-access-cost effect? (raised by mokyr)
- **[measurement]** Construct a continuous propositional-maturity proxy (e.g. count of antecedent flight-validated relevant-environment tests and reconstructed atmosphere/aerodynamic models on which each lever rested before its first Mars flight) that is logically prior to and separable from the prescriptive dummy. Entered alongside the binaries, does area contraction load on the continuous propositional-depth proxy or on the bare 'inserted' dummy, and if it loads on the dummy, what licenses calling any of these a macro-invention rather than a datable artifact? (raised by mokyr)
- **[identification]** Every ellipse value in the dependent variable is itself a product of the Karlgaard statistical-reconstruction lineage, an accumulating codified modeling capability whose access cost fell over the study era, distinct from the flight hardware. State the test that separates a falling-access-cost-of-reconstruction effect from a real falling-dispersion effect in the vehicle (hold reconstruction vintage fixed, or enter reconstruction-method generation as a covariate from the NTRS record). Does the technology block survive once the measurement-codification capability is held constant? (raised by mokyr)
- **[rival]** Run the extensibility test the design already contains the materials for: place the held-out Tianwen-1 entry (Yu 2021) against the fitted US Mars technology coefficients. If guided entry and TRN are extensible levers on a deepening propositional base, the proportional ellipse reduction they buy on the independent non-US, non-Mars cases should fall within the permutation interval the US Mars curve predicts; if JPL-bespoke prescriptive artifacts, the out-of-lineage cases should sit outside. Which way does the held-out Tianwen-1 point fall, and does it corroborate or falsify the extensibility claim underwriting the 'self-correcting' framing? (raised by mokyr)
- **[identification]** ApproachAccuracy enters as a control whose H1 test hinges on a small/insignificant delta, but its structural role is undeclared. Is it a mediator (descendant of technology generation, must NOT condition on) or a confounder (upstream of both technology choice and ellipse, must adjust for)? And is it separately identified from the technology block in the nine points, or collinear and thus uninformative? (raised by pearl)
- **[mechanism]** The InSight observation is asked to break time-technology collinearity as a one-case event-study residual. What intervention does the InSight counterfactual license on the Ladder of Causation? InSight's unguided ballistic entry was a self-selected design choice (flat Elysium site, cost-constrained), not an experimental removal of guidance. Is the large ellipse evidence for do(no-guidance) -> large ellipse (rung 2), or merely selection on the outcome (missions self-select unguided entry when they can tolerate large ellipses, rung 1)? Can a testable conditional independence distinguish the two? (raised by pearl)
- **[identification]** GuidedEntry and RangeTrigger are 1 for exactly MSL and Mars 2020 and 0 elsewhere, so as design-matrix columns they are perfectly collinear and gamma_1, gamma_2 are not jointly identified. The prospectus concedes a 'bundle' but still lists separate expected signs and a Mokyrian ordering. Which conditional-independence implication, testable in this nine-to-eleven-point dataset, apportions the ellipse reduction between the two mechanisms? If none exists, on what external interventional evidence does the split rest, and should the two be collapsed to a single identified bundle coefficient? (raised by pearl)
- **[identification]** Upgrade the DAG to a selection diagram: add square s-nodes for every dimension on which the source population (US robotic Mars EDL, 1976-2020) differs from the MSR-retrieval / Human-Mars target (site-selection latitude, mission class / producer institution, atmospheric-density regime for high-mass vehicles). For each s-node, state from the EDL-reconstruction record (Karlgaard-family NTRS, Golombek site-selection, LVS / Safe-Target-Selection) whether an arrow runs INTO the technology-to-ellipse path. If even one s-node sits on that path, gamma is not transportable. (raised by pearl)
- **[rival]** State the s-admissibility verdict explicitly using do-calculus. Is the target estimand (proportional ellipse reduction a guidance lever buys in MSR / Human-Mars) expressible as a transport formula in which gamma is the SOURCE factor re-weighted by a target-measurable term, or does an s-node block transport because the technology effect itself is modified by site-selection latitude or atmosphere regime? Name: transportable, transportable-after-reweighting, or non-transportable. (raised by pearl)
- **[governance]** If non-transportable / transportable-only-after-reweighting, specify the recovery: what single target-population observation or auxiliary distribution from a real source (TechPort/TRN reference-map record for a pre-placed-asset architecture, an MSR EDL trade study, a Human-Mars high-mass entry simulation set) supplies the target-measurable reweighting term? Absent it, what scope sentence must the prospectus add so an MSR / Human-Mars planner cannot read gamma as a decision-relevant price, given 'confidence beyond the estimation population: low'? (raised by pearl)
- **[measurement]** Is the InSight-minus-Viking ellipse difference a measure of landing capability, or is it contaminated by instrumentation drift in the dependent variable itself, because both ellipses come from a reconstruction lineage whose conventions (confidence level, dispersion sourcing, atmosphere model) evolved over two decades? Which convention shifts can be documented and differenced out, and which remain confounded with the technology indicators and therefore cap the strongest claim? (raised by shadish_cook_campbell)
- **[identification]** Is the large InSight ellipse partly a designed-in consequence of choosing a flat, forgiving Elysium Planitia site (selection on the outcome), rather than solely an absence of guidance technology, and does the approach-accuracy control, which holds injection accuracy fixed but not site-selection latitude, genuinely close this rival? Show an observable from the site-selection records that separates 'ellipse large because guidance off' from 'ellipse large because the project could afford a large ellipse given a flat site.' (raised by shadish_cook_campbell)
- **[rival]** Would bringing Tianwen-1 (2021, guided modern-era entry) in-sample as a second technology-discordant observation, rather than holding it out as an external reference, convert the single-case technology-versus-time discrimination into a two-point comparison that survives the loss of either point, mirroring InSight? What design or data reason would justify leaving the only available second discriminating observation out of sample? (raised by shadish_cook_campbell)
- **[construct]** Pre-state the quantitative divergence between the design-ellipse series (NTRS) and the achieved-miss-distance series (PDS) that you will read as 'contraction is a simulation-convention artifact' vs 'convention drift is negligible'; then check whether their ratio drifts monotonically across the era in the same direction as the technology insertions, which would mean the two series cannot independently corroborate each other. (raised by shadish_cook_campbell)
- **[identification]** Name a concrete differencing/comparison series the record can supply, a contemporaneous Mars-relevant accuracy quantity that improved 1976-2020 but is causally downstream of approach navigation and modeling fidelity while independent of onboard EDL guidance (e.g. delivered entry-flight-path-angle dispersion or orbit-determination position uncertainty at atmospheric interface), and show whether, overlaid on the ellipse series, the ellipse contraction still shows discrete steps at MSL and Mars 2020 rather than tracking the comparison trend smoothly. (raised by shadish_cook_campbell)
- **[measurement]** Instrumentation is marked 'moderate' but unbounded: the Karlgaard statistical-reconstruction methodology matured across the missions it was applied to, so the instrument producing the dependent variable improved monotonically alongside the technologies. Exhibit even one reconstruction-uncertainty quantity (a stated reconstruction uncertainty, covariance floor, or CI width) that separates 'the ellipse got smaller' from 'the reconstruction got tighter,' and state whether that instrument-precision metric improves in lockstep with the guided-entry and TRN dummies, which would make instrumentation drift observationally indistinguishable from the technology effect. (raised by shadish_cook_campbell)

## Grounded claims

- **[mechanism]** Retrieval confirms the cited dependent-variable source reports exist (MSL EDL trajectory/atmosphere reconstruction 20130010087; InSight 20200002910; the MEDLI2 instrumentation family) and confirms Ackoff's standing claim that a system's behavior is governed by how its parts interact, not by how each part performs in isolation, so an additive model of a serially coupled chain severs the emergent property by construction. But NO retrieved source this turn reports the stage-conditioned covariance / additive-vs-interaction error-budget decomposition needed to state whether the MSL ellipse reduction is additive or interaction-dominated. The empirical resolution is therefore not settled by retrieval; only the conceptual hazard and the existence of the source reports are grounded.
    - Russell L. Ackoff, The Future of Operational Research is Past, J. Operational Research Society (1979) | https://doi.org/10.1057/jors.1979.22 | grade A
    - Karlgaard et al., Mars Science Laboratory Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction, NASA NTRS 20130010087 | https://ntrs.nasa.gov/citations/20130010087 | grade B
    - Karlgaard et al., Mars InSight Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction, NASA NTRS 20200002910 | https://ntrs.nasa.gov/citations/20200002910 | grade B
- **[identification]** The two source studies named as the would-be external evidence are real and retrievable: Way et al. 2011 'On the use of a range trigger for the Mars Science Laboratory Entry, Descent, and Landing' and the MSL entry-guidance line ('Entry Guidance for the 2011 Mars Science Laboratory Mission'). Ackoff's systems position supplies the identification principle: emergent interaction structure is destroyed when a coupled system is decomposed, and reducing a mess to the significant variables that explain a salient outcome is the characteristic reductionist error. But NO retrieved source reports the actual composition of the candidate's design matrix (which missions fly guided entry with vs. without the range trigger), nor per-stage footprint-reduction figures from Way/Mendeck sufficient to license three separable coefficients. The identifiability concern is conceptually grounded; the factual answer about co-occurrence in this specific design matrix is not retrievable from the corpus this turn.
    - Way et al., On the use of a range trigger for the Mars Science Laboratory Entry, Descent, and Landing, IEEE Aerospace Conf. (2011) | https://doi.org/10.1109/AERO.2011.5747242 | grade A
    - Mendeck & Craig, Entry Guidance for the 2011 Mars Science Laboratory Mission (OpenAlex W1998256208) | https://openalex.org/W1998256208 | grade B
    - Camillo & Mazzone, Tackling wicked problems in strategic management with systems thinking, Strategic Organization (2021) | https://doi.org/10.1177/14761270211038635 | grade A
- **[measurement]** Ackoff's expansionism grounds the methodological charge: a system is understood by its role in the larger containing system, emergent properties are destroyed when the system is taken apart, and improving a part while measuring the wrong whole is suboptimization, with 'doing the wrong thing right' the cardinal error. Retrieval also confirms the relevant Mars 2020 TRN reconstructions exist (Lander Vision System Flight Performance, NTRS 20230006986; Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts, NTRS 20230007014). But NO retrieved source reports whether the divert-enabled reachability gain is or is not expressible as an ellipse-area reduction, nor supplies a site-reachability quantity to compare against the area dependent variable. The expansion-grounds critique of choosing bounded ellipse-area is supportable; the specific empirical claim that area cannot represent the divert gain is not settled by retrieval.
    - Russell L. Ackoff, From Data to Wisdom / Towards a System of Systems Concepts (governing maxim: better to do the right thing wrong than the wrong thing right; expansion vs reduction; suboptimization) | https://doi.org/10.1057/jors.1979.22 | grade A
    - Mars 2020 Lander Vision System Flight Performance, NASA NTRS 20230006986 | https://ntrs.nasa.gov/citations/20230006986 | grade B
    - Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts, NASA NTRS 20230007014 | https://ntrs.nasa.gov/citations/20230007014 | grade B
- **[measurement]** Such a measure is constructible and the literature already operationalizes it: Mars 2020 Terrain Relative Navigation via the Lander Vision System cleared hazardous terrain at Jezero by building an onboard reference map from MRO/CTX imagery and diverting from hazards within the ellipse, so 'sites cleared by the EDL system' is a real, countable quantity, not an abstraction. This is Ackoff's expansion test: the containing system is the reachable-and-safe mission space, and it moves discontinuously (a site goes from un-attemptable to attemptable) precisely where a smooth log-area contraction cannot represent the change. Ellipse area is therefore a node-level proxy that severs the site-difficulty-by-precision interaction; the candidate has measured an amputated part, not the emergent whole the prospectus itself claims (Jezero 'a site earlier systems could not have attempted at all').
    - Johnson et al., Mars 2020 Lander Vision System Flight Performance, AIAA SciTech 2022 | https://doi.org/10.2514/6.2022-1214 | grade B
    - Johnson et al., Making an Onboard Reference Map From MRO/CTX Imagery for Mars 2020 Lander Vision System, Earth and Space Science 2021 | https://doi.org/10.1029/2020ea001560 | grade A
    - Ackoff, Towards a System of Systems Concepts / Systems thinking and thinking systems | https://doi.org/10.1002/sdr.4260100206 | grade A
- **[mechanism]** The reconstruction literature the candidate leans on documents exactly the cross-phase, sequential coupling that an additive log-area learning curve destroys, not independent part-effects. The MSL reconstructed-aerodynamics and Mars Entry Atmospheric Data System (MEADS) trajectory reconstructions, and the MEDLI2 inverse-aeroheating estimation, decompose the descent into coupled hypersonic, atmospheric, and powered phases whose errors propagate in sequence within one descent; each lever nulls a distinct error-budget term only conditional on the prior phase's state. This is the candidate's own concession (the MSL increment is an inseparable guided-entry-plus-range-trigger bundle) restated as physics: landing precision is an emergent property of the chain interacting. Ackoff's standard is that emergent properties vanish when the whole is taken apart into independent additive parts, so summed gamma coefficients over GuidedEntry+RangeTrigger+TRN are a reduction that quietly assumes away the interaction structure the reconstructions show; the coefficients can be fit but cannot carry the mechanistic meaning the candidate attaches to them unless additivity is demonstrated against those budgets, which the reconstructions do not support.
    - Karlgaard-family: Assessment of the Reconstructed Aerodynamics of the Mars Science Laboratory Entry Vehicle, Journal of Spacecraft and Rockets 2014 | https://doi.org/10.2514/1.a32794 | grade A
    - Karlgaard et al., Mars Entry Atmospheric Data System Trajectory Reconstruction Algorithms and Flight Results, AIAA 2013-28 | https://doi.org/10.2514/6.2013-28 | grade B
    - Inverse Estimation of Mars 2020 Entry Aeroheating Environments Using MEDLI2 Flight Data, AIAA 2022-0550 | https://doi.org/10.2514/6.2022-0550 | grade B
    - Ackoff, Systems thinking and thinking systems | https://doi.org/10.1002/sdr.4260100206 | grade A
- **[governance]** The joint-setting half of the question is grounded: the Mars Sample Return Campaign Concept Status and the MSPG2 Final Report document landing accuracy being traded jointly against retrieval mass, sample-tube reachability, and mobility, confirming Ackoff's point that the ellipse is one node inside an interacting architecture, not a free-standing investment input. Valuing the ellipse-shrinking lever in isolation, as the stated decision-relevance does, is the textbook suboptimization Ackoff names: optimizing one part extracted from the decision it feeds can degrade the whole. The candidate's framing presumes the ellipse is the binding constraint without expanding to the containing architecture that would tell you whether it is.
    - Mars Sample Return Campaign Concept Status, Acta Astronautica 2020 | https://doi.org/10.1016/j.actaastro.2020.06.026 | grade A
    - Final Report of the Mars Sample Return Science Planning Group 2 (MSPG2), Astrobiology 2021 | https://doi.org/10.1089/ast.2021.0121 | grade A
    - Ackoff, Towards a System of Systems Concepts | https://doi.org/10.1002/sdr.4260100206 | grade A
- **[measurement]** Christensen's RPV/capabilities framework holds that what an organization can and cannot do is set by its resources, processes, and values, not by the visible artifact; the hardware is the output of a process-and-values stock, so a binary 'inserted' dummy cannot, by construction, distinguish 'the technology shrank the ellipse' from 'the accumulating competence that produced the technology shrank it.' This is the operationalization defect the question targets and it is sound on Christensen's own terms.
    - Hall of Shoulders dossier: Clayton M. Christensen (RPV / capabilities framework, Resources-Processes-Values; Innovator's Dilemma 1997; Christensen & Rosenbloom 1995) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/christensen_c | grade A
- **[measurement]** A continuous capability proxy of the kind the question requests is in fact suppliable from the public record: JPL's mission-to-mission EDL reconstruction and flight-performance assessment is documented (e.g., the M2020 terrain-relative landing-accuracy assessment comparing flight performance against predicts), establishing the existence of a codified, cumulative reconstruction process that can be counted as a covariate. The proxy is therefore constructible; what cannot be retrieved is the candidate's regression result showing which regressor the contraction loads on.
    - Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts, NASA NTRS 20230007014 | https://ntrs.nasa.gov/citations/20230007014 | grade B
- **[identification]** Christensen's jobs-to-be-done test explicitly rejects segmenting by product category or satellite/technology class and demands segmentation by the job the capability is hired to do in its circumstance; a model that blocks the series by EDL-technology generation has, on this standard, not done JTBD. The demand-side rival the question raises (the ellipse shrank because the mission was hired for a harder targeting job) is the correct Christensen reframing and is a legitimate confounder of the supply-side technology block.
    - Hall of Shoulders dossier: Clayton M. Christensen (jobs-to-be-done test; segment by job not product category; Competing Against Luck) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/christensen_c | grade A
- **[identification]** The job-difficulty covariate the question asks for is suppliable from the public record: the Mars 2020 GN&C and Lander Vision System documentation establishes that the mission's targeting job was set by hazardous, science-driven terrain at Jezero that prior ellipses could not attempt, so site-demanded targeting difficulty is an observable, encodable circumstance, not a hypothetical.
    - Guidance, Navigation and Control for the Entry, Descent, and Landing of the Mars 2020 Mission, NASA NTRS 20190028435 | https://ntrs.nasa.gov/citations/20190028435 | grade B
    - The Lander Vision System for Mars 2020 Entry Descent and Landing, NASA NTRS 20190025658 | https://ntrs.nasa.gov/citations/20190025658 | grade B
- **[rival]** Christensen's standard that a theory is only as good as the rival it can rule out applies directly: with TRN flown on a single mission, the dummy's coefficient is identified by n=1 and is perfectly collinear in-sample with every other unique attribute of that mission, so within the candidate's own pre-registration logic the TRN attribution is non-falsifiable as coded. This is structurally correct, not a stylistic objection.
    - Hall of Shoulders dossier: Clayton M. Christensen (review lens: falsifiable questions; 'a theory is only as good as the rival it can rule out'; disruption as a falsifiable claim about value networks and RPV-structures) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/christensen_c | grade A
- **[rival]** The one-off confounds the question names are real and documented, which is why n=1 cannot be broken in-sample: Mars 2020 carried the first-flown Lander Vision System (an onboard map-relative localization capability) together with the mission's GN&C divert design, so 'TRN' is bundled with a denser onboard reference map and divert authority that are unique to that single mission. The discriminating observable therefore cannot come from within these data; it must come from a controlled decomposition (Monte Carlo ablations isolating LVS map-match vs. divert vs. computer, or a second TRN-flown mission) external to the one-mission record.
    - The Lander Vision System for Mars 2020 Entry Descent and Landing, NASA NTRS 20190025658 | https://ntrs.nasa.gov/citations/20190025658 | grade B
    - Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts, NASA NTRS 20230007014 | https://ntrs.nasa.gov/citations/20230007014 | grade B
- **[rival]** The job-distinction is corroborated by the named datasets: the Mars 2020 Safe Target Selection algorithm reconstruction states STS 'selected the safe landing target within a reachable region from the on-board Safe Targets Map' handed to MSL-heritage powered descent, i.e. its objective is hazard/reachability geometry, not a science-tolerance ellipse, so the future fixed-point/hazard-bounded job is a different job in kind from the science-ellipse fit. The candidate's own prospectus names the PDS achieved-miss-distance series as the secondary dependent variable (the realized-capability construct-validity proxy that 'shares none of the Monte Carlo dispersion modeling' of the design ellipse) and states divergence between the two series 'is itself reported as a finding rather than suppressed.' The named observable that would reveal a non-transferring (job-specific) coefficient is therefore already in the design: a divergence in the technology-block contraction between the design-ellipse series and the PDS achieved-miss-distance series. If a technology lever contracts the science-tolerance ellipse but does NOT proportionally contract the achieved point-miss, the design-vs-achieved divergence is the in-data signal that the gamma is hired for the science-ellipse job and will not recur on a hit-the-tube job.
    - Dutta et al., Mars 2020 Perseverance EDL GNC Safe Target Selection Reconstruction (NTRS 20230005646), abstract | https://ntrs.nasa.gov/citations/20230005646 | grade B
    - JPL_AUTONOMY_EDL_06 prospectus.md, Sec.5 Named Datasets (PDS achieved-miss-distance as secondary DV; divergence reported as finding) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/prospectus.md | grade C
- **[mechanism]** Partial, grounded observation (not a settlement): the named reconstruction lineage is real and single-institution. The NTRS Karlgaard-family EDL trajectory/atmosphere reconstructions exist as a sustained JPL/LaRC methodological line across missions (e.g. MSL EDL Trajectory and Atmosphere Reconstruction, NTRS 20130010087; MEDLI2 reconstruction series, NTRS 20210024320; InSight EDL reconstruction, NTRS 20200003204). The prospectus confirms the entire in-sample population is a single value network: US-led Mars surface missions executed by the same EDL institution, with 'institutional learning' explicitly named in the prospectus as an unobserved monotone confound that 'could load on the technology indicators and masquerade as a technology effect.' The model as specified (ln EllipseArea ~ Sequence + GuidedEntry + RangeTrigger + TRN + ApproachAccuracy) contains NO institutional-continuity regressor, so by construction it cannot separate a portable technology effect from a closed-value-network artifact.
    - NTRS, MSL EDL Trajectory and Atmosphere Reconstruction (Karlgaard lineage), citation 20130010087 | https://ntrs.nasa.gov/citations/20130010087 | grade B
    - NTRS, MEDLI2 Trajectory/Aerodynamics/Atmosphere reconstruction, citation 20210024320 | https://ntrs.nasa.gov/citations/20210024320 | grade B
    - JPL_AUTONOMY_EDL_06 prospectus.md, Sec.7 Internal validity (institutional learning named as unobserved monotone confound; no institutional regressor in the model) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/prospectus.md | grade C
- **[governance]** Settled from the prospectus text. The document DOES contain scope conditions, but on the DATA/POPULATION and CALENDAR-TIME dimensions, not on the JOB-CLASS dimension Christensen presses. Sec.6 Scope: 'The fit is on robotic Mars landings only. Lunar precision-landing navigation and human-class EDL appear as forward and out-of-sample reference points, not in-sample evidence... the work does not address landing safety, hazard density, or scientific yield except insofar as the ellipse governs which sites are reachable.' Sec.7 External validity: human-scale precision-landing assessments are 'treated as the forward edge of the capability, not as observations the curve is fit to,' with 'Confidence beyond the estimation population: low, and the work says so.' Sec.6 Significance explicitly invites MSR and Human-Mars planners to read a gamma as 'an estimate of the proportional ellipse reduction that flying that technology buys... exactly the input a requirement-setting trade needs.' Crucially, NO scope condition reframes the dependent variable from science-tolerance ellipse to a fixed-point/hazard-bounded error, and the Kuznetsian honest-extrapolation discipline (InSight counterfactual) is stated as separating 'a genuine secular improvement trend from a one-time level shift' i.e. it polices CALENDAR TIME (transient vs secular), not JOB CLASS. There is therefore no coefficient annotation, job-class boundary, or value-network-continuity precondition that makes a fixed-point-rendezvous planner stop and recompute rather than extrapolate exp(gamma)-1. The governance guardrail against the precise misuse the significance section invites is absent on the job-class dimension; the out-of-sample-honesty claim is installed for calendar time but not for job class.
    - JPL_AUTONOMY_EDL_06 prospectus.md, Sec.6 (Significance, Scope and delimitations) and Sec.7 (External validity) | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/prospectus.md | grade C
    - christensen_c thinker dossier (JTBD test: segment by the job in its circumstance, not by product/agency class; value network must be modeled, not just the artifact) | file:///D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/christensen_c/ | grade A
- **[identification]** The CRITIQUE is well-grounded as a behavioral-methods objection: WYSIATI (analyst builds a coherent story from a known monotone outcome and is overconfident in it), the law of small numbers / availability (one salient case overweighted, base rate ignored), and researcher degrees of freedom in a tiny sample are documented, named biases. These warrant exactly the three falsifiable demands raised (blind coding, base-rate residual distribution, multiverse grid). However, the ANSWERS demanded are empirical artifacts the candidate has not produced.
    - Tversky & Kahneman, Judgment under Uncertainty: Heuristics and Biases, Science 185(4157):1124 (1974) | https://doi.org/10.1126/science.185.4157.1124 | grade A
    - Kahneman hall-of-shoulders dossier (hos-kahneman): WYSIATI / coherent-story-from-available-evidence-and-overconfidence; base-rates-vs-salience; reference-class forecasting | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kahneman/ | grade A
    - Flyvbjerg, Curbing Optimism Bias and Strategic Misrepresentation in Planning: Reference Class Forecasting in Practice, European Planning Studies (2007) | https://doi.org/10.1080/09654310701747936 | grade A
- **[identification]** The hazard named in Q1 is real and is the heuristics-and-biases program's own diagnosis: hindsight/outcome knowledge primes a coherent inside-view ordering, and salient outcomes anchor judgment (the Challenger/Columbia pattern, where base-rate information was available and discounted while the salient record of successes anchored confidence). The corrective the candidate proposes (blind, pre-outcome coding from documents dated before landing) is exactly the Kahneman-standard outside-view procedure. HOWEVER, the contamination test itself cannot be SETTLED from retrieval: the pre-flight TechPort manifests and as-designed technology records needed to re-code blind are not present in any queried corpus (AMOS, ACTA, Space Economy all returned zero hits on Mars-EDL technology/manifest queries). The methodological critique is grounded; the record-settling result is absent, so no verdict on actual contamination can be asserted.
    - Kahneman dossier (Hall of Shoulders), citing Tversky & Kahneman, 'Judgment under Uncertainty: Heuristics and Biases', Science 1974, and the Challenger/Columbia salience-anchoring analysis | https://doi.org/10.1126/science.185.4157.1124 | grade A
- **[measurement]** The normative standard the candidate must meet is established and is exactly Kuznets's national-accounts measurement discipline: any reported figure must carry an explicit definition of coverage, a stated valuation/normalization convention, and a netting rule such that 'two reasonable analysts using your method would land on' the same number; absent that, the construct is undefined. Applied here, the named per-mission primary records DO exist and are individually re-derivable (Karlgaard et al., MSL EDL Trajectory and Atmosphere Reconstruction, NTRS 20130010087, 2013; Spencer et al., Mars Pathfinder Entry, Descent, and Landing Reconstruction, doi 10.2514/2.3478; Chen et al., Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts, NTRS 20230007014). What retrieval does NOT supply is any source defining a cross-mission deflator that equalizes the probability content of a Viking-era 'three-sigma' against a modern Monte Carlo 'three-sigma', and the Viking-specific reconstruction document stating its own dispersion convention did not surface (Crossref returned only USGS Viking-site topographic mosaics, e.g. doi 10.3133/i1060, not an ellipse-construction record). Therefore the candidate can ground each ln-area locally but cannot, from the assembled corpus, demonstrate the inter-era commensurability the dependent variable presupposes.
    - kuznets dossier (Hall of Shoulders, hos-kuznets), measurement-boundary challenge: explicit coverage definition, valuation convention, netting rule, two-analyst replicability test | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kuznets | grade A
    - Landefeld, Seskin & Fraumeni, 'Taking the Pulse of the Economy: Measuring GDP', J. Econ. Perspectives 22(2), 2008, conventions embedded in a measurement boundary, treatment of quality change | https://doi.org/10.1257/jep.22.2.193 | grade A
    - Karlgaard, Kutty, Schoenenberger, Shidner, 'Mars Science Laboratory Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction', NTRS 20130010087, 2013 | https://ntrs.nasa.gov/citations/20130010087 | grade B
    - Spencer et al., 'Mars Pathfinder Entry, Descent, and Landing Reconstruction' | https://doi.org/10.2514/2.3478 | grade A
    - Chen, Casoliva, Montgomery, Serricchio, 'Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts', NTRS 20230007014, 2021 | https://ntrs.nasa.gov/citations/20230007014 | grade B
- **[identification]** The demand is well-posed and grounded: Kuznets's own discipline requires separating a measurement/convention movement from the substantive movement before any coefficient is read, and treats a change in construction method as a candidate confound to be measured and netted, not assumed away. An achieved-miss-distance cross-check is a legitimate independent construct because at least one named mission publishes flight-performance-versus-predicts landing accuracy that shares none of the dispersion-percentile conventions of the predicted ellipse (Chen et al., 'Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts', NTRS 20230007014; cf. M2020 Lander Vision System flight-performance record, NTRS 20230006986). However, retrieval supplies only this achieved-miss series for the most recent mission; it does NOT establish that an analogous achieved-miss series exists with matching fidelity for Viking and Pathfinder, which is precisely what is needed to estimate the partial (collinear) drift across the full span. So the candidate can ground the netting requirement and the cross-check in principle, but the corpus does not show the cross-check has the cross-mission coverage to separate partial drift from gross divergence.
    - kuznets dossier (hos-kuznets), transient vs secular: 'distinguish the transient (a one-time cost-step) from the secular (a sustained productivity trend)'; treat a movement as a level shift/cycle until a long decomposed series proves otherwise | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kuznets | grade A
    - kuznets dossier (hos-kuznets), measurement-boundary challenge: netting rule and two-analyst replicability as precondition for any reported figure | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kuznets | grade A
    - Chen, Casoliva, Montgomery, Serricchio, 'Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts', NTRS 20230007014, 2021, independent achieved-vs-predicted landing-accuracy series | https://ntrs.nasa.gov/citations/20230007014 | grade B
    - Zheng, Shamah, Schaffner, Ravine, 'Mars 2020 Lander Vision System Flight Performance', NTRS 20230006986, 2022 | https://ntrs.nasa.gov/citations/20230006986 | grade B
- **[empirics]** On Kuznets's own methodological commitment the burden falls squarely on the candidate, and the discrete-level-shift alternative is the prior to beat, not the null to dismiss: Kuznets built his reputation on long series and the recognition that business-cycle, level-shift, and secular movements are different animals; he would treat a few-point doubling 'as evidence of a cycle or a level shift until a long enough series proves it secular,' and would require the candidate to 'show how long a comparable series' was needed to separate them. A single ~45-year epoch of ~11 bespoke vehicles, with the contraction clustered at named discrete technology insertions (Pathfinder airbag heritage, Spencer et al., doi 10.2514/2.3478; MSL guided entry, Karlgaard et al., NTRS 20130010087; Mars 2020 terrain-relative navigation, Chen et al., NTRS 20230007014, and LVS flight performance, NTRS 20230006986), is the textbook case where a step-function and a log-linear curve are observationally near-equivalent at this resolution. The grounded conclusion: the constant-proportional-rate ('learning rate') interpretation is not identified against the level-shift alternative on this sample, and the candidate must either present the model-comparison fit statistic favoring log-linearity or concede the contraction is a sequence of level shifts that merely looks log-linear.
    - kuznets dossier (hos-kuznets), 'treat a five-year doubling as evidence of a cycle or a level shift until a long enough series proves it secular'; 'show how long a comparable series you needed to distinguish them' | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kuznets | grade A
    - kuznets dossier (hos-kuznets), long-run secular series and the danger of short-window extrapolation from a few years/points | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kuznets | grade A
    - Kuznets (with Jenks), 'Shares of Upper Income Groups in Income and Savings', NBER, exemplar: build comparable, decomposed long time series before theorizing a trend | https://doi.org/10.2307/2227904 | grade B
    - Spencer et al., 'Mars Pathfinder Entry, Descent, and Landing Reconstruction', Pathfinder airbag-heritage insertion point | https://doi.org/10.2514/2.3478 | grade A
    - Karlgaard et al., MSL EDL Trajectory and Atmosphere Reconstruction, NTRS 20130010087, MSL guided-entry insertion point | https://ntrs.nasa.gov/citations/20130010087 | grade B
    - Chen et al., Assessment of M2020 Terrain Relative Landing Accuracy, NTRS 20230007014, Mars 2020 TRN insertion point | https://ntrs.nasa.gov/citations/20230007014 | grade B
- **[empirics]** GROUNDED in method, not in the candidate's data. The Kuznetsian discipline supplies the diagnostic in principle and the binding inference rule: distinguishing a genuine secular trend from a transient level shift or cycle requires a long enough series, and an extrapolation is 'not yet warranted' when the secular movement is indistinguishable from a level shift within the observation window. Translated to the eleven-point design-ellipse series, the operative within-sample saturation diagnostic is exactly the one the question names: a monotone decline in successive log-decrements (second difference of ln-area across guidance generations) and/or systematic curvature of the residuals against sequence position. On the inference rule, the kuznets dossier is explicit that a slope fit on a short, possibly-flattening window cannot be extrapolated forward, and the structural-change literature (McMillan & Rodrik 2011) reinforces that an aggregate rate decomposes into level-conditional components whose sign and magnitude depend on the regime already reached. Consequence: if the eleven points cannot statistically separate steep-regime from floor-approach, the learning rate is a regime-local, level-conditional quantity conditioned on the level already attained, NOT a structural constant licensed to price the next vehicle. Reporting it as a transportable price is unwarranted on the candidate's own bounded-mechanism premise.
    - kuznets dossier (hall_of_shoulders, hos-kuznets) - secular-vs-transient extrapolation and short-window caution | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kuznets/ (search_brain.py --brain kuznets) | grade B
    - McMillan & Rodrik, Globalization, Structural Change and Productivity Growth, NBER WP No. 17143, 2011 (aggregate rate decomposes into level-conditional within- and between-components) | https://doi.org/10.3386/w17143 | grade A
- **[identification]** GROUNDED in method, refused on the data-specific verdict. The Kuznetsian apparatus settles the LICENSING PRINCIPLE the question demands: forward extrapolation off a fitted rate is warranted only when the realized (out-of-convention) series and the design series show the SAME proportional decrements at the recent end of the window; if the realized-capability series already flattens relative to the design ellipse, the design-ellipse rate over-predicts the marginal gain available to a vehicle near the floor, and no forward project may read a price off the design curve. This is the direct application of the dossier's rule that a secular rate indistinguishable from a level shift within the window is not yet extrapolation-warranted, and that long-run inference must be built on long series, not short post-event windows. The licensing test is therefore correctly specified: require recent-end agreement between the PDS realized-miss-distance decrements and the design-ellipse decrements at the last two guidance generations; absent that agreement, refuse forward pricing.
    - kuznets dossier (hall_of_shoulders, hos-kuznets) - 'if your secular growth is indistinguishable from a level shift or a cycle within your window, your extrapolation is not yet warranted'; danger of short-window extrapolation | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/kuznets/ (search_brain.py --brain kuznets) | grade B
- **[identification]** The candidate cannot fully separate the lever from the directed institution, and concedes as much in the design. In the entrepreneurial-state / mission-oriented frame the directed program is not a nuisance trend but the precondition that converts a propositional base into a deployable prescriptive artifact (Mazzucato 2018; Kattel & Mazzucato 2018), so a binary on first-flight and a Sequence index on the program are confounded by construction. The prospectus itself flags this as the dominant internal-validity threat: 'any unobserved factor that improved monotonically across the era (modeling fidelity, atmospheric knowledge, flight-computer throughput, institutional learning) could load on the technology indicators and masquerade as a technology effect' (sec. 7) and 'Three things changed together, and monotonically' (sec. 2). The single observation the design offers as the separator is the InSight counterfactual: a late (2018) mission deliberately flown with a Phoenix-heritage ballistic entry into a large Elysium ellipse, holding the calendar at the modern era while turning the guidance levers off (sec. 4, sec. 7). But InSight discriminates technology-generation from calendar date only; it does NOT decouple the lever from the directed program, because InSight was itself produced inside the same sustained NASA/JPL Mars institution. No observation in the nine-to-eleven census flies the levers OUTSIDE that institution, so the data cannot separate 'the hardware' from 'the directed program that made the hardware trustworthy.' The honest answer the candidate can give: InSight separates artifact-from-time, not artifact-from-institution, and the latter separation is not identified.
    - Mazzucato (2018), 'Mission-oriented innovation policies: challenges and opportunities', Industrial and Corporate Change | https://doi.org/10.1093/icc/dty034 | grade A
    - Kattel & Mazzucato (2018), 'Mission-oriented innovation policy and dynamic capabilities in the public sector', Industrial and Corporate Change | https://doi.org/10.1093/icc/dty032 | grade A
    - JPL_AUTONOMY_EDL_06 prospectus, secs. 2, 4, 7 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/prospectus.md | grade C
- **[measurement]** Nothing in the decision artifact carries the institutional precondition; the prospectus leaves it in prose and explicitly excludes the layer that would carry it. The gamma coefficient is defined purely as 'the proportional ellipse reduction that flying that technology buys' (sec. 6, sec. 8), with no scope condition or provenance band attached to the COEFFICIENT itself. The provenance bands the design does carry are attached only to the dependent variable's measurement convention (Viking-era simulation conventions, sigma-level normalization per the Kuznetsian boundary discipline, sec. 5, sec. 7), not to the institutional preconditions of the independent variable. Decisively, the candidate writes the institutional precondition OUT of the decision artifact on purpose: 'The architecture-traceability layer of the underlying argument framework is deliberately out of scope ... the decision relevance is carried in plain prose rather than as an architecture artifact' (sec. 6). In Mazzucato's terms this is the mispricing risk: a lever's measured value was realized inside a public lead-investor program that bore the pre-commercial risk and built the propositional base (Mazzucato & Semieniuk 2017; Edler & Fagerberg 2017's mission-oriented vs market-failure contrast), so reading gamma as the standalone price of TRN prices the artifact while silently assuming the institution is free. The candidate's defensible concession: as packaged, gamma can be read as the in-isolation price, and only the surrounding prose, not the coefficient, warns against it; a planner who lifts the number out of the prose would misprice a lever bought outside the directed program.
    - Mazzucato & Semieniuk (2017), 'Public financing of innovation: new questions', Oxford Review of Economic Policy | https://doi.org/10.1093/oxrep/grw036 | grade A
    - Edler & Fagerberg (2017), 'Innovation policy: what, why, and how', Oxford Review of Economic Policy | https://doi.org/10.1093/oxrep/grx001 | grade A
    - JPL_AUTONOMY_EDL_06 prospectus, secs. 5, 6, 8 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/prospectus.md | grade C
- **[rival]** A directed-program proxy is a genuine rival cause that the current H0/H1 dichotomy cannot adjudicate, because both hypotheses condition only on Sequence and the three first-flight binaries (sec. 1, sec. 4); neither contains a program-direction term, so the test is silent on institution-vs-artifact. Mazzucato's diagnostic is exactly this 'mission or procurement?' question, falsifiable against funding-history and procurement records, and her Apollo-as-archetype reading makes a sustained directional program the candidate binding variable (Mazzucato 2018; Kattel & Mazzucato 2018). Candidate proxies do plausibly exist in the named sources: mission cadence is directly derivable from the census launch/landing dates the design already encodes; sustained Mars-program funding by fiscal year is in public NASA budget records; retained in-house EDL engineering capacity is partially recoverable from NTRS authorship continuity across the Karlgaard-lineage reconstruction reports the design already cites (sec. 5). So the candidate cannot claim no proxy is available. The defensible position is the degrees-of-freedom one the prospectus itself states: with nine-to-eleven observations and 'five candidate slope parameters' already straining exact/permutation inference (sec. 7), a fourth structural term is not separately estimable and would be near-collinear with Sequence, so a program-direction covariate is excludable on statistical-power grounds, NOT on the grounds that the institution is not a rival cause. The candidate must therefore concede the omitted-variable rather than deny it: the directed program is a live rival the design cannot adjudicate at this n, and the honest move is to downgrade the gamma reading from 'price of the lever' to 'price of the lever as produced inside the directed program,' treating the institution as a fixed, unestimated precondition rather than a controlled-for confound.
    - Mazzucato (2018), 'Mission-oriented innovation policies: challenges and opportunities', Industrial and Corporate Change | https://doi.org/10.1093/icc/dty034 | grade A
    - Kattel & Mazzucato (2018), 'Mission-oriented innovation policy and dynamic capabilities in the public sector', Industrial and Corporate Change | https://doi.org/10.1093/icc/dty032 | grade A
    - JPL_AUTONOMY_EDL_06 prospectus, secs. 1, 4, 5, 7 | file:///D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/prospectus.md | grade C
- **[measurement]** The critique is conceded as well-founded and the proposed test is the correct one. Mazzucato's market-shaping/market-fixing distinction is precisely the diagnosis: coding the contraction to inserted artifacts is a market-fixing ontology that treats the public role as discrete corrections, whereas the entrepreneurial-state frame holds that the load-bearing causal object is the retained public dynamic capability that generated those artifacts (Mazzucato & Semieniuk 2017; Robinson & Mazzucato 2018). Kattel & Mazzucato (2018) formalize 'dynamic capabilities in the public sector' as the agency's internal capacity to design, learn, and adapt, and warn of the hollowing-out risk when that capacity is not measured. The candidate's three-dummy specification has no term for this stock, so by construction it cannot attribute the ellipse contraction to it; the artifacts are collinear proxies for the standing JPL EDL/GNC capability and will absorb its coefficient. The records to build the stock measure exist and are non-fabricated: NTRS publishes the Mars 2020 LVS and TRN flight-performance reconstruction lineage with named authors (e.g., NTRS 20190025658, 20230006986, 20230007014, 20160009627), from which author-recurrence across the reconstruction lineage and a capability-continuity index can be coded. The honest finding the candidate must report: until ln(EllipseArea) is regressed on a continuous public-capability stock alongside the binary block, the design cannot distinguish lever from stock, and the prudent prior (per Kattel & Mazzucato) is that the dummies are partly proxying retained public capability rather than standalone insertions.
    - Kattel & Mazzucato, 'Mission-oriented innovation policy and dynamic capabilities in the public sector,' Industrial and Corporate Change | https://doi.org/10.1093/icc/dty032 | grade A
    - Robinson & Mazzucato, market-shaping vs market-fixing analysis of NASA/ESA (Mazzucato dossier, graded A) | https://doi.org/10.1093/icc/dty034 | grade A
    - Mazzucato & Semieniuk, public investment and innovation capability | https://doi.org/10.1093/oxrep/grw036 | grade A
    - NASA Technical Reports Server: Lander Vision System for Mars 2020 EDL (20190025658); M2020 LVS Flight Performance (20230006986); Assessment of M2020 TRN Landing Accuracy (20230007014); Real-time TRN Test Results (20160009627) | https://ntrs.nasa.gov/citations/20230007014 | grade C
- **[governance]** Conceded in full: a learning rate is a scalar on rate and is silent on direction and spillover, which are exactly the properties Mazzucato makes the test of a genuine mission versus a procurement line item. A genuine mission is directional, capability-building, and spillover-generating (Kattel & Mazzucato 2018; the dossier's Apollo-as-archetype reading); a re-badged procurement buys an output for one vehicle and generates none of these. The discriminating observable is cross-program technology infusion: in TechPort, a directional mission lever shows technology-transfer/infusion edges from the originating project into downstream, non-originating projects (and ideally other centers), whereas a procurement line item shows a single project node with no outward infusion. The same signature is visible in the public reconstruction record: TRN/LVS heritage demonstrably propagates from the Mars 2020 vehicle into a reusable reference-map and flight-validation pipeline and into precision-landing testbeds beyond the originating EDL (NTRS 20160009627 relevant-environment TRN tests; 20170009437 COBALT ALHAT closed-loop terrestrial flights; 20220001565 LiDAR-inertial precision landing), which is positive evidence of spillover for the TRN lever specifically. The candidate's design, however, has not coded these infusion edges, so the learning rate as built cannot certify any of the three levers as directional rather than path-dependent. The governance warning the candidate must surface: a coefficient that quantifies rate but not direction can rationalize concentrating spend on the same three levers inside the same prime, which is the path-dependence risk, not market-shaping of a broader EDL supplier base; only the TechPort infusion observable, once coded, lets the coefficient earn the 'directional mission lever' label.
    - Kattel & Mazzucato, mission-oriented innovation policy and public-sector dynamic capabilities (Apollo-as-archetype mission reading) | https://doi.org/10.1093/icc/dty032 | grade A
    - Edler & Fagerberg, innovation policy: what, why, and how (cited in Mazzucato dossier, graded A) | https://doi.org/10.1093/oxrep/grx001 | grade A
    - NTRS: Real-time TRN Test Results from a Relevant Environment (20160009627); COBALT GN&C payload ALHAT closed-loop flights (20170009437); LiDAR-Inertial precision landing (20220001565) | https://ntrs.nasa.gov/citations/20170009437 | grade C
- **[economics]** The distributional objection is correct and is the core risk-reward question of the entrepreneurial-state frame: codified prescriptive knowledge is a public good only if the public retains it, otherwise early public risk-bearing is paired with privatized value capture (Mazzucato & Li 2021 on retained IP/equity/royalty/reinvestment as the condition for public upside; the dossier's risk-reward-distribution diagnostic). H1's 'extensible' claim is therefore not self-justifying from Mokyr; it is conditional on provenance, and the design as built never runs the provenance audit. The audit is feasible from existing records: NTRS author affiliations across the Karlgaard reconstruction lineage and the LVS reference-map pipeline reveal whether codification authors are JPL/NASA civil servants (retained-public) or prime-contractor staff (concentrated-private), and TechPort technology-transfer/infusion records reveal whether each lever can be re-instantiated by another center without re-procuring the prime. On the partial public evidence retrieved, the LVS/TRN codification appears predominantly authored and curated inside the public agency and reused across NASA testbeds (NTRS 20190025658, 20230006986, 20230007014), which would support 'extensible-as-retained-public-asset' for the TRN lever; but the candidate has run no affiliation classification for GuidedEntry or RangeTrigger, so the prescriptive-knowledge premise survives only for the lever whose codification provenance has actually been audited and demonstrated to stay inside the public agency. For the other two, 'extensible' is currently an assumption, not a finding, and could mask a capability concentrated in a single prime and transferable only by re-procurement, which would invert the public-good claim into a privatized-value-capture outcome.
    - Mazzucato & Li, 'The Entrepreneurial State and Public Options' (risk-reward distribution / retained-IP condition, Mazzucato dossier graded A) | https://doi.org/10.1017/9781108767552.003 | grade A
    - Kattel & Mazzucato, dynamic capabilities in the public sector | https://doi.org/10.1093/icc/dty032 | grade A
    - NTRS: Lander Vision System for Mars 2020 EDL (20190025658); M2020 LVS Flight Performance (20230006986); Assessment of M2020 TRN Landing Accuracy (20230007014) | https://ntrs.nasa.gov/citations/20230006986 | grade C
- **[identification]** A propositional-knowledge metric independent of the ellipse series IS specifiable from the dated design literature: the depth and codification age of the theory each lever rests on. Guided entry inherits Apollo entry-guidance theory codified since the 1960s and re-flown as MSL Entry Guidance (NTRS 20110003649, 2011), giving it a wide, mature propositional base. TRN's computer-vision map-matching theory matured visibly later, first demonstrated in a relevant-environment flight test only in 2015 (NTRS 20160009627). Mokyr's framework (Gifts of Athena, 2002) holds that a prescriptive technique resting on a wider, deeper propositional base is the durable macro-invention, so the ranking CAN be anchored to theory-maturity dates that precede and are independent of the ellipse outcome. Where the candidate has NOT done this, the macro/micro label is an assumption imported into the covariate design, not a result.
    - Mokyr, J. (2002). The Gifts of Athena: Historical Origins of the Knowledge Economy. Princeton University Press. | https://doi.org/10.1515/9781400829439 | grade A
    - NASA Technical Reports Server, Mars Science Laboratory Entry Guidance (2011). | https://ntrs.nasa.gov/citations/20110003649 | grade C
    - NASA Technical Reports Server, Real-time Terrain Relative Navigation Test Results from a Relevant Environment for Mars Landing (2015). | https://ntrs.nasa.gov/citations/20160009627 | grade C
- **[measurement]** An extensibility indicator independent of the single level-shift IS available and partly documented for TRN: the dated trajectory of cross-mission reuse-with-improvement. TRN moved from a 2015 relevant-environment flight test (NTRS 20160009627) to operational Mars 2020 use, where its landing accuracy was independently assessed against pre-flight predicts (Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts, NTRS 20230007014). A lever whose accuracy gain is re-validated and improved across successive vehicles, with flight-versus-predict error shrinking, exhibits the Mokyrian self-correcting/extensible signature (Gifts of Athena, 2002); a lever that produced one drop and never re-flew with improved performance does not. So the measurement that distinguishes deepening from a one-off staircase is the slope of the predict-versus-flight residual across reuse instances, not the size of any single insertion drop.
    - Mokyr, J. (2002). The Gifts of Athena: Historical Origins of the Knowledge Economy. Princeton University Press. | https://doi.org/10.1515/9781400829439 | grade A
    - NASA Technical Reports Server, Assessment of M2020 Terrain Relative Landing Accuracy: Flight Performance vs Predicts (2021). | https://ntrs.nasa.gov/citations/20230007014 | grade C
    - NASA Technical Reports Server, Real-time Terrain Relative Navigation Test Results from a Relevant Environment for Mars Landing (2015). | https://ntrs.nasa.gov/citations/20160009627 | grade C
- **[rival]** The rival cause is well-founded in Mokyr's framework and has an empirical analogue. Mokyr holds that the binding constraint on whether a technique is improvable is the access cost to its propositional base, which is partly tacit and institutionally localized at the producing institution (Gifts of Athena, 2002); the canonical empirical demonstration that useful-knowledge spillovers are geographically and institutionally localized is Jaffe, Trajtenberg and Henderson's patent-citation study (QJE 1993). Because the technology dummies encode WHEN a lever flew, not the access cost to the modeling capability that produced it, and because the InSight control shares the same JPL reconstruction apparatus, the technology coefficients and the InSight counterfactual cannot isolate a knowledge-access-cost effect from a technology-insertion effect. The dated evidence that WOULD separate them, mirroring the candidate's own NTRS-dated lever timeline, is the codification date of each reconstruction/modeling method (when it became documented and reusable across teams) set against the first-flight date of each guidance lever: if accuracy improvements track method-codification dates rather than lever-flight dates, the effect is access-cost, not insertion. TRN supplies a worked instance of such dating (method maturation 2015, NTRS 20160009627; reuse/assessment at Mars 2020, NTRS 20230007014).
    - Mokyr, J. (2002). The Gifts of Athena: Historical Origins of the Knowledge Economy. Princeton University Press. | https://doi.org/10.1515/9781400829439 | grade A
    - Jaffe, A. B., Trajtenberg, M., & Henderson, R. (1993). Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations. Quarterly Journal of Economics, 108(3), 577-598. | https://doi.org/10.2307/2118401 | grade A
    - NASA Technical Reports Server, Real-time Terrain Relative Navigation Test Results from a Relevant Environment for Mars Landing (2015); Assessment of M2020 Terrain Relative Landing Accuracy (2021). | https://ntrs.nasa.gov/citations/20160009627 | grade C
- **[measurement]** The design as written cannot adjudicate this: every lever is coded ONLY as a 'first-flew' binary (GuidedEntry, RangeTrigger, TRN), and Section 4.3.3 explicitly states the binary coding 'deliberately discards within-technology variation' and tests 'the INSERTION of a discrete capability... not the gradual refinement of that capability', defending the choice as sample-constrained (9-11 obs exhaust degrees of freedom). No propositional-depth covariate is built. Therefore the macro-vs-incremental claim rests on the RELATIVE MAGNITUDES of the gamma dummies (gamma_1 guided entry and gamma_3 TRN expected to dominate gamma_2 range trigger, Sec 2.5/2.6), which carry zero information about the depth of the underlying base. The materials to build the proposed proxy DO exist in the candidate's own corpus and are separable from the dummy: the Karlgaard statistical-reconstruction lineage (MSL NTRS 20130010087; InSight NTRS 20200002910; MEDLI2/Mars2020 era JSR 2022) supplies a datable, countable series of antecedent flight-validated atmosphere/aerodynamic reconstructions, and Johnson 2015 documents a flight-validated 'relevant environment' TRN test that PRECEDES the Mars 2020 first flight, i.e. an antecedent maturity event distinct from the insertion date. Because the candidate did not enter such a covariate, and because guided-entry/TRN propositional depth is perfectly collinear with their insertion mission in this 9-11-point census (the levers were inserted monotonically, Sec 4.4.4), the dummy and any depth proxy cannot be separately identified in-sample; the design admits at most a 'signed, ordered, counterfactual-surviving attribution', not a depth-vs-artifact test. On Mokyr's own criterion (the dossier: a macro-invention earns the label only if it rests on a WIDENING propositional base that makes the technique extensible and self-correcting), the 'macro-invention' designation is asserted from the relative dummy magnitudes, not measured: nothing in the present dataset licenses calling guided entry or TRN a macro-invention rather than a datable insertion artifact until the continuous antecedent-maturity proxy is constructed and shown to carry the contraction.
    - JPL_AUTONOMY_EDL_06 dissertation Sec 4.3.3 (measurement table + third operationalization defense) | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade C
    - JPL_AUTONOMY_EDL_06 dissertation Sec 2.6 conceptual model / augmented specification | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade C
    - Hall of Shoulders Mokyr dossier (Gifts of Athena, 2002), citation-grounded | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/mokyr/ | grade B
    - Karlgaard, Kutty, Schoenenberger, Shidner, 'MSL EDL Trajectory and Atmosphere Reconstruction' (NTRS) | https://ntrs.nasa.gov/citations/20130010087 | grade B
    - Karlgaard, Korzun, Schoenenberger, Bonfiglio, 'Mars InSight EDL Trajectory and Atmosphere Reconstruction' (NTRS) | https://ntrs.nasa.gov/citations/20200002910 | grade B
    - Johnson, A. E., et al., 'Real-Time Terrain Relative Navigation Test Results from a Relevant Environment for Mars Landing', AIAA 2015 | https://doi.org/10.2514/6.2015-0851 | grade A
    - JPL_AUTONOMY_EDL_06 dissertation Sec 4.4.4 coverage limitations | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade C
- **[identification]** The candidate identifies the confound but treats it as a Kuznetsian COMPARABILITY (boundary) problem, not as a measurement-codification COVARIATE, and so does not run the separation test the question asks for. Section 4.4.2 names the exact mechanism: the atmosphere model assumed in the Monte Carlo dispersion 'improved markedly across the era, in part because of the MEDLI and MEDLI2 instrumentation', and Section 4.4.3 states explicitly that 'later missions are reconstructed from richer flight data (MEDLI, MEDLI2) than earlier ones, so the dispersion characterizations of later missions are tighter and better-founded... measurement quality is correlated with mission sequence, the very axis the regression uses.' That is precisely the propositional-capital-in-the-instrument confound: the Karlgaard reconstruction lineage (NTRS 20130010087 MSL, 20200002910 InSight, JSR 2022 MEDLI2-era) is codified modeling capability that improved independently of the flight hardware. The candidate's chosen remedy is NOT to enter reconstruction-method generation as its own covariate; it is (i) provenance recording per cell, (ii) conversion to common three-sigma area, and (iii) the achieved-miss-distance secondary DV from PDS as an independent construct check. Critically, the candidate concedes the residual variance is heteroskedastic across sequence and that this correlation could load on the technology axis. The proposed clean test, hold reconstruction vintage fixed (compare only ellipses produced under one Karlgaard-family methodology generation) OR enter reconstruction-method generation as a covariate from the NTRS publication record, is feasible from the corpus (the NTRS reconstruction vintages are individually datable) but is NOT in the pre-registered specification (Sec 2.6 / 5.10), which contains Sequence, three technology dummies, and ApproachAccuracy only. Consequently the present design cannot state whether the technology block survives once measurement-codification capability is held constant: part of the attributed 'technology' contraction may belong to the cheapening of the reconstruction knowledge used to construct the metric, and the only partial defense offered, the PDS achieved-miss-distance re-fit (Re-fit one, Sec 5.8), addresses convention drift in the DV but is itself sparse, divert-conflated for Mars 2020, and subject to cartographic-frame drift, so it does not isolate reconstruction-method vintage. The honest reading: the confound is acknowledged and bounded by provenance discipline, but it is not differenced out, and the technology coefficients are therefore not shown to be net of reconstruction-codification improvement.
    - JPL_AUTONOMY_EDL_06 dissertation Sec 4.4.2 comparability problem | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade C
    - JPL_AUTONOMY_EDL_06 dissertation Sec 4.4.3 known biases by dataset | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade C
    - JPL_AUTONOMY_EDL_06 dissertation Sec 2.6 augmented specification + Sec 5.10 pre-registration | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade C
    - Karlgaard et al., 'MSL EDL Trajectory and Atmosphere Reconstruction' (NTRS) | https://ntrs.nasa.gov/citations/20130010087 | grade B
    - Karlgaard et al., 'Mars InSight EDL Trajectory and Atmosphere Reconstruction' (NTRS) | https://ntrs.nasa.gov/citations/20200002910 | grade B
    - Karlgaard, C. D., et al., 'Mars EDL Instrumentation 2 Trajectory, Aerodynamics, and Atmosphere Reconstruction', J. Spacecraft & Rockets 2022 | https://doi.org/10.2514/1.a35440 | grade A
    - JPL_AUTONOMY_EDL_06 dissertation Sec 5.8 Re-fit one + Sec 4.4.3 | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade C
- **[identification]** The structural-role question is dispositive and resolvable on the DAG alone: under Pearl's back-door criterion a variable that is a descendant of treatment (a mediator) must be excluded from the adjustment set, and conditioning on it biases the technology (gamma) coefficients toward zero by blocking part of the causal path; a true confounder (common cause upstream of both treatment and outcome) must be included. The two roles imply opposite adjustment decisions, so the variable's role cannot be left undeclared. Whether delta is separately identified from the technology block is a property of the design matrix, not of priors: if guidance generation and approach accuracy are collinear across the nine points, delta is not separately identified and its insignificance is uninformative either way. This is the exact failure mode Pearl flags as conditioning that 'could manufacture' the signature read as confirmation. The candidate must (a) assign ApproachAccuracy a role on an explicit DAG and (b) report the within-sample collinearity / variance-inflation between ApproachAccuracy and the technology indicators before reading delta's insignificance as support for H1.
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) [pearl dossier, Hall of Shoulders brain] | https://doi.org/10.1017/cbo9780511803161 | grade A
    - Drawing Credible Directed Acyclic Graphs for Causal Inference (2025) [retrieved via pearl brain] | https://doi.org/10.31234/osf.io/u4yta_v4 | grade B
- **[mechanism]** Pearl's see/do distinction settles the semantics: P(ellipse | guidance=absent) observed is rung-1 association; P(ellipse | do(guidance=absent)) requires that guidance status be set independently of the outcome. InSight's documented record shows the unguided ballistic entry was CHOSEN because the Elysium Planitia site was flat/benign and a large ellipse was acceptable, i.e. treatment (entry type) was assigned as a function of the tolerated outcome (ellipse size). That is selection on the outcome, which is precisely the back-door path that makes the single residual associational rather than interventional. The InSight EDL reconstruction (Karlgaard 2020) and the InSight landing-site characterization/assessment literature (Golombek-family Space Science Reviews 2016; Assessment of InSight Landing Site Predictions 2020) confirm both that InSight flew an unguided Phoenix-heritage entry and that the site was selected for flatness and engineering benignity, not that precision technology was withheld in an otherwise precision-demanding context. Therefore the single point licenses an associational reading only, unless the candidate produces a conditional independence (e.g. ellipse independent of entry-type given site-difficulty and mission-cost class) that holds in the data and removes the selection back-door. Absent that, the linchpin residual is a rung-1 artifact, not the rung-2 do-evidence the design treats it as.
    - Pearl, Causality (2nd ed., 2009); do-operator / ladder-of-causation exposition [pearl dossier, Hall of Shoulders brain] | https://doi.org/10.1017/cbo9780511803161 | grade A
    - Karlgaard et al., Mars InSight Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction (AIAA SciTech 2020) | https://doi.org/10.2514/6.2020-1271 | grade B
    - Golombek-family InSight site characterization, Space Science Reviews (2016); Assessment of InSight Landing Site Predictions, J. Geophys. Res. Planets (2020) | https://doi.org/10.1029/2020je006502 | grade A
- **[identification]** Two perfectly collinear indicator columns (identical 1/0 pattern across all observations) span a one-dimensional subspace, so the design matrix is rank-deficient in those columns and gamma_1, gamma_2 are not separately identified: only their sum (the bundle coefficient) is estimable from the data. No conditional-independence test internal to a dataset in which the two indicators never vary independently can apportion the effect, because apportionment requires at least one observation where one lever fires without the other. Reporting separate signs and a Mokyrian magnitude ordering is therefore importing the split from physical/historical priors, not estimating it. The Pearl-consistent honest representation is to (a) collapse to a single identified bundle coefficient and report the within-bundle split as UNIDENTIFIED, or (b) supply external quasi-experimental leverage, a mission flying one lever without the other (e.g. a guided entry without range-trigger, or a reconstructed counterfactual deploy-dispersion budget that attributes ellipse reduction to range-trigger timing separately from bank-angle guidance). The MSL guided-entry reconstruction is the natural source for such a deploy-dispersion decomposition: MSL's guided entry delivered a 19.1 km x 6.9 km ellipse with the rover ~2 km from target, and the Entry Terminal Point Controller modulates bank angle to control range to deploy, which is the lever the candidate would need to separate from the range-trigger contribution. But that decomposition is external interventional/engineering evidence imported into the model, not an estimate from the nine-point regression.
    - pearl review-lens dossier (falsifiable-questions section) [Hall of Shoulders brain]; Pearl, Causality (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
    - Post-Flight EDL Entry Guidance Performance of the 2011 Mars Science Laboratory Mission (NASA NTRS 20120015817) | https://ntrs.nasa.gov/citations/20120015817 | grade B
- **[identification]** The site-selection s-node sits ON the technology-to-ellipse causal path, so gamma is not transportable by the candidate's own machinery. The robotic census self-selected flat, hazard-sparse ellipses: the published Golombek site-selection record makes geological/hazard factors (centimeter-to-kilometer slopes, rock abundance) an explicit constraint on which sites were even attempted, and the Mars 2020 LVS / Safe-Target-Selection reconstruction shows terrain-relative navigation choosing the divert target as a function of in-frame hazard. Both are selection ON the EDL outcome (achievable touchdown footprint), so the realized ellipse reduction attributed to a guidance lever is co-determined by the hazard field the mission was allowed to fly into. A target that pre-places assets or accepts hazard-bounded sites the robotic curve never sampled changes the value of that s-node, so the technology effect gamma is modified along this path, not merely shifted. Per Pearl's selection-diagram apparatus, an s-node on the causal path between treatment and outcome makes the estimand population-specific.
    - Golombek et al., Broad Perspectives on Mars Landing Site Selection: Geological Factors from Centimeter to Kilometer Scale (NTRS) | https://ntrs.nasa.gov/citations/20010019301 | grade B
    - Mars 2020 Perseverance EDL GNC Safe Target Selection Reconstruction (NTRS) | https://ntrs.nasa.gov/citations/20230005646 | grade B
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
- **[rival]** Verdict: non-transportable as stated; at best transportable-after-reweighting only on the atmosphere-regime s-node, and not on the site-selection s-node. Do-calculus reasoning: transport licenses P*(ellipse | do(guidance)) = sum over the s-node distribution of P_source(ellipse | do(guidance), s) weighted by P_target(s) ONLY when each s-node is s-admissible, i.e. the technology effect is invariant given s (s acts purely as a population shift, not an effect modifier). The atmosphere-density s-node is plausibly a pure shift the source can condition on, because the high-mass entry-guidance literature treats density regime as a parameter the guidance law is designed against rather than an unmodeled selector, so gamma could in principle be re-weighted by a target-measurable density distribution (transportable-after-reweighting on that node). But the site-selection s-node is selection ON the outcome the panel already established for InSight (Elysium Planitia was chosen for being flat/hazard-sparse), and that same mechanism operates across the whole robotic census via the Golombek constraints and the Karlgaard-family reconstructions that report achieved performance only inside self-selected ellipses. When the selector is the outcome, gamma is a population-specific quantity: it does not survive transport because no target-measurable reweighting of P(s) can recover an effect that was estimated only over hazard-sparse draws the target population does not respect. Net verdict therefore collapses to non-transportable.
    - Pearl & Mackenzie, The Book of Why (2018), formalized in the do-calculus transport apparatus (pearl dossier, COLLEGIUM Hall of Shoulders) | https://doi.org/10.1017/cbo9780511803161 | grade A
    - Jiang, Li & Furfaro, Integrated guidance for Mars entry and powered descent using reinforcement learning and pseudospectral method, Acta Astronautica (2018) | https://doi.org/10.1016/j.actaastro.2018.12.033 | grade A
    - Karlgaard et al., Mars Science Laboratory Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction (NTRS) | https://ntrs.nasa.gov/citations/20130010087 | grade B
- **[governance]** Recovery datum: the single target-population observation that would license the atmosphere-regime reweighting is a target P(s) over high-mass entry density/dynamic-pressure regimes, available from a real Human-Mars high-mass EDL architecture study or an MSR propulsive-EDL trade study; the site-selection s-node, by contrast, cannot be reweighted and instead requires a target-measurable conditional ellipse distribution drawn from a pre-placed-asset / hazard-bounded reference architecture (e.g. a Human Mars EDL architecture study or MSR commercial propulsive-EDL study reporting achievable footprints at hazard-bounded sites). Concretely, the NASA Human Mars Entry, Descent and Landing Architecture Study (Rigid Decelerators) and the Mars Sample Return Using Commercial Capabilities propulsive-EDL study are real records that could supply the target footprint distribution at hazard-tolerant sites. Absent that observation, the prospectus must add a scope sentence of the form: 'Gamma is the proportional ellipse reduction estimated over the 1976-2020 US robotic Mars EDL census, conditional on self-selected flat, hazard-sparse landing sites; it is NOT a transportable price for MSR-retrieval or Human-Mars architectures that pre-place assets or accept hazard-bounded sites, and must not be read as a decision-relevant guidance-trade coefficient for those populations until a target-measurable footprint distribution at hazard-tolerant sites is supplied.' This guards the coefficient given its stated 'confidence beyond the estimation population: low.'
    - Human Mars Entry, Descent and Landing Architecture Study: Rigid Decelerators (NTRS) | https://ntrs.nasa.gov/citations/20180006608 | grade B
    - Mars Sample Return Using Commercial Capabilities: Propulsive Entry, Descent, and Landing (NTRS) | https://ntrs.nasa.gov/citations/20140006930 | grade B
    - Pearl, Causality: Models, Reasoning and Inference (2nd ed., 2009) | https://doi.org/10.1017/cbo9780511803161 | grade A
- **[measurement]** PARTIAL ANSWER WITH A CAP. The instrumentation threat is real and named, not a rhetorical worry: the Viking and InSight ellipses are both products of statistical entry/atmosphere reconstruction, and the Viking reconstruction (2nd International Planetary Probe Workshop) and the InSight reconstruction (Karlgaard et al., J. Spacecraft & Rockets 2021, NTRS 2020) are separated by the IMU/FADS instrumentation suite, aerodynamic databases, and atmosphere models available in each era. In Shadish-Cook-Campbell terms this is exactly instrumentation: the measuring instrument changed monotonically alongside the technology to be credited, so an uncontrolled before-after difference confounds the two. The candidate's single technology-off observation cannot, by itself, close this internal-validity rival. CAP: retrieval this turn surfaced the existence and authorship of both reconstruction papers but NOT their internal convention parameters (confidence level, dispersion sourcing, atmosphere model). I therefore cannot document which specific convention shifts are differenceable versus confounded. The candidate must extract those parameters from the NTRS reconstruction papers mission-by-mission before claiming the ellipse difference reflects capability rather than redefinition; absent that, instrumentation remains a live, unrefuted threat that bounds the strongest claim.
    - Trajectory, atmosphere, and wind reconstruction from Viking entry measurements (2nd International Planetary Probe Workshop); Mars InSight Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction (Karlgaard et al., NTRS) | https://ntrs.nasa.gov/api/citations/search?q=Viking%20entry%20atmosphere%20reconstruction | grade B
    - Karlgaard, Korzun, Schoenenberger, Bonfiglio, Kass, 'Mars InSight Entry, Descent, and Landing Trajectory and Atmosphere Reconstruction,' Journal of Spacecraft and Rockets, 2021 | https://doi.org/10.2514/1.a34913 | grade A
    - Shadish, Cook & Campbell review lens (dossier); Bernal, Cummins & Gasparrini, 'Interrupted time series regression for the evaluation of public health interventions,' Int. J. Epidemiology, 2017 | https://doi.org/10.1093/ije/dyw098 | grade A
- **[identification]** PARTIAL ANSWER, RIVAL REMAINS OPEN. The selection threat is correctly identified and the approach-accuracy control does NOT close it. Retrieval confirms the InSight site was assessed and chosen on engineering-safety grounds (regolith/flatness pre-landing assessments in the Golombek-lineage InSight site characterization literature), which means site-selection latitude is correlated with the outcome (ellipse size) independently of guidance technology. In Shadish-Cook-Campbell terms this is selection operating directly on the dependent variable: the approach-accuracy control fixes injection accuracy but leaves site-selection latitude free, so it cannot difference out the 'a flat site afforded a large ellipse' channel. The control therefore does not close the rival; selection-on-the-outcome stays open and bounds the evidentiary weight of the InSight residual. WHAT I CANNOT SUPPLY: retrieval did not return a specific separating observable (e.g., a documented required-ellipse-size or landing-hazard margin that was specified independently of achievable guidance accuracy) for InSight. Without that observable the rival cannot be experimentally settled, only acknowledged. The honest position is that the candidate must produce, from the site-selection records, an observable that fixes site-selection latitude; until then the InSight residual is suggestive, not dispositive.
    - 'A Pre-Landing Assessment of Regolith Properties at the InSight Landing Site' (InSight site-selection / Golombek-lineage characterization) | https://doi.org/10.1007/s11214-018-0537-y | grade A
    - Shadish, Cook & Campbell review lens (dossier) - selection and 'which threat does your design rule out, and which is still alive' falsifiable questions | https://doi.org/10.1207/s15327752jpa4601_16 | grade A
- **[rival]** GROUNDED ANSWER. Tianwen-1 has a reconstructed entry/atmosphere trajectory and a positioned landing site in the published record, so it is a usable second observation, not merely a reference. Wei et al. (2022, Astrodynamics) reconstructed the Tianwen-1 entry-vehicle trajectory and atmosphere from IMU + flush-air-data-system data with a CFD aerodynamic database; Wang et al. (2022, Astrodynamics) reconstructed the descent trajectory and positioned the landing site in Utopia Planitia to 1.56 m. Tianwen-1 flew guided entry in the modern era, i.e., it varies technology while holding the calendar roughly fixed against InSight, making it the mirror-image discordant point. The Shadish-Cook-Campbell prescription for exactly this inference is an interrupted time series with a COMPARISON series, not a single interrupted observation: a two-point design with two technology-discordant cases (InSight technology-off, Tianwen-1 technology-on) survives the loss of either point in a way the single-case drop-InSight refit cannot, because each point independently differences the technology axis against the time axis. Therefore bringing Tianwen-1 in-sample would harden the discrimination, and holding it out leaves the strongest available identification on the table. A defensible reason to hold it out would be a documented non-comparability of the Tianwen-1 reconstruction conventions (different sigma definition, different atmosphere model, different ellipse construction) that would reintroduce the very instrumentation confound of Q1 across programs/agencies; retrieval did not establish such a convention break, so the candidate cannot lean on it as a settled justification and should treat in-sampling as the stronger design absent that evidence.
    - Wei, Rao, Chen, Wang, Zou, 'Tianwen-1 Mars entry vehicle trajectory and atmosphere reconstruction preliminary analysis,' Astrodynamics, 2022 | https://doi.org/10.1007/s42064-021-0116-y | grade A
    - Wang, Wang, Zhang, Hua, Li, 'Landing site positioning and descent trajectory reconstruction of Tianwen-1 on Mars,' Astrodynamics, 2022 | https://doi.org/10.1007/s42064-021-0121-1 | grade A
    - Bernal, Cummins & Gasparrini, 'Interrupted time series regression for the evaluation of public health interventions,' Int. J. Epidemiology, 2017; Shadish, Cook & Campbell review lens (dossier) | https://doi.org/10.1093/ije/dyw098 | grade A
- **[construct]** The construct check as written is unfalsifiable. The prospectus states verbatim that 'agreement between the two series validates the construct, and divergence is itself reported as a finding rather than suppressed' (lines 89, 113), with no pre-stated discrimination rule. By the panelist's own program, a before-after change read off a single self-contained series, with the corroborating series's interpretation defined post hoc so that both agreement and divergence count as success, is a one-group design whose construct interpretation is unprotected: the rubric demands a named rival explanation (here, simulation-convention drift) be ruled out by a stated observable, not absorbed into an outcome-independent narrative. The candidate must therefore set a numeric threshold before fitting (a maximum tolerated drift in the design-ellipse-to-miss-distance ratio across the era) and treat exceedance as falsifying the construct, not as a 'finding.'
    - shadish_cook_campbell dossier (Hall of Shoulders brain) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/shadish_cook_campbell | grade A
    - JPL_AUTONOMY_EDL_06 dissertation.md, lines 89, 113 | D:/Claude_Code/brain/collegium/candidates/dissertations/JPL_AUTONOMY_EDL_06/dissertation.md | grade B
- **[identification]** The panelist is correct that the design lacks the comparison series his tradition prescribes. The interrupted-time-series-with-comparison design exists precisely so a shared secular trend differences out; the canonical methodological statement is that ITS uses a control series to remove shared history and instrument drift, and an uncontrolled single series confounds the intervention with both. The candidate's 'identification' rests on one decoupling point (the InSight counterfactual) plus an in-model approach-accuracy covariate (prospectus lines 77, 109), which is a one-group design with a single off-treatment observation, not a differenced control series. The approach-accuracy covariate the candidate already invokes (delivered entry-state dispersion at the atmospheric interface) is itself the natural candidate comparison quantity, but the candidate uses it as a regressor inside the same census rather than as an external series to difference against, so the shared monotone trend is never actually removed.
    - Bernal, Cummins & Gasparrini, 'Interrupted time series regression for the evaluation of public health interventions,' Int. J. Epidemiology (2017) | https://doi.org/10.1093/ije/dyw098 | grade A
    - shadish_cook_campbell dossier (Hall of Shoulders brain) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/shadish_cook_campbell | grade A
- **[measurement]** The instrumentation threat is real and the candidate's 'moderate' label understates it: the reconstruction methodology demonstrably changed across the missions it grades. The statistical-estimation technique for Mars EDL reconstruction was itself a subject of comparative methodological study (multiple estimation techniques compared for MEDLI-like data, 2012), and the instrument suite producing the late-mission dependent variable (MEDLI2 on Mars 2020) is a different, more capable sensor set than what supported earlier reconstructions, so the measuring instrument co-evolved with the treatment it is meant to evaluate. In the panelist's terms this is the classic instrumentation threat to internal validity: a drifting instrument changes what is measured and confounds any before-after comparison, exactly as the catalogue-detection-floor example in his dossier. Because the instrument improvement and the guided-entry/TRN insertions are coincident in time, the candidate must net the reconstruction-precision change out rather than let it load on the gamma coefficients.
    - Comparison of Statistical Estimation Techniques for Mars EDL Reconstruction (AIAA, 2012) | https://doi.org/10.2514/6.2012-400 | grade B
    - Karlgaard et al., 'Mars Entry, Descent, and Landing Instrumentation 2 Trajectory, Aerodynamics, and Atmosphere Reconstruction,' J. Spacecraft and Rockets (2023) | https://doi.org/10.2514/1.a35440 | grade B
    - shadish_cook_campbell dossier (Hall of Shoulders brain) | D:/Claude_Code/brain/collegium/hall_of_shoulders/brains/shadish_cook_campbell | grade A

## Gaps

- **[mechanism]** No retrieved source supplies the reconstructed stage-by-stage conditional-dispersion / covariance decomposition for MSL that would let anyone state whether the ellipse reduction equals the sum of isolated guided-entry and isolated range-trigger effects or departs from that sum by a sign/ordering-overturning margin. The named reconstruction reports recover trajectory and atmosphere, not (as retrieved) a guided-entry-vs-range-trigger additivity test. The additivity-vs-interaction question is unresolved on the evidence in hand. (raised by ackoff)
- **[identification]** Retrieval does not return the candidate's design-matrix membership, so whether any mission flies guided entry without the range trigger cannot be answered from sources. Nor did retrieval return the per-stage footprint-reduction magnitudes inside Way 2011 / Mendeck that would be needed to defend three separable coefficients over a single coupled-subsystem term. The perfect-co-occurrence identifiability claim cannot be confirmed or refuted on the evidence retrieved. (raised by ackoff)
- **[measurement]** No retrieved source reports a reachable-site / divert-authority metric for Mars 2020 and its relation to ellipse area, so whether the divert-enabled gain is representable as an area reduction (or only as a site-reachability quantity the dependent variable cannot express) is unresolved. The LVS and TRN-accuracy reconstructions confirm flight performance was characterized but do not, as retrieved, settle the area-vs-reachability representability question. (raised by ackoff)
- **[governance]** No retrieved source this turn documents the specific falsification the question demands: a trade-study record that takes a per-lever ellipse-reduction rate read off the candidate's curve and checks it against whether ellipse area is the BINDING constraint on the MSR or human-Mars architecture (versus mass, hazard density, or mobility being binding). The MSR concept and MSPG2 records show joint requirement-setting but do not isolate the marginal value of ellipse reduction against the binding-constraint frontier. Until such a record (TechPort/HEOMR architecture sensitivity study) is produced, the candidate's stand-alone-investment framing of the precision lever cannot be falsified from the retrieved evidence and remains an assumed, not demonstrated, binding constraint. (raised by ackoff)
- **[measurement]** Whether the contraction empirically loads on the binary 'inserted' dummy or on the continuous capability stock cannot be settled from retrieval: it is an output of the candidate's own model (the head-to-head fit of dummy vs. capability proxy), which is not in any queried corpus. The framework defect is grounded; the empirical verdict is asserted-pending and must be produced by re-estimating with the continuous proxy entered alongside the dummies. (raised by christensen_c)
- **[identification]** Whether the technology block remains jointly significant after controlling for site-demanded targeting difficulty cannot be settled from retrieval: the joint-significance test is an output of the candidate's model run with the difficulty covariate added, which is not in any queried corpus. The JTBD framing and the covariate's availability are grounded; the absorb/no-absorb result is asserted-pending. (raised by christensen_c)
- **[rival]** The QUANTITATIVE half of the test cannot be settled from retrieval. No retrieved source supplies (a) a computed proportional response of a fixed-point-rendezvous / hit-the-tube error to the same lever, (b) the candidate's own fitted gamma magnitudes, or (c) any cross-job elasticity comparison; the prospectus itself states no regression has been run on the assembled dataset and all numeric statements are illustrative. The Dutta STS reconstruction reports as-flown STS performance qualitatively but no retrieved series ties a science-ellipse lever to a proportional fixed-point-error reduction. Whether the lever has the SAME proportional effect across the two job classes is therefore unproven on the evidence retrieved this turn. (raised by christensen_c)
- **[mechanism]** The co-movement test itself cannot be performed from retrieval. The TechPort query for program-of-record EDL technology-development records returned zero usable hits this turn, and no retrieved source supplies any quantified institutional-continuity proxy (sustained EDL engineering headcount, mission-cadence series, or calendar-gap-between-insertions series) nor any regression of the per-generation contraction rate against such a proxy. Whether the gamma is invariant to or co-moves with institutional continuity is therefore undetermined on the evidence retrieved this turn; the prospectus flags institutional learning as a confound but does not measure it, so the value-network-artifact-versus-portable-effect question is left open. (raised by christensen_c)
- **[identification]** REFUSED (no-confabulation 3.3): no blind-reconstruction test exists to report. The dissertation is design-stage; Ch6 states 'the regression has not been executed on the assembled dataset' and 'every number is expected or illustrative.' No blind-coding panel exercise was run, so there is NO inter-rater agreement rate and NO documented blind step-ordering/breakpoint result. AMOS and ACTA corpus queries on Mars EDL landing-ellipse precision returned zero hits, so no external source supplies a substitute blind-coding outcome. The objection is valid and unanswered: the candidate should run a label-blinded reconstruction with a reported agreement statistic (e.g. kappa / Kendall's W on recovered ordering) before claiming the step structure is carried by the data rather than fitted backward from the known improvement. Asserting any agreement number now would be fabrication. (raised by kahneman)
- **[empirics]** REFUSED (no-confabulation 3.3): no base-rate residual distribution exists to report. The regression is unexecuted, so InSight has no computed residual; Ch4 acknowledges the per-mission reconstruction uncertainty is heteroskedastic across the sequence but does NOT assemble a tabulated single-mission residual / noise-band distribution, and Ch6 routes InSight's leverage through an as-yet-unexecuted InSight-excluded re-fit. AMOS queries for Mars EDL Monte-Carlo dispersion / reconstruction-uncertainty residuals returned zero hits; no retrieved peer-reviewed source quantifies the empirical scatter of a single Mars EDL ellipse around a fitted trend or the convention-drift noise magnitude. The linchpin's weight therefore cannot be adjudicated against a documented noise band on present evidence. Stating an InSight residual magnitude or a chance-residual band now would be fabrication. (raised by kahneman)
- **[measurement]** REFUSED (no-confabulation 3.3): no specification-grid (multiverse) robustness count exists to report. Ch6 NAMES the defensible alternatives (with/without Viking, program-vs-mission experience unit collapsing the Viking and MER pairs, log vs linear, achieved-miss-distance DV, cumulative-landings axis) and commits to a permutation test, but the table is 'specified-but-unpopulated by design' and every cell is labeled illustrative; the regression has not been run. Therefore the fraction of the defensible specification grid that keeps the technology block jointly significant and correctly signed is unknown and unreported. ACTA queries on small-sample specification robustness / multiverse returned zero hits. The objection is valid and unanswered: a full multiverse count is exactly what a 9-to-11-observation, five-slope design needs to show H1 is not a single favorable path. Reporting any supporting fraction now would be fabrication. (raised by kahneman)
- **[identification]** Q1 contamination test cannot be settled from retrieval. The pre-flight TechPort design-ellipse predictions and as-designed technology manifests dated before each landing are not in any queried corpus (AMOS 0 hits, ACTA 0 hits, Space Economy 0 hits on Mars-EDL technology-coding queries). No source retrieved this turn establishes whether the macro-vs-incremental assignment is or is not stable under blind, pre-outcome coding. Refused rather than confabulated: the candidate must produce the pre-flight document set and the blind re-code; the record needed to adjudicate it is not in hand. (raised by kahneman)
- **[empirics]** Q2 out-of-sample physics check cannot be settled from retrieval. The independent physical error-budget decomposition (per-mission downrange/crossrange, parachute-deploy, and position-knowledge error magnitudes in the Karlgaard-family reconstruction and MEDLI2 documents) is absent from every queried corpus (AMOS 0, ACTA 0 on error-budget / reconstruction / downrange-crossrange queries). Without the physics-derived magnitude ordering, no source retrieved this turn can confirm or refute whether it matches the pre-registered economic ordering. The WYSIATI concern (a coherent story used to choose the labels, undetectable by a label-permutation test) is a valid framing, but the discriminating physics record is not retrievable, so the question is refused, not answered. (raised by kahneman)
- **[identification]** Q3 reversal-probability computation cannot be settled from retrieval. The published reconstruction uncertainty bands on the InSight ellipse and on the modern fitted trend are not present in any queried corpus (AMOS 0, ACTA 0 on InSight ellipse reconstruction uncertainty). Without those bands no residual-sign reversal probability can be computed, and the candidate's decision rule has no stated reversal-probability refusal threshold in any retrieved source. The base-rate-vs-single-salient-case critique is well-founded methodologically (one vivid observation standing in for a base rate is the canonical availability/salience error), but the quantitative robustness check the question demands the record settle has no supporting data in retrieval. Refused. (raised by kahneman)
- **[measurement]** No retrieved source supplies the specific inter-era deflator: nothing characterizes whether a 1976 Viking 'three-sigma' and a 2020 'three-sigma' encode the same tail mass under non-Gaussian dispersions, and no Viking-era reconstruction document stating its dispersion convention was recovered (only USGS topographic mosaics surfaced). The quantitative claim that two analysts would re-derive the Viking ln-area within an order of magnitude is therefore unverified from the corpus; the candidate must produce the Viking primary record and a stated percentile-harmonization rule before the dependent variable can be treated as one commensurable series rather than a stack of incommensurable objects. (raised by kuznets)
- **[measurement]** REFUSED on the numeric commitment. No retrieved source supplies the candidate's specific quantities: no asymptotic three-sigma ellipse-floor area, and no decomposition into the four named physical sub-budgets (entry-interface density/wind residual, IMU/FADS resolution limit, MRO/CTX georeference accuracy, divert-fuel reachability). AMOS, ACTA, and Space-Economy corpora returned zero hits on Mars EDL ellipse error budgets and TRN map-relative georeference accuracy; the kuznets brain supplies only the methodological demand (state your measurement boundary; an aggregate without an explicit coverage/lower-bound rule is undefined), not the candidate's EDL numbers. The grounded half of the answer is the Kuznetsian demand itself: the candidate must publish the floor and its four-way decomposition before any contraction rate is reported, because a contraction series read without its floor draws no boundary. The number itself cannot be asserted without confabulation and is left to the candidate's own MSL/Mars-2020 error-budget source. (raised by kuznets)
- **[empirics]** PARTIAL REFUSAL on the empirics. The method-level diagnostic and inference rule are grounded (c1), but the question also implicitly asks whether the candidate's actual eleven-point design-ellipse series in fact passes or fails this saturation test. No retrieved source contains the candidate's eleven-point series or its log-decrement profile, so the empirical verdict (steep vs flattening) cannot be asserted and is left to the candidate to demonstrate on the series itself. (raised by kuznets)
- **[identification]** REFUSED on the empirical verdict. No retrieved source supplies the PDS achieved-miss-distance series, the NTRS reconstruction provenance bands, or a comparison of realized-capability decrements against design-ellipse decrements at the last two generations. ACTA and AMOS returned zero hits on MSL/Mars-2020 landing-accuracy reconstruction. Therefore whether the realized series in fact agrees with or already flattens relative to the design ellipse at the recent end cannot be asserted and is left to the candidate to demonstrate from the PDS/NTRS record. The licensing test is specified (c2); its pass/fail outcome is unretrieved. (raised by kuznets)
- **[identification]** Theory-maturity dates can order guided entry versus TRN, but no retrieved source supplies an independent propositional-base metric for the range trigger relative to TRN, nor a quantitative scale (theory breadth, citation depth, codification age) that would convert the qualitative macro/micro labels into a graded ranking testable against, rather than read from, the ellipse series. Until the candidate specifies such a metric and dates it independently, the three-way macro/macro/micro ordering remains a partly untested prior. (raised by mokyr)
- **[measurement]** The dissertation's own estimator measures only the per-insertion level shift; it does not, on the retrieved evidence, fit any across-reuse slope of the predict-versus-flight residual, so as specified it cannot detect the extensibility property even though the dated reuse record exists externally. No retrieved source supplies the analogous reuse-improvement series for the range trigger or for guided entry, so the candidate cannot yet show whether the level-shift staircase reflects cumulative deepening for those two levers or disconnected one-time gains. The extensibility indicator is therefore available in principle but unincorporated into the design. (raised by mokyr)
- **[rival]** No retrieved source supplies the codification dates of the specific reconstruction-and-modeling methods named (Karlgaard statistical reconstruction, the Lander Vision System map-building pipeline) as a series independent of the guidance-lever flight dates. Without that dated method-codification series, the candidate cannot empirically separate the technology-insertion effect from the knowledge-access-cost effect, and the InSight counterfactual, reconstructed by the same apparatus, leaves the rival cause unbroken. The disentangling evidence is specifiable in principle but is absent from the retrieved record, so this remains an open confound on the strongest causal claim. (raised by mokyr)
- **[rival]** The dissertation cannot answer which way Tianwen-1 falls against its fitted technology coefficients, because by explicit pre-registered design choice it HOLDS TIANWEN-1 OUT of the estimation sample and uses it 'only as an out-of-sample, non-United-States REFERENCE POINT FOR DISCUSSION' (Sec 5.7.2), and the lunar crater-navigation/TRN citations are likewise admitted 'as a reference point and not as in-sample evidence'. No residual of the Tianwen-1 ellipse against the fitted curve is computed, and no permutation interval is constructed for an out-of-lineage point; the design's permutation machinery (Sec 5.9) is defined only over relabelings of the technology indicators WITHIN the 9-11 US Mars missions, and the candidate states it 'is not warranted to extrapolate to... other programs without the supporting knowledge chain in place', with external-validity confidence rated LOW. The corpus DOES contain the materials the test would need, Tianwen-1 flew an independent guided/lifting entry program (Yu 2021 GNC; Huang 2022 powered-descent GNC flight results) and the lunar crater-navigation source establishes TRN as a general planetary capability, so the test is constructible in principle. But the candidate never specifies the prediction, never extracts a comparable three-sigma ellipse for Tianwen-1 under its own measurement boundary, and never states whether the point falls inside or outside the US-Mars permutation interval. The extensibility/self-correcting claim that underwrites the Mokyrian framing is therefore left as an in-sample assertion (the relative gamma magnitudes and the InSight residual) rather than the out-of-lineage falsification test Mokyr's criterion demands; on the corpus alone the question is unanswerable without performing the analysis the candidate declined to run, so it is recorded as an open gap rather than answered. (raised by mokyr)
- **[identification]** No retrieved source supplies the candidate's nine-to-eleven-point dataset or the per-mission ApproachAccuracy-vs-guidance-generation values, so the empirical sub-question (is there a mission where guidance generation and approach accuracy move in OPPOSITE directions, or did they improve in lockstep?) cannot be settled from retrieval this turn. The published EDL-reconstruction literature exists (MSL: 10.2514/1.a32708; InSight: 10.2514/6.2020-1271) but a within-sample collinearity verdict on the candidate's reconstructed approach-state dispersions was not retrievable. This must be answered from the candidate's own assembled data, not asserted. (raised by pearl)
- **[measurement]** Retrieval did not return the specific reconstruction convention parameters (confidence level, dispersion sourcing, atmosphere model) for Viking versus InSight, so the question's core demand, to document which convention shifts can be differenced out and which remain confounded with the technology indicators, cannot be settled from retrieved sources this turn. The convention-by-convention differencing is asserted nowhere I retrieved and must not be fabricated; the candidate owes a mission-by-mission extraction from the NTRS reconstruction papers. (raised by shadish_cook_campbell)
- **[identification]** No retrieved source supplied an InSight-specific separating observable (a required ellipse size or hazard margin specified independently of achievable guidance accuracy) that would discriminate 'ellipse large because guidance off' from 'ellipse large because a flat site permitted it.' The discriminating observable the question demands is absent from retrieval and is not invented here; the candidate owes it from the site-selection census. (raised by shadish_cook_campbell)
- **[construct]** Whether the design-ellipse-to-PDS-miss-distance ratio in fact drifts monotonically across Viking->Mars 2020 in the same direction as the technology insertions (the empirical correlated-instrument test) cannot be answered from retrieved sources. No retrieved source assembles both a per-mission design-ellipse value and a per-mission PDS-localized miss distance into a paired ratio series, so the direction and monotonicity of any shared-instrument drift is unestablished. The methodological defect (no pre-stated threshold) is grounded; the empirical drift direction is not. Refused rather than fabricated. (raised by shadish_cook_campbell)
- **[identification]** Whether a usable time series of delivered entry-flight-path-angle dispersion or orbit-determination position uncertainty at atmospheric interface actually exists across all nine to eleven missions, and whether overlaying it leaves discrete ellipse steps at MSL and Mars 2020 rather than a smooth shared trend, cannot be answered from retrieved sources. NTRS and OpenAlex queries for a Mars approach-navigation EFPA-dispersion delivery series returned no assembled cross-mission quantity. The requirement for the comparison series is grounded; the existence and shape of the overlay is not. Refused rather than fabricated. (raised by shadish_cook_campbell)
- **[measurement]** Whether a specific reconstruction-precision metric (a stated reconstruction uncertainty, covariance floor, or confidence-interval width) exists per mission AND improves in lockstep with the guided-entry and TRN dummies cannot be answered from retrieved sources. No retrieved source tabulates a per-mission reconstruction-uncertainty value across the census that could be regressed against the technology indicators to net the instrument drift out. The existence of the instrumentation threat and the instrument's co-evolution are grounded; the lockstep correlation between a reconstruction-precision series and the dummies is not. Refused rather than fabricated. (raised by shadish_cook_campbell)