# FRISA Broad Research Round Report ## Synthesis of Three Exploratory Cycles on the Part 121 1,500-Hour First Officer Experience Floor --- ## 0. Charter restatement and headline lock The locked net-ledger headline, restated verbatim and unaltered: > **"sign indeterminate, magnitude small, modest adverse lean."** Everything in this report is **EXPLORATORY** and changes nothing about that result. This round operates under CHARTER LOCK (Amendment 2, hash 15d5109c). Nothing below re-estimates, re-runs, re-weights, or re-labels the locked H_A/H_B net-lives result or any of its parameters (lives_saved_per_decade, the four cost dials, rule_attributable_share, rule_binding_fraction, the displaced-air comparator, the road-substitution chain, rho_AB), nor H_C/H_D. No band-swap, no Colgan-as-causal, no benefit-at-hard-zero, no GA-rate substitution. Every finding here is held to a single identifiability bar applied symmetrically to favorable and unfavorable signals, and every "not estimable" disposition is reported as a data-power or identifiability failure — never as a measured null, and never as affirmative evidence of equivalence. The three cycles below are reported with equal prominence; no cycle is elevated over another, and unflattering, null, and "not estimable" findings are given the same weight as suggestive ones. The red-team flagged real asymmetries in each cycle's drafting memos. Those verdicts are reported honestly in-line and reconciled here so the synthesis does not inherit the leakage. --- ## 1. Cycle C1 — Alternative policy designs and reform-package attribution **What was asked.** Two arms. The design arm (A1–A3) asked whether any credentialed alternative experience pathway — R-ATP reductions under 14 CFR 61.160 (military 750 hr, four-year aviation degree 1,000 hr, two-year degree 1,250 hr), the 2024 FAA Reauthorization Enhanced Qualification Program (EQP), or structured-academy / AQP / approved simulator-credit hours — plausibly dominates the raw 1,500-total-hour instrument on a competence-per-hour or safety-relevant-experience basis, and whether composition of experience can be separated from quantity at all. The attribution arm (B1–B3) asked whether the separately-timed post-Colgan reforms can have their safety credit partitioned across the package. **Design-arm findings.** The rule already embeds structured-pathway hour credit: 61.160 is law, with credit slopes of 750/1,000/1,250 hours tied to military service or accredited degree curricula and a 200-hour cross-country carve-out. The live question is therefore the *slope* of that credit schedule, not whether structured credit exists. The 2024 Reauthorization Act (Pub. L. 118-63) statutorily created EQP — a standardized airline/training-center R-ATP curriculum — but per GAO-26-107856 (Feb 2026) the FAA had **not** publicly issued EQP requirements as of February 2026, missing the November 2024 statutory deadline. The simulator-credit provision (up to 150 hours) that passed the House in H.R. 3935 was **struck on the floor and is not law**, notwithstanding endorsement by eight former FAA administrators and two former ALPA presidents. The Air Carrier Training ARC Recommendation 24-1 (Aug 2024) recommended additional structured-training R-ATP pathways and a PFTO accreditation tier. The closest empirical series — the Pilot Source Study (UND/Embry-Riddle; ~6,734 post-law regional pilots, 19 carriers) — found that pre-law, fewer total hours, recent collegiate graduation, and accredited programs were associated with **better training outcomes**, with those background effects weakening post-law. Crucially, these are **training-completion proxies** (non-completions, extra training, IOE), **not operational safety outcomes**. Per the red-team's own caution, even this training-proxy evidence is associational and observational and must not be characterized as the alternative designs "performing acceptably" in any safety sense — only that the data exist and are training-readiness associational. The ICAO MPL pathway (competency-based, no raw-hour floor; ~4,018 graduates worldwide by 2021) has **no located comparative MPL-vs-traditional safety or accident study**; as of 2013 no MPL graduate had even upgraded to captain, and captain-survey acceptance is perception, not outcome data. **Design-arm disposition (A2).** Whether the *composition* of experience (structured/scenario/simulator/AQP hours) can be distinguished from raw quantity as a safety mechanism is a **data-power failure**: there is no design-level outcome data linking pathway type to Part 121 safety outcomes. **Not estimable.** On the registry side (A3), pathway *type* is in-principle a queryable exposure field (PRD phased 2021+, ATP-CTP completion, R-ATP-by-institution), but the **outcome side is not densely linkable**, so attribution is identifiability-limited, not effect-size-estimable. The disposition is about linkage existence, not effect magnitude. **Attribution-arm findings (B1–B3).** The three core reforms cluster within roughly 18 months on the same small fleet — the 1,500-hour/R-ATP floor (Aug 1, 2013), Part 117 fatigue/duty rules (Jan 4, 2014, which **exclude** Part 135/cargo — a real differential-exposure seam), and enhanced Part 121 training — with SMS (final 2015, phased to ~2018) and PRD (phased 2021+) more separated in time. The denominator problem is decisive: there is essentially **one pilot-causal-informative Part 121 mainline fatal event since 2010** (industry framing: ~99.8% fatality reduction). One event cannot identify a five-regressor staggered partition. **Attribution-arm disposition (B2).** The experience floor's specific safety share is **not estimable — a data-power failure**, reported as such and not as a measured null in either direction. Non-identifiability cuts both ways: no pro-floor and no anti-floor causal share is recoverable from this near-empty numerator. **Red-team verdict on C1: "needs caveats."** The panel raised one major flag the synthesis must carry forward. The FAA-2013 regulatory-evaluation statement ("unable to find a quantifiable relationship between the 1,500-hour requirement and airplane accidents and hence no benefit") was treated in a drafting memo as a load-bearing, near-confirmatory adverse-to-floor anchor. That is an **asymmetric-bar violation**: an "inability to find a quantifiable relationship" is itself a **data-power / identifiability statement about a 2013 ex ante evaluation**, not a measured null of zero benefit, and it must be framed strictly as such. It does not bear on the locked benefit parameters and is not imported here as evidence the floor lacked benefit. A second flag (minor): the mechanistic mapping that both Colgan pilots already exceeded 1,500 hours, used to argue the floor's mechanism is the "weakest fit," is the **inverse of the Colgan-as-causal prohibition** — using the single index accident as a causal lever in either direction is barred, and the observation is quarantined behind an explicit symmetric non-inference caveat. A timing-framing inconsistency (one memo calling the cluster "severe co-timing confounding," another emphasizing "some timing variation") is reconciled here to the conservative reading: the 2013–2014 core is severely co-timed; only UPRT (2019) and PRD (2021) carry enough isolation to keep any future design alive. --- ## 2. Cycle C2 — International comparison beyond Europe and candidate mechanisms **What was asked.** A rate-denominated international natural-experiment arm (A1–A4) across ICAO MPL adopters, Australia, and Canada, plus a mechanism arm (B1–B4) probing whether fatigue, manual-flying skill decay / automation dependency, or CRM / multi-crew exposure carries signal that raw total hours does not proxy — each gated to "not estimable" absent an at-risk-base denominator. **International-arm findings.** The rest of the world's airline-entry floor is roughly 10× lower than the US instrument: EASA/ICAO frozen-ATPL seats a first officer near 195–250 hours; ICAO MPL is competency- and simulator-based with very limited real-aircraft time. But **no regulator or industry body (EASA, ICAO, IATA) publishes an MPL-vs-traditional accident or serious-incident rate with an exposure denominator.** EASA's own July 2020 evaluation reports MPL only as career-path share (~2%), licence-count growth, and training success (~95%) — **no accident/incident rate stratified by pathway.** The only located academic MPL evaluation (Wickander & Dahlstrom, Lund, 2014) is a qualitative host-airline feedback review, explicitly not rate-denominated and underpowered for accidents. Accidents sometimes invoked against low-experience crews (Asiana 214, Atlas 3591) involved **hours-based crews, not low-hour MPL cadets**, and must not be coded MPL-causal. **International-arm disposition (A1–A4).** Australia (ATSB/BITRE) publishes rates by **operation type, not crew pathway**, and reports no scheduled-commercial fatalities since 2005 (near-empty numerator). Canada (TSB) has occurrence counts but no pathway-stratified commercial rate. The A4 pre-data audit verdict is that in **every jurisdiction examined**, an experience-/pathway-stratified accident exposure denominator is **not published at estimand granularity**. This routes A1–A3 to **not estimable / data-power failure** before any rate is computed — explicitly **not** evidence of pathway equivalence, and with the **GA-rate substitution prohibited** (no substituting a general-aviation accident rate for the air-transport estimand). **Mechanism-arm findings.** Within-study, hours-separable performance contrasts do exist in the simulator/handling literature: Casner et al. (2014, n=16, mean ~17,844 career hours) found automation-supplanted cognitive skills decayed with **recency, not career-hour totals** (with FMS/map off, 44% missed the missed-approach point); Ebbatson et al. (2010) likewise indexed handling skill to very recent experience, not accumulated hours. LOC-I is the largest fatal accident class (~one-quarter of scheduled commercial fatalities, 2009–2018). Regulators answered the mechanism with **exposure/training requirements** (EASA Advanced UPRT mandatory pre-type-rating, Dec 20, 2019; Part 117 fatigue, Jan 4, 2014), not more hours. Dawson & Reid (1997) establish fatigue as a graded, experience-independent impairor (~17 h wakefulness ≈ 0.05% BAC). CRM meta-analyses show large effects on proximal (attitudes/behaviors) outcomes but "insufficient evidence" on distal accident outcomes — a surrogate-endpoint gap. **Mechanism-arm disposition (B1–B4).** Critically: the simulator/handling studies have **no accident-rate or at-risk-base denominator**. The pre-registered B2/B4 gate requires the candidate ingredient to be denominated against an at-risk base and to separate from hours under a contrast **before** it may be called a credible active ingredient. That denominator does not exist. Therefore B1 (fatigue), B2 (manual-skill decay), and B3 (CRM/multi-crew) are each **not estimable** as accident-rate associations — and per the gate, this is **not** read as evidence that hours is the right proxy. **Red-team verdict on C2: "needs caveats" — three major flags, all carried forward.** First, an **asymmetric-bar violation**: a drafting memo graded manual-skill-decay as "the best-supported mechanism" / "an active ingredient that total hours do not proxy" on Casner/Ebbatson, which are simulator-performance studies with no at-risk denominator — the exact deficit that routed the A-track to "not estimable." Same evidentiary deficit cannot earn "supports" in one direction and "not estimable" in the other. This synthesis therefore reports the mechanism arm as **not estimable at the accident-rate estimand**, with the simulator findings logged only as hypothesis-generating performance contrasts. Second, **Colgan-as-causal**: a memo deployed Colgan's NTSB probable cause ("competency and state failures, not an hours deficit") as a directional pillar for competency-over-hours. That is the banned move; Colgan is here an index case offered only as analogy, never as a causal pillar, and is **not** counted among any "converging" strands. Third, **exploratory-dressed-as-confirmatory / manufactured convergence**: a memo's bottom line ("most plemently identifies the active ingredient as recency + structured multi-crew/CRM exposure rather than total hours") and its "CONVERGE" framing across three independently confounded strands (FAA RIA, Colgan, Pilot Source Study) generate confirmatory-feeling weight the underlying confounded, analogy-grade evidence cannot support. The synthesis explicitly **does not** adopt any such mechanism-direction conclusion. The net-ledger arithmetic itself was untouched; the leakage was on the mechanism-direction side, and it is quarantined here. --- ## 3. Cycle C3 — Labor-market / equity-access and cost-effectiveness **What was asked.** An equity-incidence arm (A1–A4): the dollar and forgone-income magnitude of building from ~250 hr to 1,500 hr, and on whom the cost falls by socioeconomic and demographic strata. And a cost-effectiveness arm (B1–B4): the rule's implied cost per statistical life **conditioning on the locked net-lives result as a fixed, never-recomputed input**, a bounded societal-cost numerator, a comparator menu of alternative safety investments, and a QALY/discount sensitivity sketch. **Equity-arm findings.** Total cost to airline qualification is commonly cited at $80,000–$100,000+, with a pure-rental time-building upper tail of $150k–$300k. But the **modal path is the CFI route**, where the time-building increment is near-zero out-of-pocket because instructors are paid while building — converting the barrier to roughly two years of **forgone/low income** rather than tuition. The R-ATP carve-outs (1,000/1,250/750 hr) route partial mitigation through degree and Title-IV pathways. The financing architecture is **plausibly regressive as a documented mechanism**: standalone flight schools are generally not Title-IV/Pell eligible unless degree-embedded, and private flight-training loans cap low (~$30k–$40k) against an $80k–$100k+ cost. GAO-14-232 reported schools citing high costs and low entry pay as deterrents but found **"mixed evidence" of a shortage** and noted employment and earnings had decreased since 2000 — the rule effect is **not cleanly isolated**. Occupation demographics (~6.3–6.5% women, ~84% White, ~2.76% Black; women of color ~1%) are **stock, not flow**, and do not attribute composition to the rule. United Aviate Academy's 2021 diversity-and-cost initiative is industry advocacy framing, **not** a causal estimate of the 1,500-hr rule. A documented two-sided tension: post-rule scarcity raised regional first-year FO pay (~$35/hr in 2017 to ~$45/hr in 2021, and onward to ~$93/hr by 2024), a post-entry wage gain partly offsetting the up-front regressive barrier — presented at the **same identifiability bar** as the barrier claim. **Equity-arm disposition (A2/A3).** The rule's marginal SES/demographic screen-out is **not estimable / confounded** — entangled with tuition inflation, the 2013–19 hiring cycle, and COVID. Reported as such per pre-spec: not a measured effect and not a measured null. No peer-reviewed study isolating the rule's marginal compositional effect was located. **Cost-effectiveness-arm findings.** The operative DOT/FAA VSL for 2024 analyses is $13.2M (2023$), range $7.9M–$18.5M. (The Feb 2026 VSL-guidance withdrawal was **CPSC, not DOT**; any claim DOT's VSL changed in 2026 is unverified.) The FAA's 2013 Final Rule (78 FR 42324) scored the statutory 1,500-hour requirement at ~$6.4B nominal / ~$2.2B present value and assigned it verbatim "No Identifiable Accident Benefits"; the only cost-beneficial pieces were the discretionary add-ons (ATP-CTP + type rating), **not** the hour threshold. Comparator RIAs exist: the 2012 fatigue rule (~break-even, with all-cargo carved out because cost exceeded benefits) and TAWS/EGPWS (~95% CFIT prevention; CFIT airline fatality rate down ~89%, exact $/life unverified from primary sources this round). **Cost-effectiveness-arm disposition (B1).** This is the decisive finding and the one most exposed to leakage. **Conditioning on the locked denominator — sign-indeterminate and small — the implied cost per statistical life is not a well-posed point estimate.** It is undefined or sign-indeterminate (division by a near-zero or wrong-signed denominator) and **must be reported as "not a well-posed point estimate," never as a favorable or unfavorable cost-per-life number.** **Red-team verdict on C3: "needs caveats" — four major flags, and the synthesis explicitly declines the leaked conclusions.** The panel found that a drafting memo converted the charter-mandated "not well-posed" ratio into a one-directional unfavorable conclusion ("~150x–440x VSL," "ranks at or near the bottom"), building those defined ratios **only** from the positive-lives / FAA-retrospective side (~1.1 lives/decade from a ~55% retrospective reduction, Colgan ~35%) while parking the favorable-to-rule tail as "undefined." That is four distinct violations at once: (1) an **asymmetric ratio** that suppresses the favorable tail; (2) an **alternative-denominator re-litigation** — constructing ~1.1 lives/decade from attributable-share-style figures is a de facto re-derivation of the locked denominator and Colgan-as-causal-adjacent; (3) a **benefit-at-hard-zero lean**, anchoring a directional verdict on the "No Identifiable Accident Benefits" scoring; and (4) an **implied policy recommendation** ("would flag it," "would have excluded it"), which scope-out bars. **This synthesis adopts none of those conclusions.** Per charter, B1 stands at: **the cost-effectiveness ratio is not a well-posed point estimate while the locked denominator's sign is indeterminate.** No "150x VSL," no "bottom-ranked," no defined ratio is carried — favorable or unfavorable. The illustrative bounding arithmetic is set aside as a charter violation, not a finding. The B2 societal-cost numerator and B3 comparator menu are retained only as itemized, externally-sourced **benchmark context**, with the rule positioned against them qualitatively given B1's ill-posedness and with **no back-solving** of the locked lives number. B4 confirms a QALY/discount sketch **cannot** convert an indeterminate-sign denominator into a defined ratio. The minor flags (statute-vs-rule numerator forking path; multiplicity discount not demonstrated before nominating candidates) are noted and addressed in §5. Memo 1 (equity arm) was found largely compliant — Colgan fenced as context, A2/A3 marked not estimable, advocacy framing kept non-causal, the wage-gain defense held at the same bar. --- ## 4. Candidate hypotheses for future confirmatory pre-registration The following are sharpened candidates each cycle produced. **None is confirmatory now.** Each is stated as a hypothesis-and-test to be pre-registered in a fresh future round, not as a directional result; where a drafting memo pre-stated a direction (e.g., a non-inferiority *conclusion*), it is reworded here to a neutral *test of whether* the relation holds, per the red-team. **From C1 (designs & attribution):** - **Precursor non-inferiority (reworded to neutral):** *Test whether* R-ATP-degree-pathway first officers show equivalent / non-inferior non-fatal precursor rates (FOQA/ASAP exceedances, unstable-approach/go-around, LOC-I-relevant events) versus raw-1,500-hour FOs at matched carriers/fleets/IOE-period, via a pre-registered equivalence test on a precursor with an a-priori-established base rate. Requires PRD-linked or carrier-supplied data. - **Part 117 cargo-exclusion difference-in-differences:** the cleanest natural-experiment seam — fatigue-attributable precursor incidents between Part 135/cargo (excluded) and Part 121 (included) across the Jan 2014 boundary. Dead on fatal counts; estimable only on precursors, contingent on comparator-validity. - **Staggered SMS/AQP cross-operator event-study** to isolate SMS/AQP's marginal contribution from the co-timed 2013–14 cluster; conditional on operator-level adoption dates and FOQA access. - **UPRT (2019) / PRD (2021) interrupted-time-series on higher-frequency surrogates** — the only package members temporally separated enough to support any ITS; pre-register the surrogate ladder and the explicit no-fatal-outcome power limit. - **Surrogate-endpoint power study as a pre-registration prerequisite** (not itself confirmatory of the floor): compute minimum detectable effect per reform before any attribution, converting today's "not estimable" into a documented power-failure boundary. - **PRD-linked registry-attribution feasibility** (addresses linkage, not effect size): whether PRD + ATP-CTP + R-ATP codes can tag right-seat pilots by pathway in principle. - **Competence-per-hour readiness equivalence:** *test whether* structured EQP/MPL-style curriculum hours are non-inferior to an equal-readiness quantum of raw GA hours on a pre-specified competency battery at a common checkpoint — explicitly bracketed as a **readiness** outcome, not a Part 121 safety claim. Feasibility contingent on FAA issuing EQP requirements and PRD maturing (both currently unmet per GAO-26-107856). **From C2 (international & mechanisms):** - **H-MECH1 (recency over hours, hours-fixed):** *test whether* manual-handling/recovery performance or LOC-I-precursor exposure is predicted by recency holding total hours fixed — credible only if recency separates from hours in an hours-matched design denominated against an at-risk base (scaling Casner/Ebbatson). - **H-MECH2 (multi-crew/CRM dosage vs hours):** requires an LOSA-coded dosage denominator linkable to event rates; absent it, not estimable. - **H-MECH3 (fatigue carries signal hours does not):** fatigue proxy associated with event rate after conditioning on hours, on an at-risk base. - **H-MECH4 (pathway RATE contrast):** *if and only if* an A4-gate pathway-stratified rate denominator is ever published, test whether structured-multi-crew pathways differ in rate from hours-based pathways. - **H-MECH5 (UPRT/Part 117 as quasi-experiment):** ITS/DiD against jurisdictions without the change; pre-register pre-trend tests and the multiplicity correction. The MPL-vs-frozen-ATPL serious-incident-rate design within a dual-pathway carrier group, and the LOE/line-check first-attempt fail-rate surrogate design, sit here too — both contingent on a provider supplying pathway-tagged exposure denominators, with "not estimable" pre-committed if they cannot. **From C3 (equity & cost-effectiveness):** - **H-A2*/H-A3*:** DiD/ITS and triple-difference on the student→commercial→CFI→R-ATP funnel around the 2013 date, testing whether conditional retention or composition shifted by SES/demographic strata net of tuition-inflation and hiring-cycle controls, with a pre-committed "not estimable / confounded" stopping rule. - **H-A1*/H-A4*:** whether the rule-attributable entry-cost increment is dominated by forgone income (CFI effective wage × months-to-threshold), and whether reliance on family support / concurrent employment falls disproportionately on under-represented strata and predicts slower accrual — symmetric identifiability bar. - **H-FIN*:** whether non-Title-IV-eligibility of standalone training shifts low-wealth entrants into degree pathways or out of the pipeline. - **Cost-ledger re-derivation candidates:** independent re-derivation of the FAA-2013 RIA cost tables against the docket, and a benchmark-ranking against a fixed comparator menu — **with the locked net-lives verdict held fixed and every ill-posed (undefined-denominator) case pre-declared "not estimable" rather than scored**, and with no implied policy recommendation attached. --- ## 5. Cross-cycle multiplicity and limitations This round opened a wide exploratory surface: C1 contributed roughly four design/attribution candidate families, C2 roughly four-to-five mechanism/international candidate families, and C3's eight pre-commit questions yielded a further cluster of equity and cost candidates — on the order of sixteen-plus independent looks across the three cycles. **None of this surface is multiplicity-corrected confirmatory evidence.** No look-penalty or family-wise correction has been applied or demonstrated; the red-team correctly flagged (C3, minor) that candidates were nominated before any pre-specified multiplicity discount across cycles was shown. That discount is a prerequisite for any future confirmatory round and is **not** satisfied here. Three structural limitations recur across all three cycles and dominate the round: 1. **The empty numerator.** Roughly one pilot-causal-informative Part 121 fatal event since 2010 makes both reform-attribution (C1-B) and pathway-safety comparison (C1-A, C2-A) data-power failures on fatal/hull-loss outcomes. Every promising design migrates to non-fatal surrogates (FOQA/ASAP exceedances, unstable approaches, LOE fail rates), which are themselves contingent on data-access that does not currently exist in collectible, linkable form. 2. **The missing stratified denominator.** No jurisdiction — US, EASA, Australia, Canada — publishes accident exposure cross-tabulated by crew pathway or experience at estimand granularity. This routes the entire international arm and the mechanism accident-rate estimands to "not estimable," and it is the single most durable identifiability ceiling in the round. It is reported as a ceiling, not as equivalence. 3. **The asymmetric-bar temptation.** Every cycle's drafting memos drifted, at least once, toward grading a favorable-to-prior (or, in C3, unfavorable-to-rule) finding above the identifiability bar that the same memo applied to the opposite direction — manual-skill-decay graded "supports" on sim-only evidence (C2); the FAA "no benefit" statement imported as adverse anchor rather than as a data-power statement (C1); a "150x VSL / bottom-ranked" ratio carried off a suppressed favorable tail (C3). The synthesis has, in each case, pulled the finding back to the symmetric bar: simulator and training-proxy evidence is hypothesis-generating only; "no quantifiable relationship" is an identifiability statement, not a measured null; and the cost-effectiveness ratio is not well-posed in either direction. The banned moves — Colgan-as-causal (and its inverse), benefit-at-hard-zero, alternative-denominator re-derivation, GA-rate substitution, and implied policy recommendation — were each flagged where they surfaced and are **not** adopted as findings. A final non-repackaging guard: cumulative exploratory findings across C1, C2, and C3 **may not** be recombined into a confirmatory claim. Only a fresh future pre-registration — with its own multiplicity correction, its own a-priori power floor, and its own symmetric identifiability bar — can promote any candidate above exploratory status. --- ## 6. Closing — relation to the three published papers The three published FRISA papers stand on the locked net-ledger and on the H_C/H_D results. This broad research round re-opens none of them. The locked headline remains, verbatim: > **"sign indeterminate, magnitude small, modest adverse lean."** What this round adds is **scaffolding, not score**. It maps where the question *could* be sharpened — the composition-versus-quantity mechanism, the Part 117 cargo-exclusion seam, the UPRT and PRD timing isolation, the pathway-stratified denominator that no authority yet publishes, the regressive financing architecture, and the ill-posedness of any cost-per-life ratio while the locked sign is indeterminate — and it documents, with equal prominence, where the question is simply **not estimable today**: the near-empty Part 121 fatal numerator, the absent cross-jurisdiction stratified denominators, and the confounded post-2013 entry pipeline. None of the locked parameters (lives_saved_per_decade, the four cost dials, rule_attributable_share, rule_binding_fraction, the displaced-air comparator, the road-substitution chain, rho_AB) has been touched, re-weighted, or re-labeled, and neither has H_C/H_D. Every finding above is EXPLORATORY. Every "not estimable" is a data-power or identifiability failure honestly reported as such, never as a measured null and never as evidence of equivalence. The favorable and the unfavorable have been held to the same bar. The candidate hypotheses are exactly that — candidates for a future pre-registration that does not yet exist — and nothing here changes the published result.