{
  "hypotheses": [
    {
      "id": "H_A",
      "name": "Monotonic experience-floor (pro-rule)",
      "statement": "Across the 250->1,500 TFH band the rule actually moves a co-pilot through, the correctly-denominated Part 121 first-officer accident rate -- and specifically its catastrophic left tail (loss-of-control, icing, startle, unstabilized approach) -- carries a genuine, materially-negative marginal slope that is a within-pilot skill-dose effect surviving exposure correction and survivor/healthy-worker adjustment. The rule therefore buys a positive expected number of Part 121 lives per decade. Distinguishing falsifiable prediction: the benefit term has a median materially above zero (rule prevents on the order of 1+ relevant events per decade). Forbidden moves under H_A: importing the out-of-band Li/Baker 5,000-9,999 hr RR=0.43 gradient into 250-1,500 (cardinal band-swap); reading Colgan as hours-causal (both pilots exceeded 1,500 hr).",
      "benefit_param_override": "lives_saved_per_decade ~ lognormal with median ~2 lives/decade (low/P5 ~0.5, mode ~2, high/P95 ~8), negligible left tail, anchored to the observed base rate of ~2 Part 121 passenger fatalities in 15 yr post-2009 so it cannot exceed physical plausibility. Generated by a monotone-decreasing in-band slope (risk proportional to hours^-b, b>0). All cost-side parameters identical to H_B."
    },
    {
      "id": "H_B",
      "name": "Structure-and-selection-dominated, non-monotonic shoulder (anti-rule-as-instrument)",
      "statement": "The 250->1,500 band sits on the slow-declining shoulder of a curve whose hump lies below ~350 hr (Knecht/Craig) and whose only large documented protective gradient (Li/Baker RR=0.43) lies out of band at 5,000-9,999 hr. At the policy margin, structured-pathway membership and recency screen off raw TFH (Pilot Source Studies: lower-hour structured >= higher-hour time-builders on TRAINING outcomes), and high-hour risk reduction is substantially survivor/healthy-worker enrichment, not skill-dose. The rule's marginal Part 121 benefit is therefore small and the rule is an inefficient instrument. Distinguishing falsifiable prediction: the benefit term is one-sided non-negative with median near zero (~0.1-0.5 lives/decade) and P5=0, with only a thin precautionary right tail. Forbidden move under H_B: scoring the benefit at a hard zero -- the FAA 'no quantifiable relationship' and the absent MPL-vs-time-builder accident comparison are power/data failures, not measured nulls; the precautionary left-tail floor is underpowered but non-negative, so E[saved]>0 by construction.",
      "benefit_param_override": "lives_saved_per_decade ~ half-normal / lognormal one-sided non-negative with P5=0, median ~0.3 lives/decade, mode near 0.1, thin right tail to ~3-5 (high/P95 ~3), negligible left tail. Generated by a flat-in-hours in-band slope shifted by a structure/pathway dummy and a wide near-null pathway x TFH interaction. All cost-side parameters identical to H_A."
    }
  ],
  "parameters": [
    {
      "name": "lives_saved_per_decade",
      "meaning": "Rule-attributable Part 121 lives saved per decade (the benefit term, lives_saved_Part121). THE ONLY hypothesis-dependent parameter.",
      "shared_or_hypothesis_dependent": "hypothesis-dependent",
      "distribution": "lognormal (H_A) / one-sided half-normal-lognormal with P5=0 (H_B)",
      "low": 0,
      "mode": 0.3,
      "high": 8,
      "source": "FAA 'no quantifiable relationship' (TheAirCurrent); base rate ~2 Part 121 pax fatalities in 15 yr (NTSB/BTS). H_A: median ~2, P5~0.5, P95~8. H_B: median ~0.3, P5=0, P95~3. Both negligible left tail; H_B never at hard zero."
    },
    {
      "name": "communities_lost_all_service_2013_2026",
      "meaning": "Small US communities that lost ALL scheduled passenger air service over the rule window.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "triangular",
      "low": 14,
      "mode": 35,
      "high": 50,
      "source": "GAO-24-106681 (=1 over 2018-2023 lost ALL service); RAA=14 by Nov 2022; ~30+7 cuts 2013-2016 (Diio Mi/Travelweekly). Low end anchored to reproducible all-service-lost count."
    },
    {
      "name": "rule_attributable_share",
      "meaning": "Fraction of observed service loss causally attributable to the 1,500-hr rule net of fuel, consolidation, <=50-seat RJ retirement, demographic decline, COVID.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "Beta (mean ~0.25, e.g. Beta(2,6)), meaningful mass near 0",
      "low": 0.05,
      "mode": 0.2,
      "high": 0.5,
      "source": "GAO-24-106681 / GAO-26-107856 list the rule as one of several co-causes, never attributing service loss to it alone; unions dispute a shortage."
    },
    {
      "name": "rule_binding_fraction_on_EAS_service",
      "meaning": "Fraction of at-risk EAS service where the rule actually binds; gates out routes already on <=9-seat Part 135 (no SIC ATP / 1,500-hr; Part 135 PIC needs only 1,200 hr).",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "Beta, well under 0.5",
      "low": 0.2,
      "mode": 0.4,
      "high": 0.6,
      "source": "CRS R44176: ~44 of ~110 EAS communities use 8-9 seat aircraft; DOT EAS FAQ."
    },
    {
      "name": "displaced_roundtrips_per_community_per_yr",
      "meaning": "Annual round-trip O&D travelers per lost community at the EAS margin.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "lognormal centered ~3000",
      "low": 3000,
      "mode": 3000,
      "high": 10000,
      "source": "EAS 10 enplanements/service-day floor -> ~3,650 outbound/yr; matches white-paper 3,000; upper tail for better-served 50-seat communities (DOT EAS FAQ; 14 CFR 398)."
    },
    {
      "name": "fraction_who_drive",
      "meaning": "Fraction of displaced travelers who substitute a road trip (vs forgo travel or drive-to-distant-hub-then-fly).",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "Beta, mean ~0.5",
      "low": 0.3,
      "mode": 0.5,
      "high": 0.7,
      "source": "No service-availability-specific estimate; 9/11 fear-shock (Blalock-Kadiyali-Simon) over-states it. UNVERIFIED for this scenario; net scales linearly. Must NOT be set to 1."
    },
    {
      "name": "added_oneway_road_miles_to_hub",
      "meaning": "Incremental one-way drive distance to the alternate hub (RT = x2).",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "lognormal centered ~150 mi one-way",
      "low": 75,
      "mode": 150,
      "high": 250,
      "source": "White-paper planning assumption 150 one-way / 300 RT; EAS 175-mi-from-hub rule supports long-haul skew. UNVERIFIED population mean; planning dial."
    },
    {
      "name": "road_fatality_rate_per_100M_pax_mile",
      "meaning": "Road fatality rate for substituted (rural/long-haul-skewed) trips, per 100M passenger-miles.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "lognormal",
      "low": 0.73,
      "mode": 0.9,
      "high": 1.1,
      "source": "Savage 2013 car/light-truck 7.3 per billion pax-mi = 0.73 per 100M pax-mi; ~1.3x rural uplift (IIHS rural 1.65 vs national 1.26 VMT). Occupancy ~1.6 (1.2-1.8) carried separately for VMT conversion."
    },
    {
      "name": "displaced_air_leg_fatality_rate_per_100M_pax_mile",
      "meaning": "Fatality rate of the displaced scheduled commuter/Part-135/50-seat regional leg, netted OUT of the cost (avoided air deaths), per 100M passenger-miles.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "lognormal spanning mainline to modestly-worse commuter",
      "low": 0.007,
      "mode": 0.03,
      "high": 0.2,
      "source": "Savage 2013 mainline 0.07 per billion = 0.007 per 100M; commuter ~1 order above mainline, still 2-3 orders below road. CRITICAL: NEVER substitute GA rate (~10-30x higher)."
    },
    {
      "name": "vehicle_occupancy_passengers_per_car",
      "meaning": "Party size to convert vehicle-miles to passenger-miles; explicit scalar to prevent double-counting.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "triangular",
      "low": 1.2,
      "mode": 1.6,
      "high": 1.8,
      "source": "IIHS / BTS occupancy; used only where a VMT-based rate is converted to passenger-miles."
    },
    {
      "name": "part121_passenger_fatalities_per_decade_post2010",
      "meaning": "Observed base rate the benefit term must fit inside (sanity anchor, not free).",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "near point-mass ~0-2",
      "low": 0,
      "mode": 1,
      "high": 3,
      "source": "NTSB/BTS: ~2 pax fatalities in 15 yr since 2009 (PenAir 1; Atlas 3 cargo crew)."
    },
    {
      "name": "rho_AB_benefit_cost_coupling",
      "meaning": "Correlation between the benefit side and the cost side through the shared 'how-binding-is-the-rule' latent; enforces the correlated diagonal, forbids independent corners.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "fixed high-positive (Gaussian copula)",
      "low": 0.6,
      "mode": 0.75,
      "high": 0.9,
      "source": "White-paper co-movement requirement; GAO. Admissible region is the diagonal, NOT the independent-corner rectangle."
    },
    {
      "name": "pathway_x_TFH_interaction",
      "meaning": "EXPLORATORY: slope of accident risk vs TFH for structured-cadet/MPL relative to US time-builder over 250-1500 hr.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "wide near-null Normal(0, large) on log-HR interaction; BOUND not estimate",
      "low": -0.5,
      "mode": 0,
      "high": 0.5,
      "source": "No published MPL-vs-time-builder accident comparison; structurally non-identifiable (extreme selection funnel). Uncertainty-widener only, never a point shift toward zero benefit (Wikipedia MPL; FSF)."
    },
    {
      "name": "decision_standard",
      "meaning": "Normative decision rule under which the net sign is read: realized-evidence vs precautionary/maximin. A PARAMETER, reported under BOTH values.",
      "shared_or_hypothesis_dependent": "shared",
      "distribution": "categorical, both reported",
      "low": 0,
      "mode": 0,
      "high": 1,
      "source": "Methodologist/ALPA: the net sign is standard-dependent; the choice is normative-not-empirical and both must be disclosed (white-paper integrity coupling)."
    }
  ],
  "decision_criteria": "REPORTED REGARDLESS OF OUTCOME (no selection): (1) the FULL coupled net_lives distribution with pre-specified quantiles 5/25/50/75/95 and E[net], under EACH of H_A and H_B; (2) P(net_lives < 0) under EACH hypothesis, with Monte Carlo standard error from 1,000,000 draws; (3) sensitivity of P(net<0) and E[net] to rule_attributable_share and to each of the four linear cost dials (communities, displaced_roundtrips, fraction_who_drive, added_road_miles), reported one-at-a-time AND jointly; (4) the net SIGN under BOTH the realized-evidence standard (use posterior expectations) and the precautionary/maximin standard (weight the non-negative benefit floor and the worst-case cost tail) -- both signs reported, the choice labeled normative-not-empirical. HOW THE HYPOTHESES DIVERGE: H_A and H_B share every cost-side parameter byte-for-byte and the single rho_AB-coupled binding-ness latent; they diverge ONLY in lives_saved_per_decade (H_A median ~2, P5~0.5; H_B median ~0.3, P5=0, both negligible left tail). SUPPORT CRITERIA: H_A is supported if, under the realized-evidence standard, E[net]>0 and P(net<0)<0.5 driven by a benefit median materially above the cost median; H_B is supported if E[net]<=0 or P(net<0)>=0.5 under realized-evidence, i.e. the small near-zero benefit cannot overcome the coupled (attribution- and binding-gated) road-cost term. EXPLICIT PERMITTED CONCLUSION: 'sign indeterminate, magnitude small' is a valid and reportable outcome if the net distribution straddles zero under both standards; the analysis is required to stop there rather than manufacture a sign. The reported estimand set (full distribution + P(net<0) + E[net], under both hypotheses and both standards) is locked now and may not be narrowed after seeing outputs.",
  "integrity_audit_checklist": [
    "BENEFIT-SIDE BAND-SWAP: confirm the write-up does NOT import the Li/Baker 5,000-9,999 hr RR=0.43 gradient into the 250-1,500 benefit term; it appears only as an out-of-band curve-bounding reference.",
    "BENEFIT NOT SCORED AT HARD ZERO: confirm H_B's lives_saved_per_decade prior is one-sided non-negative with P5=0 but positive mass above zero (E[saved]>0), not a point mass at 0 -- the single sign-flipping illegitimate move.",
    "AIR-COMPARATOR LOCK: confirm the displaced air leg uses the scheduled commuter/Part-135/50-seat rate (0.007-0.2 per 100M pax-mi), NEVER the GA rate (~10-30x higher).",
    "SHARED-VS-SPLIT INTEGRITY: confirm every cost-side parameter is byte-identical across H_A and H_B; only lives_saved_per_decade differs. Audit the two parameter dumps for any silently-diverged shared value.",
    "COUPLING ENFORCED: confirm a single rho_AB-coupled binding-ness latent drives both sides on the correlated diagonal; confirm NO hand-paired incompatible corners (Side A floor vs Side B ceiling) appear anywhere in the reported net.",
    "OUTCOME-LEVEL MATCHING: confirm Pilot Source Study training-completion and MPL pass-rate evidence are used only as intermediate-outcome plausibility, never promoted to an accident-rate (terminal) benefit claim; confirm cost and benefit are scored at the same inferential level.",
    "FULL REPORTING / NO QUANTILE CHERRY-PICK: confirm the full net distribution (5/25/50/75/95), E[net], and P(net<0) are reported under BOTH hypotheses -- no selective quantile or scenario.",
    "BOTH DECISION STANDARDS REPORTED: confirm the net sign is given under BOTH realized-evidence and precautionary/maximin, with the choice labeled normative-not-empirical -- no decision-standard shopping.",
    "NO HARKing: confirm the two hypotheses and all priors match this locked document's git hash; any deviation is logged as a timestamped amendment and reported as such.",
    "ATTRIBUTION SYMMETRY: confirm the same causal-attribution rigor is applied to both arms -- the road-death cost is gated by rule_attributable_share AND rule_binding_fraction (not assumed =1), mirroring the strict bar applied to the benefit.",
    "NO DROPPED ANALYSES: confirm all pre-specified sensitivity sweeps (rule_attributable_share + four cost dials, one-at-a-time and joint) are reported even where null or unflattering.",
    "PRIORS SOURCED NOT TUNED: confirm every prior cites its source from this document and none was re-centered toward a desired net after seeing draws.",
    "BOTH HYPOTHESES REPORTED EVEN IF UNFLATTERING: confirm that if H_B (or H_A) is unflattering to FRISA's prior view, it is still reported with equal prominence; confirm the permitted 'sign indeterminate, magnitude small' verdict was not overridden toward a preferred sign.",
    "EXPLORATORY LABELING: confirm pathway x TFH, recency, covariate-attenuation, left-tail-vs-central, and subgroup splits are explicitly labeled exploratory and never used to flip the confirmatory headline sign.",
    "UNVERIFIED-CLAIM FLAG: confirm the European MPL/cadet accident-equivalence is flagged as unverified/non-identifiable and encoded as an uncertainty-widener, never asserted as positive evidence that the US benefit is zero."
  ]
}