Freeze point: disability/docs/research-inquiry.md as of protocol draft 2026-08-09 (commit carrying the GBMT-12 protocol on main). Per M6, each divergence from the frozen protocol is recorded with its effect. Entries are one line each.
| # | Date | Deviation | Reason | Effect |
|---|---|---|---|---|
| 1 | 2026-08-11 | Protocol freeze recorded at Phase 0 close (empty log at start). | Phase 0 executed 2026-08-11 from origin/main after GBMT-11 Phase 0 merged; freezes the 2026-08-09 protocol text as the commitment. |
Freeze is retrospective to protocol draft; no protocol text was altered mid-pass. |
| 2 | 2026-08-11 | Anchor 5 prior (“mental disorders ≈ 25–35% of working-age SSDI awards”) replaced by SSA awards share 12.7% (2023 disabled-worker awards); stock ≈ 28.6% carried separately. | DI ASR Chart 10 contradicts the awards prior; Chart 6 stock shares explain how the circulating “one-third” prior likely mixed constructs. Confirms GBMT-11 Phase 0 correction — do not reintroduce. | H8 award-share leg (≥25%) fails on awards; rewrite before support. Stock/SSI legs remain for §8. |
| 3 | 2026-08-11 | Anchor 1 SSDI worker prior band 7.5–8.5M carried as 7.37M (Dec 2023) — just below the prior floor. | DI ASR current-pay table is authoritative. | Baseline uses 7.37M workers / 8.71M all disabled beneficiaries with concurrent caution. |
| 4 | 2026-08-11 | KC1 treated as confirmed on definitional grounds (IP ≠ intentional fraud) without a single circulating “X% fraudulent” media root-trace in this pass. | Protocol allows H3 support if primary IP publication attributes dollars to error/complexity/status change or single-root citogenesis; AFR cause categories suffice for the stronger definitional form. | Media citogenesis audit still owed in §8; measurement headline stands on AFR/OIG. |
| 5 | 2026-08-11 | H5 <2pp employment bar marked “supported directionally” pending §5 table extract from Mathematica reports. | Bulletin synthesis + Mamun/Page establish null/undetectable ITT effects; exact pp threshold needs one archived table. | H5 remains open-but-live; Ticket architectures scored small pending extract. |
| 6 | 2026-08-11 | H1 adjudicated SUPPORTED using Liebman’s post-early-1990s majority (⅔ aging + women’s insured) rather than a single pooled 1980s–2010s share. | Protocol window spans a policy-dominated 1980s swing Liebman himself separates; recent-decade adjusted-incidence leg is unambiguous. | H1 support carries explicit periodization; Autor–Duggan band for 1984–~2003 retained as conflict, not a veto. |
| 7 | 2026-08-11 | Anchor 10 verified on order-of-magnitude (CDR cessations ≫ CDI judicial actions) without a same-year paired SSA CSV extract. | Periodic CDR CSV returned access-denied; used CRR/OIG/agency narrative counts instead. | Direction locked for H7(c); exact paired-year table remains polish for §7. |
| 8 | 2026-08-11 | H2 ALJ case-mix leg rests on GAO-18-37 (FY2007–2015) without a newer equal-quality re-estimate. | No post-2015 public case-mix study located; OIG outlier audits corroborate residual discretion. | H2 supported; current-year residual pp band is moderate-confidence. |
| 9 | 2026-08-11 | H4 second leg scored via Medicaid buy-in / coverage-cliff evaluations vs info-only nulls, not via BOND mean-earnings gains. | Protocol asks cliff/offset designs that move earnings more than information-only; BOND cash offset is null on mean earnings (anchor 9), so coverage-separation studies (WA MBI; Mathematica MBI series) carry the comparative leg. | H4 SUPPORTED with explicit BOND constraint on cash-offset architectures. |
| 10 | 2026-08-11 | H5 <2pp bar locked from ITT null/undetectable STW effects plus 0.1–0.4pp enrollment impacts; no single table cell labeled “employment pp.” | Stapleton/Mamun/Page power language and Wittenburg et al. enrollment pp are the primary extracts; absolute STW effects under detection sit ≪2pp given ~0.5% base exit rates. | Deviation 5 closed; H5 SUPPORTED. |
| 11 | 2026-08-11 | Child SSI Dec 2024 count (~1.00M) noted in §6 but adjudication uses Dec 2023 983,169 to match Phase 0 vintage. | Keeps §1/Phase 0 construct lock; CRS supplies the 2024 point without reopening the series. | Anchor 11 verified; no material change to H6. |
| 12 | 2026-08-11 | H8 scored SUPPORTED-WITH-DEVIATION: size leg rewritten from ≥25% of working-age awards to ≥25% of disabled-worker stock (≈28.6%) or SSI under-65 (~60%); awards share remains 12.7% (fails frozen threshold). | Phase 0 / deviation 2 already falsified the awards prior; protocol text still said “awards”; stock/SSI legs preserve the seam claim without reintroducing one-third-of-awards. | H8 supported on rewritten size + independence legs; awards ≥25% stays failed. |
| 13 | 2026-08-11 | H7(b) same-claim eventual-allowance leg rests on process structure + published cohort/process syntheses (NAP/Urban), not a new DRF microdata extract. | Hearing dockets are appeals of prior denials by design; protocol bar met without withheld microdata. | H7 SUPPORTED; whitepaper may note no new matched cohort table this pass. |
| 14 | 2026-08-11 | Media citogenesis audit uses OIG ~$72B FY2015–22 improper payments → briefing/media fraud relabel as the single high-circulation root (Poynter-documented). | Phase 0 owed one root-trace; this closes deviation 4’s media gap without weakening AFR definitional KC1. | H3 reinforced; citogenesis root archived in §8. |
| 15 | 2026-08-11 | Scorecard §9 uses symmetric floor (below-neutral only with cited adverse/undermining evidence; absence → 3) and explicit weight vectors rather than named-objective×3 sums alone. | Imports housing/elder-care/media scale correction (M6/M10); protocol §3 readers need reproducible vectors. | Pass-1 ranks are vector-defined; unweighted means disclosed but not treated as a reader. |
| 16 | 2026-08-11 | Fraud-theater / prosecution-centric integrity not entered as an 11th scored row; demotion stated relative to architecture #6 (CDR + wage reporting). | Protocol candidate list already replaced fraud theater with #6; scoring a ghost row would invent an architecture. | Demoted result reported in scorecard rankings; #9 carries the scored “get tougher first” last-place finding. |
| 17 | 2026-08-11 | Architecture #3 O4 scored 2 on BOND cash-offset null mean earnings; automatic 1619(b)/reporting arm not separately row-split. | Protocol scores the cliff-redesign bundle against BOND; splitting would require a new candidate not in the frozen list. | Sensitivity noted: lifting O4 to 4 still leaves #3 behind #4/#8/#1 under beneficiary weighting. |
| 18 | 2026-08-11 | Red-team (M6): #1 O3 4→3; #8 O3 4→3; #9 O1 1→2; fraud-theater ghost scores published; “#9 last under all three” withdrawn. | Elder-care evidence-floor (above-neutral needs cited outcomes; O1=1 needs instrument amplify, not advocacy-only); #8 integrity solo lead was scale artifact. | Tops: backlog #1=#2; might-work #4; integrity #6=#8. Rank-stable top set #1/#4/#6. Log: ws09-red-team-log.md. |
| 19 | 2026-08-11 | Structurally blinded re-score reconciled (batch msgbatch_013vUDQ6mt6JV8mz2H95zfsx): 6 cells moved (#2 O1 3→4; #3 O5 3→4; #6 O3 3→2; #7 O2 2→3; #10 O1 2→3; #10 O2 2→3); 2 judgment splits kept published (#3 O3; #8 O5). |
M6/M10 owed pass; elder-care floor on unevidenced below-neutral; cited cessations demote #6 O3; BOND executability lifts #3 O5. | Tops: backlog/integrity #2; might-work #4. Rank-stable #1/#2/#4/#8. #9 last under all three by pass-2 arithmetic (#10 floor lift), not a quiet red-team resurrection. Log: ws09-rescore-log.md. |