MED01 — Corrections & Version History
No substantive correction is silently overwritten. Corrections inherit the full verification burden of the original claim. Post-freeze clarifications to the brief are logged here too — the frozen brief itself is never edited.
Corrections
| # | Date | Affected artifact (+ version) | Original statement | Correction | Reason | Supporting evidence (S-ID) | Verification performed | Claims affected | Commit |
|---|---|---|---|---|---|---|---|---|---|
| 16 | 2026-10-14 | C077 (ledger + disposition); synthesis §8, §13; plain C077; summary | Whiting AE ORs held as two-model convergence | Verified: any AE OR 3.03 (2.42–3.80; 29 trials/3,714), serious OR 1.41 (1.04–1.92; 34/3,248), withdrawal OR 2.94 (2.18–3.96; 23/2,755); C077 back to Accepted | Table 3 of the corrected JAMA article, screenshot by Danny | S-402 | DOC via Danny’s browser | C077 | this commit |
| 15 | 2026-10-14 | C078, C077 (ledger); synthesis §8, §9, §13; plain C078 | Corrected Whiting values held as convergence | Ashworth WMD −0.12 (−0.24 to 0.01; 5 trials, NS) verified; corrected abstract figures and “increased risk … including serious AEs” verified; NRS −0.76 and Table 3 ORs still held | JAMA corrected abstract + 2016 correction notice opened by Danny (screenshots); notice records the 2015-07-13 Ashworth correction | S-402 | DOC via Danny’s browser (verifier bot-blocked) | C078, C077 | this commit |
| 14 | 2026-10-14 | C052 (ledger); plain C052 | Ware 2010 nabilone: “authors reported nabilone effective for the primary outcome, sleep quality” | ISI difference 3.2 (1.2–5.3); LSEQ restfulness 0.5 (0.0–1.0) marginal, wakefulness 0.3 (−0.2 to 0.8) not; no effect on pain, mood or QoL; more AEs. Plain ledger gains the study | Danny’s §M10a spot check, item 3 (“specify restfulness but not wakefulness”) | S-351 | abstract re-read by verifier | C052 | this commit |
| 13 | 2026-10-13 | C078 (ledger); synthesis §9; plain C078 | Ashworth WMD −0.36 (−0.69 to −0.05; 7 trials) | Original value marked as later corrected; corrected values reported (Ashworth −0.12 NS; NRS −0.76) as convergence; all Whiting pooled estimates held; disposition → Accepted with qualification | Whiting 2015 has 4 PubMed ErratumIn notices; verifier used the uncorrected abstract (cross-audit B3-01, most serious) | S-402 | PubMed CommentsCorrections read; erratum text JAMA-blocked | C078, C048 (note) | this commit |
| 12 | 2026-10-13 | C071; C077; synthesis §8, §13; plain C071, C077 | “8 IQ points” held; Whiting AE ORs held, not shown | Meier 8 IQ points added for the n = 23 subgroup (PMC full text); Whiting AE / withdrawal / serious-AE ORs added as two-model convergence | Cross-audit HELD report H8, H9; B2-04 harms parity | S-419, S-402 | Meier: PMC full text read; Whiting: not readable | C071, C077 | this commit |
| 11 | 2026-10-13 | C062; C064; synthesis §5; plain C062, C064 | UK “June 2010”, exact day held; TGA/AEMPS held; Gemini BIA counts “not supported” | UK 16 June 2010 and Spain 20 July 2010 added (verified); TGA convergence; IGAS regulatory cohort added as convergence; earlier ruling withdrawn as overbroad | Cross-audit H4, H5, B2-03 | S-446, S-447, S-394 | emc SmPC §9 and AEMPS ficha técnica §9 opened; IGAS 404 | C062, C064 | this commit |
| 10 | 2026-10-13 | C053; C055; critique §4; synthesis §4; plain C053, C055; summary | Meiri “similarly effective”; Tramèr remark against Gemini; “214 children” | Meiri response rates (54% vs 58%), not equivalence; Tramèr remark withdrawn; Devinsky 2016 ages 1–30 | Cross-audit B3-02, B3-03, B3-05 | S-355, S-264, S-364 | abstracts and PMC full text read | C053, C055 | this commit |
| 9 | 2026-10-13 | C077; synthesis §8; plain C077; summary | 96.6% of adverse events across 31 studies | 96.6% = 4,615 of 4,779 events within the 23 RCTs; serious RR 1.04 (0.78–1.39) added | Cross-audit B2-01 | S-435 | abstract read | C077 | this commit |
| 8 | 2026-10-13 | C067; C041 origin wording; critique §6b/§6d; synthesis §6; plain C067; summary | 112 / 31 / 62 / 19 / 32 / 9 chain; NCT01898520 as “no publication identified” | Re-screen: 132 / 44 / 88 / 28 therapeutic / 17 matched / 11 unresolved; NCT01898520 published 2020 (S-448, retrospective); C041 “fabricated” reworded to “presented as live, admitted estimated” | Cross-audit B2-02; verifier never ran the scheduled title search; the original filter included 12 non-therapeutic studies | S-403, S-448 | all 132 re-queried (0 errors); 28 searched by title/PI/sponsor | C067, C041, C084 (count reference) | this commit |
| 7 | 2026-10-13 | C065; synthesis §5; plain C065 | “CBD for diabetes” | “CBD for Type II diabetes” | Cross-audit B1-23 | S-399 | Google Patents claim 1 read | C065 | this commit |
| 6 | 2026-10-13 | C074 | “FEV1/FVC slightly higher” | “FEV1 and FVC were slightly higher at low exposure; nonlinear” | Cross-audit B1-30 (ratio vs separate measures) | S-430 | abstract + PMC full text read | C074 | this commit |
| 5 | 2026-10-13 | C048 | Wade 2004 “primary composite not significant” | “primary outcome, a VAS score for each patient’s most troublesome symptom, not significant”; Novotna −0.84 recorded as convergence | Cross-audit B1-4 | S-325 | abstract read | C048 | this commit |
| 4 | 2026-10-10 | ledger P2 v1.0 — end-of-era statement (not a claim) | Cancer appetite/cachexia (C054) listed under “Preliminary human evidence … single or small trials” | Moved to “Replicated or controlled human evidence identified → several randomized trials”; HIV appetite and anorexia-nervosa evidence stays in the second group | Placement must follow design, not result: two large RCTs (469 and 243 patients) are controlled evidence even though neither showed benefit (ChatGPT synthesis editorial finding 3) | S-356, S-357 | ledger C054 text re-read; claim unchanged | C054 (placement only) | this commit |
| 3 | 2026-09-26 | synthesis v1.0 §4 | “…41 subjects worldwide.” (no tag) | “…(C020; S-286 §5.20, verifier-found).” | Gemini cross-audit: verifier-found datum lacked a claim tag | S-286 | tag only | C020 | this commit |
| 2 | 2026-09-26 | ledger v1.0 C019; synthesis v1.0 §4; both plain-English v1.0; ledger C001 qualification | C019: “per-patient breakdown not in abstract”; C001: MeSH Cannabinoids “dates from 1971” | C019: breakdown (4/8, 3/8, 1; 7/1) inserted — it IS in the PubMed abstract; C001: MeSH year → 1977 | ChatGPT cross-audit; verifier re-fetched PMID 7413719 (2026-09-23 read had been truncated — verifier error) and NCBI MeSH record (“Year introduced: 1977”) | S-273; NLM MeSH D002186 | live re-fetch | C019, C001 | this commit |
| 1 | 2026-09-26 | ledger v1.0 → C024, C035 (dispositions); C018, C028, C036 (notes); synthesis v0.1 → v0.2 | C024 US component “pending verifier open”; C035 recommendation + Govt response “HOLD”; C018 both sources “AB” | C024 US verified; C035 Accepted, HOLD lifted; C018 Tashkin 1973 downgraded to BIB and its result removed from synthesis | Verifier page checks completed (FDA 2016 testimony; HoL ch. 8; Lords Hansard 3 Dec 1998 cols 704–705); PubMed carries no abstract for PMID 4718513 | S-281, S-299, S-300 (DOC); S-272 (BIB) | live open of each page; quotes recorded in ledger | C018, C024, C035 (+ notes C028, C036) | this commit |
Version history
| Artifact | Version | Date | What changed | Why |
|---|---|---|---|---|
| MED01-P2 (ledger, synthesis, plain-English pair) | P2 LOCK | 2026-10-14 | Locked by Danny after the §M10a cross-audit and corrections #5–#16; artifacts change only by logged correction from here | Open items carried as listed in round-index (Version-lock status) |
| Ledger P2 + synthesis P2 + plain-English pair P2 | v1.0 + corrections #5–#13 | 2026-10-13 | §M10a cross-audit: ChatGPT audit accepted (findings confirmed at source), Gemini audit rejected (fabricated quotations); corrections approved by Danny | Two claims → Accepted with qualification (C077, C078); held list rebuilt |
| Plain-English pair P2 (ledger + summary) | v1.0 | 2026-10-12 | Signed off by Danny; text = v0.2 (ChatGPT §M11 fidelity review in three batches: 24 findings, 23 accepted + 1 partly) + 2 consistency edits from ChatGPT’s re-review (registry denominator context; “non-serious”) | ChatGPT: sign-off recommended after the 2 edits |
| Interim synthesis P2 | v1.0 | 2026-10-12 | Signed off by Danny; text = v0.2 (ChatGPT editorial pass, 13 findings, all accepted) | ChatGPT: “ready for sign-off after these changes: yes” |
| Claim ledger P2 | v1.0 | 2026-10-10 | Signed off by Danny; text = v0.2 + 2 documentation corrections (C067 denominator, corrected to the 62-record base; held-list heading); S-308–S-445 + 9 P2 cautions appended to SOURCE_LIBRARY; correction #4 the same day (end-of-era C054 placement) | ChatGPT review of v0.2: sign-off YES |
| P2 — Gemini Stage 1 RERUN (decision revised) | — | 2026-10-06 | Danny reversed the same-day decision below: because Q1–Q5 cover major P2 topics, Gemini’s Stage 1 is rerun in a fresh Gemini Spark session under a separate Google account (second) that has no access to the project Drive. The first Gemini Stage 1 (c519cc4) and its disclosure (197dc39) stay filed, marked superseded — not used as Gemini’s Stage 1 pass; kept as postmortem evidence of how probe exposure shapes a map. ChatGPT’s Stage 1 is unaffected. Stage 2 goes to both researchers only after the rerun is filed. | Danny |
| P2 protocol deviation — Gemini Stage 1 probe exposure (decision superseded by the row above) | — | 2026-10-06 | Gemini located the brief by searching the Drive mirror (contrary to brief §0b). Search snippets showed it the Probe Appendix title, header, “How to answer” text and §6 heading with Q1–Q5, plus a round-index snippet on the two-stage design and the archive audit’s existence (no archive content). Firewall between researchers not breached (ChatGPT’s pass and archive audits are git-only). Danny’s decision: keep Gemini’s Stage 1 as filed, disclose the exposure, proceed to Stage 2; no rerun. Critique to treat Gemini’s coverage of Q1–Q5 topics (approvals, tumour biology, condition list, paediatric epilepsy, constituent interaction) and any convergence with ChatGPT on them as weaker evidence of independent retrieval. Postmortem item: mirror the Probe Appendix only after both Stage 1 passes are filed; researchers’ Drive connectors off. Evidence: raw-inputs/gemini/MED01-P2-gemini-stage1-disclosure.md (197dc39) | Danny |
| P2 brief (core + Probe Appendix) | v1.0 FROZEN | 2026-10-03 | Frozen by Danny after ChatGPT/Gemini editorial passes (v0.1 → v0.2), Gemini’s confirmation and ChatGPT’s implementation check (1 minor wording fix to appendix §6); two-stage pass adopted | Freeze gate; dispositions in reviews/MED01-P2-brief-editorial-passes.md |
| Round infrastructure | — | 2026-09-18 | Templates created | MED01 setup |
| P1 LOCK | — | 2026-09-26 | All §M10 checklist items met: passes filed; verification complete; ledger v1.0 (+#1–#3); technical synthesis v1.0 (+#2–#3); plain-English pair v1.0 (+#2); both §M10a cross-audits filed, issues resolved (#2, #3) or documented (5 HOLD components with convergence). Registered in knowledge-core/AUDIT_LOG.md. | Danny |
| Plain-English ledger + synthesis | v1.0 | 2026-09-26 | Signed off by Danny; text = v0.2 | Sign-off gate |
| Plain-English ledger + synthesis | v0.2 DRAFT | 2026-09-26 | ChatGPT §M11 fidelity review, 12 ledger items + 5 synthesis sentences + closing line, all applied. Most consequential: C030 “the authors called cannabis a cause” → “described cannabis as an independent risk factor; an observational study cannot by itself establish causation” (Claude’s drift, both documents); C027 “no honest ratio is possible” → bounded to this evidence map; C038/C039 absence language restored to “identified/located”; C028 HOLD moved into the sentence; C036 label → Open; C040 “wrong” → “not supported by this record”. | §M11: simplify the language, never the uncertainty |
| Interim synthesis | v1.0 | 2026-09-26 | Signed off by Danny after ChatGPT’s final calibration: §10 P6 stale [HOLD] removed (C035 verified); §11 end-of-era split C017/C018 so aerosolised-THC bronchodilation is not read as smoked-cannabis bronchodilation (D4/D5/D10 — Claude’s slip after the Tashkin-1973 downgrade); §1 integrity-check scope narrowed to journal DOIs. | Sign-off gate; synthesis never exceeds the ledger |
| Interim synthesis | v0.2 DRAFT | 2026-09-26 | ChatGPT editorial calibration, 13 items, all applied (status wording; “40 verified” → “40-claim signed ledger”; unledgered “working assumption” sentence removed; crossover scope; Tashkin 1973 result withdrawn pending text; “went no further” removed; C028 sentence narrowed; C033 “plausible” removed; WHO excision sentence deleted — ledger/synthesis mismatch; §11 C021/C014 precision; §12 HOLD list regenerated; §13 P2 handoffs separated from unfinished P1 verification). Verifier lifted HOLDs on C024-US and C035 by opening FDA/HoL/Hansard; C036/C028 stay HOLD on verifier side (Bookshelf blocked) with ChatGPT’s read recorded as convergence. | Synthesis may not exceed the ledger; HOLDs lift only on verifier reads |
| Claim ledger | v1.0 | 2026-09-25 | Signed off by Danny; text = v0.2; S-241–S-307 + 3 paper-specific cautions appended to knowledge-core/SOURCE_LIBRARY.md | Sign-off gate passed after ChatGPT approval of v0.2 |
| Claim ledger | v0.2 DRAFT | 2026-09-25 | ChatGPT pre-signoff editorial calibration, 13 items, all applied: C001 exact two-query counts; C002/S-242 → primary × observational (bibliometric); C003 periodization made explicitly editorial; chemistry design tag in-vitro → mechanistic (chemical/structural) in C004–C006; C006 “61” → Inconclusive subclaim; C015 Tramèr quantitative / Campbell qualitative; C022 dronabinol drift fixed, NIDA-cigarette parenthetical moved to C025; C024 Canada out of claim sentence, US pending S-281 open; C025 → Inconclusive/HOLD (sources unopened); C027 reworded as absence-of-dataset claim → Accepted w/ qual, NIMH $ figures held; C034 excision story → Q8 note; C035 recommendation + Govt response → HOLD; C036 → HOLD for verifier support check; C040 pre-1996 wording; end-of-era cognitive/psychomotor line removed (C031/C032 did not support it — Claude’s error); verifier-found marking on C020/C024; two researcher-performance cautions rows moved to verification log. No claim added or retired. | Ledger cannot let an unopened source carry Accepted, nor one unresolved component contaminate a verified claim (§M9); provenance of verifier-found facts must stay visible for §M10a |
| PROTOCOL-MEDICAL + MED01 templates | v0.5 BASELINE | 2026-09-18 | Baseline commit 7beaff6 — protocol v0.1→v0.5 through ChatGPT (3 rounds) and Gemini (adversarial) review; full trail in ../PROTOCOL-MEDICAL.md changelog and ../reviews/ |
First committed state of the medical track; all later changes are measured against it |