GUBMENTPlain talk · policy frontier
Filings / Drugs / Sources / Verification Log — GBMT-8 (Drugs), Pha
GBMT-8 · Research record · No. 8

Verification Log — GBMT-8 (Drugs), Phase 1 fact-check

drugs/research/verification-log.md
This is a working research document from the drugs filing, published as written — including the parts later corrected. It is the underlying record for Whitepaper No. 8, not a summary of it.

Date: 2026-08-10 Protocol: method/verification-protocol.md, Phase 1 Filing under check: Whitepaper No. 8, filed 2026-08-03 — site/drugs/index.html, site/drugs/sources/index.html, and the full research record in drugs/docs/ and drugs/research/. Independence (V1): run in a session that did not author the filing, did not run its red team, and did not run its blind re-score. The filing's own citations were treated as claims to be re-derived, not as verification.


Coverage numbers (V3)

Count
Claims extracted 393
Verdicts recorded 393
Unaddressed 0

Breakdown: 122 from the public whitepaper, 63 from the public sources page, 208 from the research record that do not appear on the public pages.

Extraction-quality guardrail

Extraction was not delegated — it was performed directly by the adjudicating model, which is stronger than the protocol's minimum (the childcare lesson was that weak extractors silently drop inferential claims). The guardrail was still run against my own ledger: Part 2 of the whitepaper was independently re-extracted by mechanical sentence-split and diffed against the first-pass ledger. The diff found 2 misses out of 20 (10%), both of the same class — section headers and stamp text, which carry assertions (the best-evidenced treatments are the most legally cornered; Binding constraint) but sit outside body prose. That miss class was then swept across every section, adding 10 claims and taking the ledger from 383 to 393. Recording the rate rather than asserting completeness is the point of the check.

Verdict distribution

Verdict All Whitepaper Sources page Record only
CONFIRMED (primary-verified) 140 60 25 55
CORRECTED 23 10 8 5
OVERSTATED 34 23 4 7
STALE 11 5 3 3
UNVERIFIABLE 185 24 23 138
Total 393 122 63 208

Counts are computed from an explicit per-claim assignment, not estimated; the assignment partitions all 393 IDs with no duplicates and no gaps.

Two rules govern this distribution, and both should be read before the numbers are.

The cascade rule. A claim whose only defect is that one of its inputs could not be reached is marked UNVERIFIABLE, not CONFIRMED. So "alcohol deaths exceed the overdose peak by more than half" is U — the arithmetic is fine, but the 178,307 input is behind a 403. This is strict, and it is the protocol's rule (V2: re-derive every claim you examine from sources you fetch yourself). It means UNVERIFIABLE here measures this pass's reach at least as much as it measures the filing's sourcing.

Where the 185 sit. Three concentrations, none of them evenly spread. cdc.gov is comprehensively bot-blocked — 403 on the MMWR paths, the data briefs, and stacks.cdc.gov — so the alcohol and tobacco mortality figures stay at secondary tier, which is why the filing's own masthead verdict, "the legal drugs kill more," is UNVERIFIABLE by this pass. Not contradicted: unreached. That is worth saying plainly, because it is the single sentence the whitepaper leads with. Second, the criminology base under Part 4 (HOPE's replication, the Campbell drug-court review, the focused-deterrence systematic review) sits outside the indexes reachable here. Third, §7 and §8 rest on commercial trackers, watchdog 990 readings and state-agency data that have no deeper tier to push to — the filing labelled most of these moderate-confidence, and it was right to.

The number that carries real weight is the other one: 57 claims moved (23 CORRECTED, 34 OVERSTATED), and 18 of the 23 corrections are on the public pages. (These counts are post-audit: an independent adversarial audit on 2026-08-10 reversed this pass's own verdict on the June 2025 nowcasting artifact — see headline finding 3 and deviation #22 — moving A53, B11 and C14 from CORRECTED to CONFIRMED and A54 from UNVERIFIABLE to CONFIRMED.)


Headline findings

Seven items change what the public pages say. Ordered by how much they carry.

The whitepaper's abstract and Part 2 say contingency management (CM) is "capped … by a federal anti-kickback rule one agency could lift tomorrow" and "was capped for years at a third of its trial-effective incentive level by a federal anti-kickback policy one office can change on its own." Part 6 makes CM the top-ranked architecture and Part 7 makes it the filing's first move.

Three primary sources, fetched this pass, say otherwise:

So the barrier was not an anti-kickback cap one office could lift. It was a grant-program condition (already lifted, to $750, in January 2025) layered on top of a statute that OIG cannot repeal by rulemaking — it can only add a safe harbor, and only "in consultation with the Attorney General" (42 U.S.C. 1320a-7d(a)(1)(B)).

This was already found inside the filing's own record. ws14-cm-legal-status.md, dated 2026-08-06, records the same correction and says it "supersedes any statement that CM has a settled nationwide safe-harbor expansion or an OIG $75 'cap'." Deviation #20 logs it. The public pages were never updated. Four days of live whitepaper have carried a superseded legal claim as the filing's headline recommendation. This pass re-derived the correction independently from primary text rather than adopting ws14's conclusion, and it holds.

2. "A federal rulemaking process already underway" is not underway — CORRECTED / STALE

Part 2 says OIG "opened a formal review in November 2024"; Part 6 credits CM with "a federal rulemaking process already underway"; §11 calls it "already in motion."

Calling a mandatory annual notice "a formal review" of CM, and a rule with a 2027 target "already underway," is the CHARACTERIZATION failure mode (V4) applied to the filing's own recommendation.

3. The June 2025 "spike that wasn't" — CONFIRMED (this pass's own first verdict was wrong, and is reversed)

Part 3 says the June 2025 reversal "turned out to be a statistical artifact in CDC's own forecasting model — corrected months later." The anchor table cites this to Post & Ciccarone, AJPH 2025, at PMC13066679.

Citation fix (stands). The article is "The 2025 Drug Overdose Spike That Wasn't: Neither Politics nor Data Errors Explain the Anomaly," Am J Public Health 2026 May;116(5):591–593 (PMID 41950462, doi 10.2105/AJPH.2025.308412) — 2026, not 2025.

Substance (reversed). This pass first recorded the claim as CORRECTED/withdrawn, reasoning that the title asserted the opposite and that the body was paywalled. Both premises were false. The article is open access at https://pmc.ncbi.nlm.nih.gov/articles/PMC13066679/ and was fetched in full on 2026-08-10 by an independent adversarial audit and re-fetched to confirm. It says, verbatim:

"Instead, the anomaly was a model artifact"

"A second revision released in August 2025 … clarifying that the January 2025 'spike' was an artifact"

"the anomaly resulted from applying growth-era algorithms to a period of decline"

The title rules out political manipulation and coding/reporting error as the explanation — leaving the forecasting-model artifact, which is exactly what the filing said. The paper also records that neither administration had an apparent motive to inflate mortality.

Verdict: CONFIRMED at primary-source tier. The withdrawal is reversed on the public pages and in the ledger below. The failure mode was V-title-only: a verdict rendered on a title plus an unchecked paywall assumption, without fetching the body. Logged as deviation #22.

4. Methadone's OTP monopoly is not "written directly into statute" — OVERSTATED

The abstract says methadone is dispensable only through 2,000 clinics "by a statute Congress alone can change"; Part 2 calls it "a restriction written directly into statute (21 U.S.C. §823(h))."

21 U.S.C. §823(h) (current text, in the atlas at method/sources/usc21-823h-methadone-registration.md) requires practitioners dispensing narcotics for maintenance or detoxification treatment to obtain a separate annual registration, and directs the Attorney General to register applicants who meet standards established by the Secretary. It is facility-neutral on its face. It does not name opioid treatment programs, does not confine dispensing to them, and does not cap the number of them. The OTP-exclusive structure is built by SAMHSA's 42 CFR Part 8 certification regime and DEA's registration standards issued under that section.

SAMHSA's own 2024 final rule supports the narrower reading. Asked by commenters to allow community-pharmacy dispensing, SAMHSA did not say it lacked statutory authority — it said "The final 42 CFR part 8 rule only applies to dispensing of methadone in OTPs. SAMHSA continues to work with Federal partners to explore ways through which access to MOUD might be expanded."

The separate-registration requirement is statutory. The monopoly is the agencies' implementation of it. This matters beyond wording: §11 puts methadone deregulation in Tier 3 ("requires new statute") on the strength of this reading.

5. A quotation attributed to Spencer 2023 is not in Spencer 2023 — CORRECTED

ws09-oregon-findings.md states: "Spencer's own abstract attributes this to decriminalization occurring 'in a context… not accompanied by substantial public health investments'." The published abstract (J Health Econ 2023;91:102798, PMID 37556870) reads: "This paper evaluates the causal effect of drug decriminalization on unintentional drug overdose deaths in a context with relatively poor access to drug treatment services."

The substance survives — Spencer does locate the effect in a poor-treatment-access context, and the finding of 182 additional 2021 deaths (~23%) is confirmed verbatim. The quotation does not. Per the protocol's known traps, a figure or phrase never seen inside the fetched source is exactly what this pass exists to catch.

6. The Switzerland gap the honesty box says is open has been closed — and it closed against the filing — STALE

The honesty box says: "we haven't yet root-traced Switzerland, the case most often cited as the one that funded treatment properly. That comparison is still open."

ws12-switzerland-root-trace.md, dated 2026-08-06, ran that trace. Its verdict: REFUTES the use of Switzerland as a positive counterfactual — the Swiss model is a four-pillar framework that retains enforcement, and its exceptional intervention is heroin-assisted treatment for a treatment-refractory population (~1,700 people, ~8% of Swiss heroin users), not decriminalization. Finding 3: NOT MEASURED — "Retire the phrase 'Switzerland funded treatment properly'."

The record found something that weakens the filing's framing, and the public page still tells readers the question is open. ws12's own implementation note says the site edit was deferred because PR #39 was touching site/drugs/index.html; that PR is closed and the edit never landed.

7. Three numbers on the public pages disagree with each other and with the record — CORRECTED


Secondary findings — recorded, no public change required


What could not be reached, and what was tried (V5)

Target Channels tried Result
CDC alcohol-attributable deaths (178,307, 2020–21) cdc.gov MMWR direct, stacks.cdc.gov, PubMed title search 403 / 403 / no match. Figure remains secondary-tier only
CDC NCHS Data Brief 549 (54,743 opioid deaths; 88% fentanyl) cdc.gov PDF 403
CDC tobacco mortality (450–480k) cdc.gov 403
AJPH editorial body text (the "spike that wasn't") PMC, Europe PMC fullTextXML, NCBI efetch db=pmc Not open access; title + metadata only
Chua et al. NEJM 2024 figures (1,938 prescribers; 5,245 patients; 810–830k flat) PubMed, PMC11103581 Citation confirmed (N Engl J Med 2024;390(16):1530-1532) — but it is a Correspondence, not a full article, and the specific figures could not be read
April 2026 Oregon changepoint study PubMed (multiple strings) No index record. Consistent with the record's own admission
BJS yearend-2007 state drug-offender count (for "down 46%") BJS p22st/p23st Not in either; Prisoners in 2008 not retrieved
D.C. Circuit consolidation order; Aug 17 2026 briefing date CourtListener v4 search + docket-entries Entries endpoint requires authentication
Campbell drug-court review (154 evaluations, 38% vs 50%) PubMed Criminology coverage gap
Focused-deterrence 2025 systematic review (50 evaluations, 9 RCTs) PubMed Criminology coverage gap

Retrieval notes worth carrying forward to other filings: data.cdc.gov's Socrata API serves the VSRR provisional overdose series cleanly over curl and is the working route around the cdc.gov block for mortality counts — it settled the peak and decline numbers that Phase 0 had to take from search summaries (deviation #3). library.samhsa.gov serves SAMHSA store PDFs after a 301 from store.samhsa.gov (follow redirects). reginfo.gov's eAgendaViewRule serves Unified Agenda entries directly and is the way to check whether a rulemaking is actually live.


The ledger

Verdict key: C = CONFIRMED · X = CORRECTED · O = OVERSTATED · U = UNVERIFIABLE · S = STALE. "as-of" columns per V6 are given only where the two differ.

A. Public whitepaper — site/drugs/index.html (122 claims)

# Claim Verdict Basis
A1 Whitepaper No. 8, GBMT-8, filed 2026-08-03 C Repo record
A2 Largest sustained decline in 40 years C Post et al., JAMA Netw Open 2025;8(6):e2514997
A3 Almost no policy anyone chose caused it O True of the two decomposition studies; contested by Dasgupta et al. 2026
A4 ~113,000 peak overdose deaths C VSRR peak 111,466 (12-mo ending Jun 2023)
A5 The legal drugs kill more U Depends on A17/A18, both blocked by the cdc.gov 403. Unreached, not contradicted
A6 Binding constraint is law, not evidence O Weakened by finding 1 — the CM leg of this claim does not hold
A7 CM capped by a federal anti-kickback rule one agency could lift tomorrow X 85 FR 77791–92; SAMHSA PEP24-06-001; 42 CFR 1001.952(hh)
A8 Methadone best-studied opioid treatment on Earth U No primary tier for a superlative
A9 Dispensable only through 2,000 federally licensed clinics C ws04 two-source; ~2,000–2,100 OTPs
A10 By a statute Congress alone can change O 21 U.S.C. §823(h) — see finding 4
A11 Fell ~38% from the 2023 peak S 37.3% at filing; 39.4% at check date
A12 Explained by supply shock + shrinking at-risk population O Per ws13, describe as multi-cause and unresolved
A13 Not by any program this filing could recommend C Dowell 2025; Vangelov 2026
A14 Portugal and Oregon are the two most-cited precedents C ws09, phase0
A15 Money arrived more than a year after leniency, both cases U Interval arithmetic checks (Feb 2021 → Jun 2022 = 16 months) but the disbursement date rests on ws09; Oregon SoS audit not retrieved
A16 Cheapest first move is a federal rule change X Depends on A7; see finding 1
A17 ~178K alcohol deaths/yr U cdc.gov 403; secondary corroboration only
A18 450–480K tobacco deaths/yr U cdc.gov 403; secondary corroboration only
A19 ~113K peak overdose deaths C VSRR 111,466
A20 43.1% deaths involving both opioid & stimulant C MMWR 2025;74(32):491–499
A21 Alcohol figure is a CDC 2020–21 average U cdc.gov 403
A22 Alcohol exceeds overdose peak by more than half U Arithmetic holds (178,307 / 111,466 = 1.60); the numerator is unreached
A23 Tobacco 4–7× peak overdose deaths U Arithmetic holds (450–480k / 111,466 = 4.0–4.3, and 4–7× against non-peak years); numerator unreached
A24 Tobacco toolkit already in the architecture list C ws10 A9 row
A25 Src line: CDC ARDI/MMWR 2024, tobacco estimates, Data Brief 549 U None retrievable
A26 Two legal drugs kill more than the fentanyl era at its worst U Same dependency as A5
A27 Methadone has decades of evidence C Uncontested; ws04
A28 Dispensable via pharmacies in UK, Australia, Canada C ws04 two-source
A29 US: only ~2,000 federally licensed clinics C ws04
A30 Restriction written directly into statute (§823(h)) O Finding 4
A31 Untouched by 2024's reform C 89 FR 7528: "only applies to dispensing of methadone in OTPs"
A32 Waiver repeal produced real prescriber growth C ws04; Chua NEJM 2024
A33 Flat growth in patients treated U NEJM item is a Correspondence; figures unreadable
A34 Confirmed by two independent studies U Second study (J. Addict. Med. 2025) not retrieved
A35 The barrier wasn't the binding constraint C Consistent with A32/A33 direction
A36 CM has the strongest evidence base for stimulant UD C OIG concurs: "the most effective currently available treatment for stimulant use disorders" (85 FR 77791)
A37 Polysubstance now the single largest overdose category O True as category comparison; drops ws03's "without being an outright majority"
A38 Capped for years at a third of trial-effective level X No cap as described; "a third" traces to nothing in the record
A39 By a federal anti-kickback policy one office can change on its own X Finding 1; 1320a-7d(a)(1)(B) requires AG consultation
A40 Src: 42 CFR Part 8 · Chua NEJM 2024 · PEP24-06-001 C All three exist; retrieved
A41 Stamp: rulemaking-executable, no statute required O Safe harbor is rulemaking; the statute it shelters from is not
A42 The CM fix is already in motion X RIN 0936-AA13 → Long-Term, NPRM 07/2027
A43 OIG has standing authority from a 1987 law C 42 U.S.C. 1320a-7d(a)(1)(A)(i) cites Pub. L. 100-93 §14(a)
A44 Opened a formal review in November 2024 X 89 FR 93545 is the mandatory annual notice; no CM mention
A45 Nothing about this needs Congress O A safe harbor doesn't; repealing the statutory exposure does
A46 Peaked at ~112–114K in 2023 C Reported peak 111,466; nowcast-adjusted series runs higher
A47 Fallen ~38% since S See A11
A48 Longest sustained decline in over 40 years C Post et al. 2025
A49 Exactly two studies decompose the cause quantitatively O Dasgupta et al. 2026 is a third analysis, contra
A50 Both point to shrinking at-risk population + supply shock C Dowell 2025; Vangelov 2026
A51 Plausibly a Chinese precursor crackdown C Stated as plausible; Vangelov 2026
A52 Not naloxone/treatment, which are secondary and unisolated C ws03; both papers
A53 June 2025 reversal was a CDC forecasting-model artifact C Post et al., AJPH 2026;116(5):591–593 (PMC13066679): "the anomaly was a model artifact" — see finding 3
A54 Corrected months later C Same source: "A second revision released in August 2025 … clarifying that the January 2025 'spike' was an artifact"
A55 Src: … Post & Ciccarone, AJPH 2025 X AJPH 2026;116(5):591–593
A56 Deaths falling for reasons no policy here controls C ws03; ws11
A57 Real cocaine & heroin prices fell ~80% since 1981 U ONDCP/RAND STRIDE series not retrieved
A58 13% of state prisoners are drug offenders C BJS Prisoners in 2023, p. 28
A59 Down 46% since 2007 U BJS yearend-2007 baseline not retrieved
A60 89% of US prisoners held in state systems O 88.2% (2023); 87.0% (2022)
A61 HOPE: 4-site replication found no advantage C Replication null real; see A66 for the settledness issue
A62 Focused deterrence: RCT-confirmed for gang/gun violence, not drug markets U 2025 systematic review not retrieved
A63 Drug courts 50%→38% recidivism U Campbell review not retrieved
A64 Benefit concentrated among serious, dependent offenders U NIJ MADCE not retrieved
A65 Stamp: steelman applied, not decorated C red-team.md Attack 4
A66 Enforcement's job is violence/reoffending, not price or supply O Overstates settledness of HOPE — Humphreys & Kilmer 2020
A67 Portugal HIV win genuine and strongly corroborated U Primary Portuguese/EMCDDA series not retrieved
A68 Causal contribution "cannot be firmly established" U Hughes & Stevens PDFs unreachable (deviation #6 stands)
A69 Most-cited pro-decrim US source is funded advocacy U Cato/MPP funding not verified at primary tier
A70 Three of four peer-reviewed studies find little-to-no effect O 2 verified, 1 unlocatable — see secondary findings
A71 They point to regional fentanyl timing C Zoorob 2024
A72 Funding didn't reach providers until 16 months after M110 took effect U Arithmetic checks; the June 2022 date was not re-derived from the Oregon audit
A73 Both precedents broke identically O Rests on A70's count and on Portugal claims held at U
A74 Src: Hughes & Stevens; Joshi; Zoorob; Spencer C Joshi, Zoorob, Spencer all verified as cited
A75 Spencer 2023 finds a real Oregon effect C J Health Econ 2023;91:102798 — 182 deaths, 23%
A76 Spencer attributes part of it to missing treatment buildout C Abstract: "a context with relatively poor access to drug treatment services"
A77 Eleven architectures on eight dimensions C ws10
A78 The eight dimensions as listed C ws10 D1–D8
A79 Four objective weightings C ws10
A80 Top-ranked, confirmed by an independent blind second scorer C ws10-rescore-log.md (with disclosed leak)
A81 CM has the strongest trial evidence in the filing O ws13: 2026 meta-analysis of 26 trials finds no overall retention effect; trials are pre-fentanyl-era
A82 A federal rulemaking process already underway X Finding 2
A83 Next to no cost C ws10 D3
A84 Two first-pass claims withdrawn C ws10-rescore-log.md
A85 Supply-side no longer ranks last C ws10 Σ16 vs A4/A9 Σ13
A86 Interdiction leg refuted by forty years of price data U Depends on A57
A87 Precursor leg carries the best observed link to the decline O Contested by Dasgupta et al. 2026
A88 Prevention row fell from second to bottom-tier C ws10-rescore-log.md (21→13)
A89 Decrim and enforcement steelman show largest weighting swings C ws10
A90 Src: 11 × 8 × 4, anchored scales, red-team applied C ws10; 11×8 = 88 cells, 87 scored (A11 has no D1)
A91 Stamp: clearest first move, not "the" recommendation C red-team Attack 2 amendment
A92 Rulemaking-executable list: CM safe harbor, IMD waivers, settlement conditions O CM item depends on A42
A93 Can move in the next year with no new appropriation or statute X CM NPRM target is 07/2027
A94 Congress-required: methadone monopoly, cannabis banking O Methadone leg weakened by finding 4
A95 Oregon and SF reversed within two to four years C ws08; HB 4002 (2024); SF ordinance (Jul 2026)
A96 By locally organized, donor-backed coalitions regardless of mortality data C ws08
A97 A politically invisible policy can bank results first C Argument, not fact claim; consistent with ws11
A98 Src: sequencing synthesis, 11 workstreams S Record now has 14 workstreams (ws12–ws14)
A99 Scorecard dimensions favor cheap, fast fixes C red-team Attack 2
A100 Switzerland not yet root-traced; comparison still open S ws12 completed it 2026-08-06 — finding 6
A101 Independent blind second scorer has re-scored every cell C ws10-rescore-log.md, 87 cells
A102 Reconciliation withdrew two claims and confirmed the headline C ws10-rescore-log.md
A103 A status line leaked the headline to the second scorer C Disclosed in ws10-rescore-log.md
A104 11-workstream protocol S See A98
A105 Pre-registered adjudication criteria (M3) O Deviation #13: scoring method "was set by the synthesizing session's judgment, not pre-registered"
A106 Anchor table preserving corrected priors C research-inquiry.md §3
A107 Deviations log (14 entries) X 20 entries
A108 Red team applied C red-team.md
A109 Steelman built for enforcement per M3 C ws05 H5.3; A10 row
A110 All committed and public C Repo
A111 Meta description: falling for reasons almost no policy caused O Same basis as A3
A112 Rail links (PDF, sources, corrections) C Files present
A113 Part 1 header: alcohol and tobacco kill more than the whole overdose crisis U Same dependency as A5
A114 Part 2 header: best-evidenced treatments are the most legally cornered O CM leg fails per finding 1
A115 Part 2 stamp: "Binding constraint" O Same
A116 Part 3 header: "The decline nobody caused" O Same basis as A3/A49
A117 Part 4 header: what enforcement buys and doesn't C ws05
A118 Part 5 header: "The precedents, root-traced" C ws09, phase0
A119 Part 5 stamp: "Contested, not settled" C ws09
A120 Part 6 header: eleven designs, four scorings, one first move C ws10
A121 Part 7 header: the cheap fix moves first C ws11
A122 Part 3 stamp: "Strong comparator, not a strawman" C ws10 A11 row; ws11

B. Public sources page — site/drugs/sources/index.html (63 claims)

# Claim Verdict Basis
B1 Series GBMT-8, filed 2026-08-03 C Repo
B2 Scope statement C research-inquiry.md
B3 Imports M1–M9 in full C Pre-M10 filing; correctly not retrofitted
B4 Phase 0 verdict GO C phase0-findings.md
B5 11 workstreams executed 2026-08-03 S 14 now exist
B6 Workstream list C Files present
B7 Red team applied four amendments before publication C red-team.md
B8 Peak ~112–114k in 2023 C VSRR 111,466 (Jun 2023)
B9 Cumulative decline ~37–38% C 37.3% at filing
B10 "Largest on record" confirmed via Post et al. JAMA Netw Open 2025 C 2025;8(6):e2514997 verified
B11 June 2025 reversal was a CDC nowcasting-model artifact, later corrected C Post et al., AJPH 2026 body fetched; finding 3
B12 "1 in 10" is a decade-stale 2015 root U NSDUH primary tables not retrieved
B13 2024 rate is 12.4–19.3% U Same
B14 Perceived need 3.3–6.6% replicates U Same
B15 Federal drug budget $44.2–44.5B FY23–25 U ONDCP PDFs not retrieved
B16 Split ~55/45 demand-majority U Same
B17 Historical 65–71% supply-side U Same
B18 No unified state/local accounting; one 1990–91 survey U Negative claim; Census survey not retrieved (V5: cannot be CONFIRMED)
B19 Anchor 4 prior (13–15% state, ~45% federal, ~7× ratio) C BJS: 13%, 45–46%, 7.24×
B20 Portugal HIV ~1,287→16 U Primary series not retrieved
B21 Death data incompatible across sources U Same
B22 Greenwald/Cato 2009 confirmed funded advocacy U Funding not verified at primary tier
B23 3 of 4 studies find no detectable effect O 2 verified
B24 Spencer dissent; funds didn't disburse until 16 months after C Spencer verified; timing verified
B25 France 79–80% single-root (Auriacombe 2004) U Paywalled (deviation #7 stands)
B26 French GP buprenorphine fell 43.8% 2009–2015 U Not retrieved
B27 Methadone now France's leading substitution-drug-death cause U DRAMES not retrieved
B28 Anchor 8 queued, not verified C Honest disclosure; accurate
B29 CM capped at ~$75–599/yr by anti-kickback rules X Finding 1
B30 The cap is OIG subregulatory policy, not statute X OIG: no OIG-imposed CM limit; $75 was a SAMHSA grant condition
B31 Alcohol ~178,307/yr; tobacco 450–480k U cdc.gov 403
B32 Methadone only via ~2,000 OTPs; peers use pharmacies C ws04
B33 Restriction written into §823(h), untouched by the 2024 reform O Finding 4; second half CONFIRMED
B34 Settlements $50–58.6B depending on tracker U Trackers are secondary by nature
B35 Only 12 states commit to detailed public reporting U Same
B36 Real prices fell ~80% since 1981 U See A57
B37 Alcohol+tobacco exceed peak overdose by 1.5–4× U Arithmetic holds; inputs unreached
B38 6.9% of rural counties had a Medicare-enrolled OTP in 2022 U HHS OIG report not retrieved
B39 ~3× spending disagreement ($13.1B vs $34–42B) C Reported as an open dispute — appropriate
B40 Polysubstance 43.1% is the single largest category C MMWR 2025;74(32)
B41 Two decomposition studies point supply/exposure-side O See A49
B42 Waiver repeal: prescriber growth, flat patients, two studies U See A33/A34
B43 CM anti-kickback cap raised 10× Jan 2025 (SAMHSA, grants only) X It was SAMHSA's own grant cap, not an anti-kickback cap
B44 Every binding constraint examined is federal C ws06; verified for CSA scheduling, §823(h), 1001.952
B45 Narrow Schedule III order April 2026, active D.C. Circuit litigation, not legalization C 91 FR 22714; 3 petitions on CourtListener
B46 CM cap is OIG subregulatory policy with a process already underway X Findings 1 and 2
B47 12 states reporting; $61M+ to law enforcement in 2024 U KFF/Hopkins tracker is the primary tier available; not retrieved
B48 CA/WA 50–60% illicit U State agency data not retrieved
B49 Buprenorphine 14–50× pain-vs-OUD disparity U Not retrieved
B50 Oregon and SF reversed within 2–4 years C ws08
B51 3 of 4 studies; dissent locates harm in missing funding O See B23
B52 CM ranks first or tied-first under 3 of 4 weightings C ws10
B53 Supply-side ranks last under every weighting X Withdrawn by the re-score (ws10-rescore-log.md); page never updated
B54 Decrim and enforcement steelman show largest rank instability C ws10
B55 Public-order durability runs on a 2–4 year clock C ws08/ws11
B56 Deviations log: 16 entries X 20
B57 Deviation summary rows 1–13 C Match deviations-log.md entries 1–13
B58 Five red-team attacks C red-team.md
B59 Independent second scorer remains a standing requirement S Delivered 2026-08-04
B60 Four-site replication found no advantage C See A61
B61 Switzerland not yet root-traced S ws12, 2026-08-06
B62 Receipts: deviations log (16 entries) X 20
B63 Status: Draft pending second scorer X Re-score completed 2026-08-04

C. Research record only (208 claims)

Given in explicit ID sets so the partition is checkable: the five sets below are disjoint and their union is C1–C208.

CORRECTED — 5. (Was 6; C14 moved to CONFIRMED by the 2026-08-10 audit — deviation #22.)

OVERSTATED — 7. C51, C53 (Dowell's and Vangelov's decomposition results stated more firmly than ws13 itself allows once Dasgupta et al. 2026 is in view); C93, C94 (HOPE — the Humphreys & Kilmer 2020 reconsideration is nowhere in the record); C116 (MOTAA's existence shows a statutory route was chosen, not that it was required — see headline finding 4); C157 (the Joshi null was published late September 2023, not "by mid-2023"; the operative point — before the 2024 session — holds); C207 (ws10's A3 basis text calls the Nov 2024 annual solicitation "already mid-process").

STALE — 3. C13 (CY2025 = 69,973 in the record; the current VSRR extract gives 69,939 — a provisional revision, not an error); C205 (the standing second-scorer requirement was discharged 2026-08-04); C208 (ws12's implementation note defers the site edit because "open PR #39 touches site/drugs/index.html" — #39 is closed and the edit never landed).

CONFIRMED — 55. Verified at primary tier this pass: C12 (CY2024 vs VSRR 80,860); C44–C46 (Tanz, MMWR 2025;74(32):491–499); C60, C61 (89 FR 7528); C79 (PEP24-06-001); C82, C85, C86 (BJS Prisoners in 2023); C99–C101, C103, C105 (91 FR 22714 / 91 FR 22777); C112 (Dec. 7, 2016 policy statement); C114 (89 FR 93545); C115 and C172 (Pub. L. 100-93 §14(a) via 42 U.S.C. 1320a-7d(a)(1)(A)(i)); C147, C148, C150 (Joshi, Zoorob, Spencer); C181 ($750 and the $605 2025 safe-harbor figure); C187 (Dasgupta et al., PMID 42078391 — noting it is a preprint, which ws13 does not say); C190–C194 (all four of ws14's legal statements, each independently re-derived); C195 (20 deviations entries); C14 (the nowcasting artifact — see finding 3; this pass's own first verdict on it was wrong and is reversed). Verified as accurate descriptions of the committed record: C160–C171 (scorecard and sequencing — the reconciled Σ values were re-added and check: A3 20; A5, A10, A11 18; A7 17; A2, A8 16; A1 15; A6 14; A4, A9 13; A11's per-dimension mean is 18/7 = 2.571 against A3's 20/8 = 2.500, so the "edges A3" claim is arithmetically right); C179 (ws12's REFUTES verdict); C182; C196–C204; C206.

UNVERIFIABLE — 138. C1–C11, C15–C43, C47–C50, C52, C54–C59, C62–C76, C78, C80, C81, C83, C84, C87–C92, C95–C98, C104, C106–C111, C113, C117–C146, C152–C156, C158, C159, C173–C178, C180, C183–C186, C188. These break into the three groups named under the verdict distribution: CDC-attributed figures behind the 403; the criminology and clinical literature not reachable from here; and the commercial/watchdog/state-agency material in §2, §7 and §8 that has no deeper tier. Four deserve individual mention because the filing itself flagged them and this pass could not close them either: C7–C10 (Portugal's primary sources — deviation #6 stands), C15–C17 (France's Auriacombe root — deviation #7 stands), C67–C68 (the Chua figures: the citation is confirmed as N Engl J Med 2024;390(16):1530-1532, but it is a Correspondence, and the 1,938 / 5,245 / 810–830k figures could not be read from it), and C152–C154 (Oregon's disbursement timeline, which carries headline claim A15).

Corrections applied on this branch

Per the corrections protocol, in dependency order — findings file → anchor table → scorecard basis → both public pages.

  1. ws04-findings.md — CM anti-kickback section rewritten to the primary-source position.
  2. ws06-findings.md — CM ceiling and "process underway" corrected; methadone statutory characterization narrowed.
  3. ws09-oregon-findings.md — Spencer quotation corrected to the published abstract; the April 2026 changepoint study downgraded and its arithmetic fixed.
  4. ws03-findings.md — Post & Ciccarone citation corrected to AJPH 2026. (The nowcasting attribution was withdrawn here and then restored the same day — see finding 3 and deviation #22; the AJPH 2026 body is open access and confirms the artifact explanation.)
  5. research-inquiry.md anchor table — rows 1, 9 and 11 corrected.
  6. ws10-findings.md — A3's basis text corrected (it cited the "Nov 2024 solicitation" as a live CM process).
  7. site/drugs/index.html — abstract, Part 2, Part 3, Part 6, Part 7, honesty box, receipts.
  8. site/drugs/sources/index.html — anchors 1/9/11, §4, §6, §10, deviations count, red-team Attack 5, document status.
  9. deviations-log.md — entry 21 (this pass).

Corrections are made in the open, per the protocol and the GBMT-2/GBMT-10 precedent. The whitepaper's scorecard rank order is not changed by this pass — that is a Phase 2 / re-score question. What changes is the basis text under A3 and the legal claims the public pages make about it. A3's D2 ("legal executability = 3, achievable via existing agency rulemaking alone") and D8 ("speed = 3, effects within ~1 year") are now visibly contradicted by RIN 0936-AA13's July 2027 NPRM target, and this log records that as owed to the next re-score rather than silently re-scoring a cell here.

Sources fetched and cached

Added to the atlas under method/sources/ this pass: oig-contingency-management-and-safe-harbors.md (85 FR 77684 at 77791–92; 42 CFR 1001.952(hh); OIG inflation table; 89 FR 93545; RIN 0936-AA13; SAMHSA PEP24-06-001) and dea-2026-marijuana-schedule-iii-order.md (91 FR 22714; 91 FR 22777; 91 FR 22778; the three D.C. Circuit dockets). Both carry the provenance header the atlas requires.

← All Drugs research documents Sources digest Read the whitepaper