Phase 0 Findings (GBMT-8)

Date: 2026-08-03 Scope: Verify the five ★ Anchor Table rows (overdose peak/decline, treatment-gap definition, federal drug budget split, Portugal 2001, France 1995 buprenorphine). Go/no-go input for full execution.

Verdict: GO, with one prior broken, two corrected, and one downgraded

No kill condition fired outright, but two of the three kill-condition triggers named in the protocol's execution notes came close enough to matter:

1. Kill condition (a) — the treatment-gap number is definitionally underdetermined, just as feared

The protocol's kill condition (a) asked: if the treatment-gap number turns out to be definitionally underdetermined the way GBMT-2's housing-shortage number was, that's the headline. It is. Under 2024 NSDUH data alone, the "treatment received" rate is:

• 12.4% of the SUD population (SUD-only denominator)
• 19.3% of the "needing treatment" population (SUD-or-already-treated denominator)
• ~17.0% for the narrowest, most evidence-based cut (MOUD receipt among opioid use disorder specifically)

— a factor-of-1.6× swing from denominator choice alone, before touching the deeper problem: the famous "1 in 10" figure is not even a current number. It traces to a single 2015 root (specialty-facility-only treatment, DSM-IV criteria, single-mode survey collection) that SAMHSA itself says is not comparable to 2021+ data after two independent methodology breaks (the DSM-5 transition and the 2020 collection-mode redesign). Every architecture in §10 premised on "closing the treatment gap" needs to restate its target in one of these identified quantities, not the borrowed "1 in 10."

The one part of the anchor that is not an artifact of definition: self-perceived need for treatment is genuinely and robustly single-digit (3.3–6.6%, stable across 2015/2023/2024 and a peer-reviewed reanalysis). Most people who screen positive for SUD and aren't in treatment don't think they need it. That finding survives every definitional cut tried.

2. Kill condition (c) — Portugal's causal story does not fully survive, but doesn't collapse either

Kill condition (c) asked whether Portugal's decriminalization precedent would survive root-tracing. Verdict: it survives as a real, mostly-positive natural experiment, but not as the clean causal story either side of the US debate uses it as. Specifically:

• The HIV-transmission win is genuinely strong and well-corroborated (~1,287 → 16 new IDU-attributed diagnoses, 2001–2019).
• The most-cited pro-decriminalization source in US discourse — Greenwald/Cato 2009 — is a Marijuana Policy Project–funded advocacy report, not independent scholarship. This is worth stating plainly in the whitepaper: a huge share of US "Portugal proves decriminalization works" argumentation runs through one funded document.
• The scholars who actually study this (Hughes & Stevens) frame their own work as a corrective to both the pro- and anti-decriminalization overclaims, and state that decriminalization's causal contribution — separate from the simultaneous treatment-system expansion Portugal also built — "cannot be firmly established."
• Death-rate data is a genuine mess: three sources give three different, non-comparable 2001 baselines. This isn't spin, it's real: Portugal used two incompatible national measurement systems (toxicology-based vs. cause-of-death-certificate-based) for years.

Net effect on §10: Portugal remains the strongest available precedent for decriminalization, but the whitepaper cannot present it as a controlled experiment that isolates decriminalization's effect — it has to be presented as a package (decrim + dissuasion commissions + a large simultaneous treatment buildout), with the HIV finding carrying more evidentiary weight than the death-rate finding.

3. The overdose decline is bigger, later-peaking, and more fragile-looking than the seed anchor said

Not a kill condition, but the single biggest correction: the true peak was ~112–114k in 2023 (not "2022–23" — 2022 alone wasn't the peak year), and the cumulative decline through 2025 is ~37–38%, not the ~25% the seed anchor guessed. This is good news for whoever's theory of the case the decline validates — except nobody's theory is settled yet, because a June 2025 apparent reversal ("deaths rising again") was later shown to be a CDC nowcasting-model artifact, corrected in August 2025. That single episode is worth a prominent honesty-box entry: real-time overdose data has already produced one nationally-reported false alarm in this project's own research window. §3 and §11 (sequencing) both need to treat any future apparent reversal with the same skepticism until it's confirmed against final, not provisional, data.

4. France's buprenorphine "natural experiment" is the shakiest precedent in the seed set, and it's structurally, not just evidentially, shaky

The famous 79% figure is single-root, imprecisely sourced, and — more importantly — the policy it's supposed to validate partially reversed: French GP buprenorphine-initiation fell 43.8% from 2009–2015 ("the end of the French Model"), and French national surveillance now shows methadone, not buprenorphine, is the leading cause of substitution-drug death in France. A 2023 peer-reviewed paper explicitly debunks the naive US transferability calculation (the "37,000 lives saved" extrapolation) that some US policy advocates have run with this figure. §4's methadone-deregulation architecture can still lean on France as directional evidence that deregulating a MOUD changes prescribing at scale, but should not repeat the 79% figure as a settled causal estimate, and should note the French system's own subsequent drift back toward methadone as a live caution about assuming any one-time deregulation is durable.

5. Federal drug spending is majority demand-side already — a fact useful against both sides of the standard debate

Contrary to the "war on drugs = mostly enforcement spending" framing common in harm-reduction advocacy, the federal drug control budget has been demand-majority (treatment+prevention, ~55–56%) since sometime in the FY2012–FY2021 window, and total federal spending ($44.2–44.5B) is not dominated by enforcement the way discourse often assumes. The historical "~2:1 supply-side" claim is confirmed for the 1990s–2000s — so the shift is real, not just a redefinition — but it did happen partly through a 2012 accounting change that also inflated supply-side numbers (Medicaid/Medicare/BOP corrections got added to both sides of the ledger, not just demand). This nuance matters for §7 and §11: "fund treatment more" is not obviously the marginal-dollar answer anymore if federal treatment dollars already outweigh enforcement dollars at the federal level — the binding constraint (per §4's seed hypotheses) may be regulatory/legal capacity to deliver evidence-based treatment (methadone's OTP monopoly, contingency management's anti-kickback ceiling), not appropriations. State/local enforcement spending remains genuinely unmeasured — the only attempt at a unified count is a single Census Bureau survey from 1990–91, never repeated in 35 years — so any comparison of "enforcement dollars vs. treatment dollars" nationally is still missing its state/local enforcement leg entirely.

Anchor verifications (full detail in the table, §3 of research-inquiry.md)

Notes for full execution

• §4 (treatment economics) and §9 (precedents) carry the most information per hour on current findings — both the France and Portugal precedents turned out to need far more nuance than the seed anchors assumed, and that pattern likely extends to the other §9 cases (Oregon M110, BC, Switzerland) not yet scanned.
• §2's baseline pass should settle on one treatment-gap denominator convention (recommend "SUD population, any-treatment-received" = 12.4% for 2024, since the "needing treatment" denominator is circular) and use it consistently for the rest of the filing, exactly as §2's seed hypothesis (H2.1) predicted would be necessary.
• The CDC nowcasting-artifact episode (row 1) should be flagged prominently wherever the whitepaper cites provisional (not yet finalized) mortality data.
• All five verifications used at least one primary or official source (CDC/NCHS, SAMHSA/NSDUH, ONDCP, EMCDDA-derived Portuguese data) per M1's two-source rule, with peer-reviewed independent reanalyses as the second leg where available; three of five (rows 1, 5, 7) hit real access limitations (cdc.gov 403s, paywalled primary papers) that are logged in deviations-log.md rather than papered over.