Date: 2026-08-04 Method: Per M6 — a second scorer (separate agent, procedurally blinded: the D1–D8 anchored dimensions + architecture list + the ws02–ws09 evidence base; the scorecard, sequencing, red-team log, and site pages forbidden) scored all 87 cells independently. Reconciled cell-by-cell. Blinding procedural and disclosed — including by the scorer itself, which reported one leak honestly: the protocol's own status block (an allowed file) states the A3 headline conclusion verbatim, so agreement on A3's top rank is partially contaminated and is discounted as confirmation; A3's per-cell bases were independently derived and stand. The other ten rows are uncontaminated.
Two headline catches
1. The supply-side row contradicted the filing's own headline finding — resolved by splitting the architecture's two legs. The original scored A8's mortality-mechanism cell at 1 ("no credible mechanism to overdose deaths") on the strength of the 40-year price/purity paradox. But the filing's own §3 attributes the historic 2023–25 death decline to a fentanyl supply shock (Vangelov/Reuter/Humphreys, Science 2026; converging with Dowell 2025) — the record's single best observed mortality correlation is supply-side. The paradox refutes the interdiction leg; the precursor-diplomacy leg carries the §3 evidence, with named unresolved caveats (fentanyl prices fell — the wrong direction for a shortage; West Coast timing misfit). Resolved: D1 1→2, D2 2→3, D3 1→2, D4 1→2, D6 2→3, D8 1→2. The old claim "supply-side modernization ranks last under every weighting" is withdrawn — the architecture is internally split, and the scorecard now says so.
2. The prevention row was scored on the size of its target, not the record's evidence about the instrument. A9's original 21 points included D4=3 (mortality) under a reading of "mortality" as all drug deaths — but the anchor says "overdose deaths specifically," and nothing in the record connects alcohol/tobacco levers to overdose mortality; D5=3 (CJ footprint) credited the tobacco model's absence of possession arrests as a reduction in the existing drug-arrest footprint, which it is not; D1=3 rested on a tobacco-control track record no workstream ever examined (§9's tobacco arm was never executed — the record verifies the body counts, not the policy efficacy). Resolved: D1 3→2, D4 3→1, D5 3→1, D6 3→2, D7 2→1, D8 2→1 — A9 falls from 2nd-ranked to bottom-tier (21→13). The rubric note carried with it: D4's overdose-only anchor structurally penalizes the architecture aimed at the numerically larger mortality pool — a scoring-frame limitation now stated on the scorecard rather than papered over in either direction.
Other material reconciliations (identical cells omitted; full matrices below)
| Cell | Orig | Blind | Resolved | Basis |
|---|---|---|---|---|
| A1 D7 | 2 | 3 | 3 | Removing daily-observed-dosing compulsion is coercion-reducing on its face |
| A2 D6 | 2 | 3 | 3 | Documented bipartisan statutory history (SUPPORT Act, CAA 2024) |
| A2 D4, D7 | 2, 2 | 3, 3 | 2, 2 | Kept: the X-waiver null is direct in-record evidence that delivery-variant access alone doesn't move volume; the in-custody offered-vs-compelled question is unresolved. Logged |
| A3 D6 | 2 | 3 | 3 | Zero public-order exposure + bipartisan action in both branches; OIG-reversibility is a different vulnerability than backlash and stays in the basis text |
| A3 D7 | 3 | 2 | 2 | CM lifts a provider payment restriction; it does not reduce state power over individuals |
| A4 D4 | 1 | 2 | 2 | Quality-conditioning has a plausible indirect path; anchor-1 overshot |
| A5 D2 | 3 | 2 | 2 | The SCS leg's legal posture is unresolved in the record |
| A5 D4 | 2 | 3 | 3 | Naloxone is definitionally the most direct mechanism to overdose death; §3's attribution caveat concerns the decline, not the drug's established efficacy — a distinction §3 itself draws |
| A5 D5, D8 | 1, 2 | 2, 3 | 2, 3 | FTS decriminalization (45 states) is a banked footprint reduction; naloxone acts within the year |
| A6 D1 | 2 | 1 | 1 | Contested evidence with incompatible baselines, an advocacy-funded root, three nulls and one adverse study — and the package variant never cleanly tested (Switzerland gap, deviations #15) |
| A6 D3 | 2 | 1 | 1 | Prompt, adequate treatment appropriation is the design's defining feature — and its documented failure mode |
| A6 D8 | 1 | 2 | 2 | The CJ effect is immediate on enactment (Oregon, Feb 2021) |
| A7 D2, D8 | 1, 1 | 2, 2 | 2, 2 | The narrow Schedule III slice already moved via rulemaking = mixed; §280E/banking relief bites fast once effective |
| A10 D1 | 2 | 3 | 3 | 50 controlled evaluations incl. 9 RCTs + a Campbell review is anchor-3 convergence for the architecture's actual core mechanism (violence/recidivism) — the original discounted the evidence for pointing somewhere unhelpful, which is a D4 concern, not a D1 one |
| A10 D4 | 1 | 2 | 2 | Drug courts' treatment routing is a plausible indirect mortality path |
| A10 D5 | 2 | 1 | 2 | Kept: participants' supervision-for-incarceration substitution is a real, modest reduction; net-widening caution stays in the basis. Logged |
| A11 D6 | 2 | 3 | 3 | The 2024–26 political drift runs toward the status quo — documented in ws08/ws09 |
Reconciled matrix (D1–D8; Σ)
| # | Architecture | D1 | D2 | D3 | D4 | D5 | D6 | D7 | D8 | Σ |
|---|---|---|---|---|---|---|---|---|---|---|
| A1 | Methadone deregulation | 2 | 1 | 3 | 2 | 1 | 2 | 3 | 1 | 15 |
| A2 | MOUD-everywhere | 2 | 2 | 2 | 2 | 1 | 3 | 2 | 2 | 16 |
| A3 | CM safe-harbor expansion | 3 | 3 | 3 | 2 | 1 | 3 | 2 | 3 | 20 |
| A4 | Treatment-quality regime | 1 | 2 | 2 | 2 | 1 | 2 | 2 | 1 | 13 |
| A5 | Harm-reduction scale-up | 2 | 2 | 2 | 3 | 2 | 1 | 3 | 3 | 18 |
| A6 | Decrim + dissuasion + funding | 1 | 1 | 1 | 2 | 3 | 1 | 3 | 2 | 14 |
| A7 | Cannabis federal redesign | 2 | 2 | 3 | 1 | 2 | 2 | 3 | 2 | 17 |
| A8 | Supply-side modernization (split verdict) | 2 | 3 | 2 | 2 | 1 | 3 | 1 | 2 | 16 |
| A9 | Prevention/tobacco playbook | 2 | 2 | 3 | 1 | 1 | 2 | 1 | 1 | 13 |
| A10 | Enforcement steelman | 3 | 3 | 2 | 2 | 2 | 3 | 1 | 2 | 18 |
| A11 | Status quo drift | — | 3 | 3 | 2 | 2 | 3 | 2 | 3 | 18/7 |
What survives, what changed
- A3's top rank survives — now with the contamination caveat stated rather than hidden, and with two cells moved in opposite directions (D6 up, D7 down). Its case never rested on the leaked sentence: the D1/D2/D3/D8 cells (trial evidence, live OIG rulemaking, near-zero cost, demonstrated one-year speed) are each independently grounded.
- The bottom is now A4 and A9 (13 each), not A8. The treatment-quality regime remains the genuinely-untested idea; prevention remains the architecture whose real strength (targeting the larger legal-drug body count) lives outside this scorecard's overdose-specific mortality anchor — both facts now stated as such.
- A9's fall and A8's split are rubric-integrity results, not evidence changes — the same lesson as the housing and media re-scores run the same day: most scorecard error is anchors applied loosely, not facts gotten wrong.
- A11 (status quo) at 18/7 dimensions edges A3 on per-dimension mean (2.57 vs 2.50) — the blind scorer's renormalization point, now carried on the scorecard: the comparator's strength is partly that the deaths are already falling for reasons no architecture here caused, which is the filing's own §3 headline applied to its own ranking.
Effects applied
ws10-findings.md revised to the reconciled matrix with corrected rank-stability text; whitepaper scorecard section + sidebar updated; report PDF regenerated; deviation appended.