Is America's drug policy actually working?

The bigger numbers: alcohol and tobacco kill more people than the whole overdose crisis

Every "drug policy" conversation that stays confined to illegal substances is, by simple body count, looking at the smaller problem. Alcohol-attributable deaths (CDC, 2020–21 average) exceed the overdose crisis's own peak by more than half; tobacco kills 4–7 times as many Americans annually — using the exact policy toolkit (taxation, age-gating, marketing limits) this filing's own architecture list already includes for the legal drugs everyone already tolerates.

This was the one claim we led with and couldn't prove. The Phase 1 fact-check marked it unverified — not contradicted, unreached — because cdc.gov blocks automated retrieval and every route to the underlying figures returned a 403. The Phase 2 pass found the block is on the route, not the source: CDC's own MMWR is deposited in PubMed Central and Surgeon General's Reports are published on NCBI Bookshelf. Both were read in full. Alcohol: 178,307 average annual deaths, 2020–2021, from CDC's Alcohol-Related Disease Impact application across 58 causes — 1.60 times the overdose crisis's own verified peak of 111,466. Tobacco: "at least 480,000 premature deaths annually." One caveat we owe you: that tobacco figure is the average for the 2005–2009 exposure period, published in 2014. It is still CDC's standing estimate, but "4–7 times as many Americans annually" rests on a twenty-year-old base, and smoking has fallen a long way since.

The legal wall: the best-evidenced treatments are the most legally cornered

Methadone — decades of evidence, dispensable in the UK, Australia, and Canada through ordinary pharmacies — can only be dispensed in the US through ~2,000 federally licensed clinics. What 21 U.S.C. §823(h) actually requires is a separate annual registration for practitioners dispensing narcotics for maintenance treatment, granted on standards the Secretary sets; the clinic-only structure on top of it is regulation (42 CFR Part 8 plus DEA registration standards), and 2024's reform left that structure alone. Buprenorphine's 2023 prescribing-waiver repeal produced real growth in prescribers — but flat growth in patients actually treated: removing a legal barrier didn't move the outcome, because the barrier wasn't the binding constraint. And contingency management — the strongest evidence base for stimulant use disorder, which drives a growing share of the polysubstance deaths now the largest single overdose category — has been held below its trial-effective incentive level by a stack of federal fraud-and-abuse rules.

This section originally said something stronger, and it was wrong. It described contingency management as capped "by a federal anti-kickback policy one office can change on its own," and said the fix was "already in motion." Checked against primary text, neither holds. HHS's Office of Inspector General states in its own 2020 rule that "there is no OIG-imposed $75 limitation on contingency management program incentives" — the $75 figure is nominal-value guidance under a different authority, the Beneficiary Inducements penalty, and OIG says it does not apply to the anti-kickback statute at all. The $75 that actually constrained programs was SAMHSA's own grant condition, and SAMHSA lifted it to $750 in January 2025. The real regulatory ceiling is the patient-engagement safe harbor, capped at $623 for 2026, in-kind only, and available only inside a qualifying value-based arrangement.

And the rulemaking is not underway. The November 2024 document we cited is OIG's statutorily required annual solicitation for safe-harbor proposals — it is published every year and does not mention contingency management. The actual CM safe-harbor proposal sits on the Unified Agenda under Long-Term Actions with a target of July 2027. OIG's authority to write such a rule is real and dates to a 1987 statute, but the law requires it to act in consultation with the Attorney General and through notice-and-comment — and a safe harbor shelters conduct from a statute rather than repealing it.

The deeper problem isn't that we described the wall wrong. It's that there was already a way around it, and we said there wasn't. CMS has approved contingency management as a Medicaid benefit under Section 1115 demonstration authority in five states — California, Washington, Montana, Hawaii and Delaware — with two more pending. Approval is not delivery, and we should not have written it as though it were: the state-by-state review this rests on says plainly that California "is the only state to have confirmed implementing" the benefit, and warns that the other four states' figures are "subject to change prior to implementing." Five approved, one confirmed running — California, since March 2023. The approved incentive ceilings run $596 to $1,092 per programme, and the programmes are 24 weeks in California and Washington and 12 in Montana — not a calendar year. Only Delaware's figure is annual: $599 a year, raised to $750 a year with CMS's approval. Washington's $1,092 over 24 weeks is 1.8 times the federal safe-harbor cap this section calls the binding ceiling. Application to approval took about a year; approval to first patient, nine months more. So the "cheapest, fastest first move" we recommended is a rule targeting 2027, and the route that has actually put the treatment in front of patients — in one state so far — is one we described as closed.

And the ceiling we named is not the only ceiling. That is where we had this wrong in the other direction. The Veterans Health Administration has run a national CM programme since 2011, funds it centrally, operates the largest integrated health system in the country, and faces no OIG safe-harbor problem. Between July 2018 and December 2020 it delivered contingency management to 1,698 of the 138,280 patients it diagnosed with stimulant use disorder — 1.2%. We read that as showing no incentive ceiling was binding anywhere. The study's own discussion says otherwise, in terms: "Primary among these [barriers] are incentive caps, which hold incentive distributions to <$600 per calendar year due to tax reporting requirements." The VA has a cap too. It is a different cap — driven by IRS reporting rules, not by fraud-and-abuse law — and it sits at roughly the same level as the safe harbor this filing wanted lifted. The honest reading is narrower than the one we published and still cuts against our recommendation: what suppresses contingency management is caps on incentive size, whatever authority they come from, and the OIG safe harbor is only one of them. Lifting it would not have moved the VA by a dollar. The delivery cost is real as well — a trained coordinator at every site, 36 point-of-care urine tests per patient, an incentive-management platform, a readiness review before launch — and our scorecard's "near-zero net new cost" priced the rule change, not the service. What we cannot claim, and originally did, is that the VA is a place where no ceiling exists.

What survives, and got stronger. The same VA data produced the first real-world evidence that contingency management saves lives: 1,481 patients who received it were 41% less likely to die within a year than 1,481 matched patients who didn't (adjusted hazard ratio 0.59). We were under-confident about that, not over. And the incentive-level complaint half survives — the first data-driven benchmark of what a trial-effective dose costs puts voucher protocols at about $1,536 per twelve weeks and prize protocols, the kind states actually run, at $660. Against the voucher figure the ceiling really does bind. Against the prize figure, the 2026 safe harbor is within 6% and Washington's approved programme clears it outright.

The decline nobody caused

Overdose deaths peaked at about 111,500 on a 12-month-ending basis in mid-2023 and have fallen 37% through the end of 2025 — and 39% through February 2026 — the longest sustained decline in over 40 years. Two studies decompose the cause quantitatively and both point away from any single policy lever: a shrinking at-risk population and a fentanyl supply shock (plausibly a Chinese precursor-chemical crackdown), not naloxone distribution or treatment expansion, which have real but secondary, unisolated contributions. A 2026 reexamination cuts against that reading, finding the fentanyl-purity series does not track regional overdose increases and that the same correlations appear with unrelated economic indicators. The honest statement is that the decline is multi-cause and unresolved. In June 2025, a widely reported "deaths rising again" reversal turned out to be a statistical artifact in CDC's own forecasting model — corrected months later, after the story had already spread.

We withdrew this claim earlier today and were wrong to. The 2026-08-10 verification pass withdrew the sentence above — that the June 2025 "deaths rising again" reversal was "a statistical artifact in CDC's own forecasting model" — on the reasoning that the cited article's title, "The 2025 Drug Overdose Spike That Wasn't: Neither Politics nor Data Errors Explain the Anomaly," asserted the opposite, and that its body was paywalled. Both halves of that reasoning were wrong. The article is freely available at PMC13066679, and its body says, verbatim: "Instead, the anomaly was a model artifact," that CDC's "second revision released in August 2025" clarified "that the January 2025 'spike' was an artifact," and that "the anomaly resulted from applying growth-era algorithms to a period of decline." The title rules out political manipulation and data error as causes — which leaves the forecasting-model artifact, exactly what this section said. The claim is restored and confirmed at primary-source tier. What stands from the earlier correction is the citation itself: Post et al., Am J Public Health 2026;116(5):591–593, not AJPH 2025.

Deaths are already falling substantially for reasons no policy in this filing controls. Every architecture we score has to beat that trend, not zero — and the false-alarm episode is a warning that the public conversation is primed to believe a reversal on thin evidence.

What enforcement actually buys — and what it doesn't

The precedents, root-traced

Portugal's 2001 decriminalization produced a genuine, strongly corroborated HIV-transmission win — but the scholars who actually study it say decriminalization's own causal contribution, separate from the treatment system Portugal built alongside it, "cannot be firmly established." The most-cited pro-decriminalization US source is a funded advocacy report, not scholarship. Oregon's Measure 110 tells a sharper story: two peer-reviewed studies find no detectable effect on overdose deaths, pointing instead to the same regional fentanyl timing that hit every neighboring state — but Oregon's own treatment-funding mechanism didn't start reaching providers until 16 months after decriminalization took effect. Both precedents broke the identical way: the leniency arrived on schedule, the treatment money didn't.

Corrected 2026-08-10: this said "three of four peer-reviewed studies." The third — an April 2026 changepoint study — could not be located in any index, and the December 2019 inflection point attributed to it conflicts with the changepoint analysis inside one of the two studies we can verify, which puts Oregon's fentanyl escalation in the first half of 2021. Two verified nulls, one verified dissent.

One peer-reviewed study (Spencer 2023) does find a real Oregon effect — and attributes part of it to that same missing treatment buildout, not to leniency itself.

The architecture that wins: eleven designs, four ways of scoring them, one clear first move

Eleven candidate architectures were scored on eight anchored dimensions — evidence strength, legal executability, cost, mortality impact, criminal-justice footprint, political durability, liberty impact, and speed — then ranked under four objective weightings (mortality-first, order-first, liberty-first, equal). The top-ranked architecture — confirmed by an independent blind second scorer — is contingency-management legalization: the strongest trial evidence in the filing — though a 2026 meta-analysis of 26 trials found no overall retention effect, and the trial base largely predates the fentanyl era — a rulemaking-only legal path, next to no cost. Two of the cells behind that ranking no longer hold. The 2026-08-10 verification pass found the rulemaking is not underway — the CM safe-harbor proposal targets a July 2027 notice — which contradicts both the "legal executability" and the "speed" cells this row scored at the top of the scale. We have not re-scored the board here: patching cells outside a fresh blinded pass is exactly the shortcut this project's method forbids. The owed re-score is recorded in the research file, and until it runs, treat the first-place finish as unsettled. Two first-pass claims did not survive the re-score and are withdrawn: supply-side spending no longer ranks last — its interdiction leg is refuted by forty years of price data, but its precursor-control leg carries the record's best observed link to the current death decline, so the row now holds a split verdict — and the prevention/tobacco-playbook row fell from second place to bottom-tier once its cells were re-anchored to what the record shows about the instrument rather than the size of its target. Decriminalization and the enforcement steelman still show the largest weighting swings — which remains the finding, not a flaw.

The Phase 2 steelman did the arithmetic, and it is not flattering. We did not re-score the board — patching cells outside a blinded pass is the shortcut this project forbids, and that re-score is still owed. But the same eleven rows can be totted up under three defensible readings of their own cells, and a reader can check every one:

Contingency management holds first place in exactly one of the three — the uncorrected one. Apply the cells our own fact-check contradicted and it falls to fourth. Add the Phase 2 evidence and it climbs back into a three-way tie, but only when the recommendation is restated as state Medicaid coverage rather than a federal rule, and only after its cost cell is repriced. The one result stable across all three readings is the comparator: status-quo drift — changing nothing — scores at or above every architecture on per-dimension mean in every version of this board. We reported that in a parenthesis. It is the most durable thing the scorecard says.

One more thing a reader should be able to do and can't. This section says the architectures were ranked under four objective weightings, and our sources page says CM ranks first or tied-first under three of them. The weights were never written down — they are named in the research file and nowhere quantified, and no committed document contains four ranked lists. Only the equal-weight totals above can be reproduced. That claim is not wrong so much as uncheckable, which for a filing that publishes its receipts is the worse failure. The owed re-score will publish its weights.

Sequencing: the cheap fix moves first, precisely because nobody's watching it

Rulemaking-executable items — the contingency-management safe harbor, Medicaid's IMD-exclusion waivers, tighter conditions on opioid-settlement spending — need no new appropriation and no new statute. Corrected 2026-08-10: this section said they "can move in the next year." For the CM safe harbor that is no longer true — the proposal targets July 2027, and nothing is pending. The tiering also needs one repair in the other direction: methadone's dispensing monopoly sits here under "items requiring Congress," but §823(h) requires only a separate registration on standards the Secretary sets, so a meaningful part of that monopoly is regulatory and does not belong in the congressional queue. A durable cannabis banking fix does. The sharpest lesson from the precedents: both Oregon's and San Francisco's harm-reduction-forward policies were reversed within two to four years by locally organized, donor-backed coalitions, regardless of what the underlying mortality data actually showed. A policy that's politically invisible — like a federal safe-harbor rule — can bank results before any backlash coalition has time to organize. A policy that's politically legible as "harm reduction" cannot count on that same runway.

The argument of this section is that invisibility is an advantage. The record says it's the opposite. We wrote that a politically invisible rule "can bank results before any backlash coalition has time to organize." But a policy nobody can see also has nobody demanding it, and on a crowded federal agenda attention is what buys calendar time. Look at what actually happened between 2022 and 2026. Four federal changes to drug treatment and drug law were legible and contested — the buprenorphine waiver repeal, over-the-counter naloxone, the 2024 methadone rule, the cannabis Schedule III order that drew three appellate challenges. All four happened. The one item in our own catalogue that nobody was watching — the contingency-management safe harbor — is the one that didn't, and its published target moved away from us: May 2026, then Long-Term Actions, then July 2027, across four consecutive years in which OIG invited exactly this kind of proposal and none was made.

Low backlash risk and low durability turn out to be the same fact seen twice. A safe harbor written by notice-and-comment can be narrowed by notice-and-comment, and an invisible rule has no constituency to defend it. Our own comparison was mis-drawn too: Oregon and San Francisco were ballot and ordinance measures reversed by voters. The right comparison for a federal safe harbor is other federal safe harbors — and there the record is a rulemaking that hasn't reached a proposal in four years of asking.

What replaces it. The first move is state Medicaid coverage of contingency management under Section 1115 — the same demonstration authority this section already credits for Medicaid's IMD waivers, and the one that has actually put the treatment in front of patients. It is not free and shouldn't be sold as free. A federal safe harbor is still worth having; it is what would let full voucher-magnitude protocols reach payers nationally. It is just not the first move, and it is not fast.