The untaken lever — my position, steelmanned, then dismantled, then what survives

Rules of engagement

Same three rules as the original Steelman page, mirrored for the other side.

First, read it in my voice, not the voice you would use to mock me. The speaker on this page is someone who has watched the addicted in their family, in their friends, in the obese-and-hypertensive patients of a thousand GP surgeries, fail to do what the willpower frame told them to do, over and over, for decades, and finally watched a drug do the thing in a fortnight that thirty years of dietary self-discipline did not. The position is not glib. It is what arrives after the moralism collapses.

Second, the steelman is not a strawman in a hat. The argument below is the position I actually hold, made as forcefully as I can make it. If, after reading it, the pharmacology-first position sounds more persuasive than it did before, the page is doing its job. The teardown comes after.

Third, the answer comes at the end, not in the middle. No fact-checking footnotes interrupting the steelman. Let me finish.

My voice — the full statement

It is not a handful. Roughly 10 per cent of any developed-country adult population has the disease — clinically diagnosable, mechanism-real, not metaphor. That is hundreds of millions of people globally. The burden is enormous, devastating, unsolved, and getting worse on most of the curves we can measure.

For the sick, education does not work. Willpower does not work. Forty years of evidence is in. We ran the experiment. DARE, Just Say No, change-for-life, the eatwell plate, the abstinence-only sex curriculum — all of them were the willpower frame in operational form, all of them were evaluated, all of them failed at the dependent variable. Adolescent drug use did not fall under Just Say No. Obesity did not fall under change-for-life. Tobacco fell — but it fell because the price went up, the smoking rooms closed, NRT and varenicline became available on prescription, and the culture re-coded smoking as low-status. The poster campaigns were the smallest part of it. The pharmacology and the price levers did the work.

The pharmacology, when you let it work, is extraordinary. Varenicline triples the sustained quit rate for tobacco. Methadone halves opioid mortality. Buprenorphine does the same with easier prescribing. GLP-1 agonists shift weight outcomes in the obese by twenty per cent of body mass and cut all-cause mortality by a fifth in the cardiovascular-disease group. Naltrexone — the most under-deployed molecule on the table — helps both alcohol and opioid use disorder, oral for one, depot for the other, the same active ingredient solving two of the five biggest addictive classes at once. Acamprosate as the alcohol partner. Five drugs. Combined adoption ceiling, roughly six per cent of the global burden of disease averted. Six per cent. WOW.

And we are at ten per cent adoption.

Of the things that work outside the drug list, the surprises are these. AA — the cheap, manualised, peer-led twelve-step model — beats CBT on continuous abstinence in the Cochrane review. That should have surprised everyone in the field. It did. Contingency management, paying people to be sober, works at large effect sizes in stimulants where no drug exists. Therapy on its own — the hour-a-week professional model — hardly works for the genuinely sick. Twenty-eight-day abstinence-only US-model rehab is a failure as an evidence-based intervention; it persists because insurance reimburses it, not because it produces outcomes. The rehab industry is the willpower frame in clinical drag.

And — this is the line that goes in everyone's notebook — extrapolating from your own ability or lack of addiction is neither a helpful policy approach nor does it make you a compassionate discussion partner with someone who has acknowledged their struggle. When you have never been hooked, the willpower frame is intuitive. You did the thing. You think anyone could do the thing. The 10 per cent are not anyone. They are a different population on the relevant biological axis. Telling them they should be more like you is, at the level of policy, a category error. At the level of personal conversation it is unkind.

I accept the perimeter has stretched. Sex addiction is not a thing. The DSM declined, the ICD downgraded, the evidence base does not support the construct. Gambling addiction, on the other hand, is — ICD-11 recognises it as a behavioural addiction, the neural circuit is the right one, and the operational pattern fits. Drawing the line carefully matters; the willpower critics have a point that the diagnostic frontier has expanded faster than the evidence in some places. Honest answer: discipline the perimeter, then treat what's inside it like a disease.

So here is the position. The sick are sick. They need drugs. The drugs exist and they work. Current adoption is ten per cent. Lifting it to a hundred — by giving it through public systems, by selling it through markets, by both, by whatever payment mechanism actually puts the molecule in the body — is the largest public-health win on the table in 2026. The blocker is not science. It is moralising. Stop moralising. Pick up the lever.

Let it sit for a moment before you answer it. If your first instinct is to argue with the ten-per-cent figure, look up the IHME global burden of disease numbers before you do.

The five drugs and the arithmetic

The numbers behind the steelman, all extrapolated from the per-drug effect sizes already cited on the Treatments and Costs pages.

What I get right

Five things — not small.

One: the 10-per-cent prevalence is real. The IHME global burden of disease numbers, the WHO addiction figures, the NESARC-III alcohol epidemiology, and the UNODC drug-use estimates all converge on a number close to 10 per cent of adults in developed economies meeting criteria for one or more substance use disorders at any given time. Lifetime numbers are higher (29 per cent for AUD alone in the US per NESARC-III). The 10-per-cent figure is the floor, not the ceiling. The willpower critics who treat addiction as marginal — a few people on the high street, mostly other people's children — are mis-sizing the denominator by an order of magnitude.

Two: education-only interventions fail at population scale. DARE, Just Say No, abstinence-only sex-ed, the change-for-life and eatwell-plate decades, the BMI letters to schoolchildren — the evidence base is consistent and damning across these. The Policy page documents this in detail. "Education works" is not a tenable empirical claim for the addicted population; it is a claim about how the speaker wishes the world worked. The data closed this question in the early 2000s.

Three: pharmacological effect sizes are large compared to anything else in the field. The mortality differentials on methadone (3.2×), buprenorphine (2.2×), semaglutide (19% all-cause), and the quit-rate trebling on varenicline are not marginal nudge-style effects. They are the largest interventions modern medicine has produced for any condition with comparable global disease burden. The willpower critics typically have not done the comparison; if they had, they would notice that addiction pharmacotherapy outperforms most cardiovascular preventatives on absolute risk reduction.

Four: the Cochrane AA finding really did surprise the field. Manualised twelve-step facilitation outperforms CBT on continuous abstinence at long follow-up at a fraction of the cost (Kelly et al., 2020). The peer-support mechanism — cheap, lay-led, distributed — does work that the expensive professional model does not. The implication is uncomfortable for the addiction-treatment industry but it is in the published evidence. Anthony is right to flag it as a load-bearing fact.

Five: the extrapolation point is correct and underweighted. Non-addicted people inferring from their own self-control to what the addicted ought to be able to do is a category error grounded in a population-mean confusion. It is also, at the personal level, the move that makes addicts not want to talk to non-addicts about their problem — which destroys the social regulation that would otherwise help. The Steelman page's lab-tech nurse is exactly this error embodied. Anthony's line — that self-extrapolation is neither good policy nor good company — deserves the underline he gives it.

What I get wrong

Four things, load-bearing.

One: "willpower does not work for the sick" elides what willpower actually does for them. The compliant tirzepatide patient, the patient who returns to the methadone clinic every day, the smoker who finishes the varenicline course, the alcoholic who attends AA twice a week — each of these is exercising substantial self-discipline. The drug shifts the biology so that the discipline can land where, without the drug, it could not. Saying willpower does not work for the sick is true in the narrow sense that pure self-control without pharmacological support rarely produces sustained recovery in severe cases. It is false in the broader sense that pharmacology and self-discipline are substitutes. They are complements. The honest version of the position is: willpower alone does not work for the sick; willpower with pharmacological support does. Eliding the complement weakens the case.

Two: "therapy hardly works" is too broad. Therapy alone, for severe AUD, hardly outperforms peer support. Therapy as a delivery mechanism for the drug + behavioural change combination is a different proposition. Project MATCH found CBT, MET, and TSF all produced significant sustained improvement; the comparative differences were small but the absolute effects were real. The Cochrane finding is that AA is non-inferior or slightly better than CBT, not that CBT does nothing. Contingency management is itself a structured behavioural intervention. "Therapy hardly works" rolls these distinctions together; the truer claim is that therapy without the right behavioural model and without pharmacological support underperforms what the field considers the standard of care.

Three: "rehab is a failure" is true of the dominant US 28-day abstinence-only model and false of the field as a whole. The Treatments page makes the case at length. Residential programmes that integrate medication, harm reduction, and post-discharge step-down do produce measurable outcomes. The American Minnesota Model in its purest form does not — and it is the model with the loudest marketing budget. Calling all rehab a failure cedes ground to the marketing department of the failed model, which then defends the broader category from inside. The sharper version of the claim: most US 28-day abstinence-only rehab is evidence-light; the format persists for insurance reimbursement reasons, not for outcome reasons.

Four: the give-it-or-sell-it framing under-prices the moralising problem at the patient level. The argument that the market or the public system will sort it out the moment moralising stops assumes the moralising is exogenous to the patient. It is not. The internalised moralism — shame about being the kind of person who needs the drug — is itself part of the disease's clinical picture (see the Pathway page on the shame spiral). Even when the drug is on the shelf and the system pays for it, the patient may not reach for it because reaching for it confirms the diagnosis. The Mounjaro case is the live counter-example, and it is striking, but Mounjaro patients overwhelmingly self-identify as having tried diet-and-exercise repeatedly and run out of options. They have already done the moral self-flagellation; reaching for the drug is the post-shame move, not a pre-shame one. The supply-side argument is necessary but not sufficient; the cultural change has to happen too.

"Extrapolating from your own ability"

This argument deserves its own section because it is the sharpest line in the steelman and it does most of the political work.

The claim is twofold. At the policy level, treating the addicted as a different population on the relevant biological axis is empirically correct (10 per cent prevalence, distinct mechanism, distinct response to interventions); pretending otherwise produces interventions calibrated to the wrong denominator. At the personal level, the non-addicted person using their own self-control as proof of what the addicted ought to be able to do is a conversational move that ends conversations, alienates the people who need help, and produces the well-documented avoidance-of-medical-care that makes outcomes worse.

The teardown of this is partial, not total. The policy-level point holds; the personal-level point is exactly right; the conversational implication — be careful with self-extrapolation when talking to an addicted friend or family member — is good advice and good politics. But — and this is the steelman's blind spot — the population that addiction policy has to manage is not only the 10 per cent who have the disease. It is also the 90 per cent who do not, and who are nonetheless making lifestyle decisions about alcohol, food, tobacco, screens, and gambling that affect their long-run health and the system's costs. The willpower frame is wrong for the 10 per cent. It is partially right for the 90. Public-health policy needs both tracks, calibrated separately. The steelman's case is a strong argument for the first track and is silent on the second. Real policy needs both.

Sex addiction, gambling, and where the perimeter actually sits

Anthony is empirically right on both halves of this. Sex addiction is not a recognised diagnosis. The DSM-5 hypersexual disorder proposal was rejected in 2013. The ICD-11 added Compulsive Sexual Behaviour Disorder in 2018 but explicitly classified it as an impulse-control disorder, not an addictive one — a deliberately conservative call by the working group. The Pathway page documents this. The neural circuit is shared; the clinical label is not yet earned. Pornography "addiction" sits in the same bucket.

Gambling addiction, on the other hand, is recognised. ICD-11 lists Gambling Disorder under "Disorders due to addictive behaviours" — the same parent category as substance use disorders. DSM-5 reclassified pathological gambling under Substance-Related and Addictive Disorders in 2013. The mechanism — incentive-salience sensitisation, variable-ratio reinforcement, comorbidity with depression and other addictions — fits. The treatment evidence base is weaker than for substance addictions but the diagnostic case is settled.

Disciplining the perimeter matters. The willpower critics are right that "addiction" has been used loosely to label patterns of behaviour that have not earned the clinical label. Defending the perimeter where it is currently drawn — substance addictions and gambling in, sex and porn out, eating-disorders separately classified, screen addiction not yet — is what gives the harder pharmacology-first thesis its credibility. The honest version of the position is not "everything is addiction, medicate it all" — it is "the things that meet the diagnostic criteria are real, and we are systematically under-treating them".

Where the position lands

What survives the teardown is a sharper, more defensible version of the steelman. Five planks.

1. 10 per cent of the adult population has a real addiction, mechanistically distinct from ordinary variation in self-control. Policy that treats them as ordinary self-control failures will fail at the dependent variable. Forty years of data confirms this.
2. For the addicted 10 per cent, willpower alone does not produce sustained recovery; willpower with pharmacological support does. Education informs but does not cure. The standard of care has to combine the drug, the behavioural model, and the social environment — not pick one and pretend it's sufficient.
3. Five drugs already exist that, fully adopted, would avert roughly 6 per cent of the global burden of disease. Varenicline + NRT. GLP-1 / GIP agonists. Methadone. Buprenorphine. Naltrexone (dual-indicated). Current adoption is ~10 per cent of the eligible population. The gap is a policy choice, not a scientific one.
4. Naltrexone is the simplest single move — one molecule, two indications, off-patent, cheap, modest NNT but enormous denominator. If you have to convince a sceptical minister of the case in one prescription, this is the one.
5. The blocker is moralising, at every layer of the system. The fix is the give-it-or-sell-it framework — public provision or competent market — together with explicit cultural permission for the addicted to reach for the drug without shame. Mounjaro is the live demonstration that the market mechanism works the moment the moralism is held at arm's length.

That is the position that survives the steelman's own teardown. Sharper than the original. Narrower than the loose form. Empirically defensible. Politically harder to dismiss because the loose claims have been pre-emptively conceded.

The lever, picked up

The original question of the microsite — is addiction real? — has an answer. Yes, for the 10 per cent who have the disease, mechanistically and consequentially. The corollary is the work this page exists to do. The drugs work. The drugs are not adopted. The arithmetic of the gap is enormous. The fix is supply-side and cultural at once: deliver the molecule, hold the moralising at arm's length, let the people who are sick reach for the thing that works.

The original Steelman page made the case that the lab-trained nurse who thinks Mounjaro is cheating is partly right and partly wrong. This page makes the case that the position arguing against her — the pharmacology-first, anti-moralising position — is also partly right and partly wrong, and that the honest version of it is sharper than its loose form. Both pages are doing the same job: holding the strong version of a position long enough to see what survives the teardown. The version that survives is the one worth defending.

The lever is right there. The five drugs are off-patent or cheap-to-generic. The trials are done. The deaths are documented. The Mounjaro precedent is in front of you in 2026. The next page documents what happens when we keep not picking it up.

→ Read the Policy page, which is what happens when the lever stays untaken.