An intervention’s intended effect — the one its mechanism predicts in isolation — is routinely wrong, because the whole-organism response adds a term the mechanism omits. What matters is the net effect across the whole system, and the omitted term is signed: usually it runs against the intervention (net smaller than the mechanism implies), sometimes it redirects onto another outcome (a real targeted effect netted away by an offsetting harm on a different pathway), and sometimes it runs with the intervention (net larger than the mechanism implies — when the intervention meets no compensatory brake and instead recruits a reinforcing response). The diagnostic is therefore not the narrower what compensates? (which presumes the term subtracts) but the signed what does the whole system do in response — damp, redirect, or reinforce?
The failure mode is dosing or recommending straight from the naive mechanism — add X kilocalories of exercise, lose Y weight; lower the marker, get the outcome — as if the body were an open loop. It is a closed loop with counter-regulation, so the arithmetic over-predicts.
The worked instances (already source-grounded on their pages)
Each restates a finding held on the linked page; the selection and framing as compensation-instances is this page’s synthesis. The first four run the attenuating or redirecting way (the term subtracts or diverts); the last is the amplifying mirror (the term adds). The two limbs are not equally evidenced: the attenuating direction rests on four instances, the amplifying direction on one, whose intake -> weight step is itself inferred, not measured — so “the term can add” is held as a real but single-instance generalization, not a symmetric law.
- Exercise -> the body compensates. Added exercise energy is partly offset by eating more and moving less the rest of the day, so the net energy deficit is smaller than the sessions “burn” -> Exercise Energy Compensation. The naive “calories out” arithmetic over-predicts fat loss.
- Weight loss -> the body defends its mass. Energy restriction triggers adaptive thermogenesis and hormonal counter-regulation (hunger up, satiety down), so net loss falls short of the deficit and weight is actively regained -> Weight-Loss Maintenance and Metabolic Adaptation. The intended linear loss meets a defended set-point.
- Drug effects can be undone by withdrawal or by an offsetting harm. A benefit sustained only while a drug is taken is not a durable net effect (weight regain after GLP-1 discontinuation); and a benefit on one outcome can be offset by a harm on another, so all-cause net effect can be null even when the targeted effect is real. (The general form; the specific drug instances live on their own pages.)
- Anti-inflammatory drugs -> events down, all-cause mortality flat. The sharpest worked instance: two RCTs (canakinumab in CANTOS, colchicine in LoDoCo2) cut cardiovascular events by dampening inflammation, yet neither reduced all-cause mortality — the vascular benefit was offset by a rise in fatal infection (CANTOS) / noncardiovascular death (LoDoCo2). A real targeted effect on the intended outcome, netted to null on survival by a harm on another pathway -> Inflammation as a Modifiable Lever.
- Alcohol -> the body does NOT compensate, and eats more (the amplifying mirror). The naive cost of a drink is its own calories; the net is larger, because those calories are not offset AND the drink provokes extra food intake in the same occasion (food energy +343 kJ, total energy +1072 kJ vs a non-alcoholic comparator) -> Alcohol and Mortality and Vascular Disease. This is the sign-flip of the exercise case: there the whole-organism response subtracts from the intended deficit; here it adds to the intended surplus. Same principle (net != intended via the whole-organism response), opposite direction — which is why the diagnostic must be signed, not assumed to subtract.
The failure mode it names
Mechanism-to-dose extrapolation without the compensation term. A real, human-corroborated mechanism (fructose -> de-novo lipogenesis; exercise -> energy expenditure; a marker lowered) licenses a direction, not a magnitude — the magnitude requires the whole-organism response, which may attenuate, redirect, or amplify it. This is why a mechanism admitted is admitted for direction only, discounted, and never dressed as an outcome finding.
Where it sits
The placement and cross-links below are this page’s reasoning.
- Sibling of The Comparator Problem (what the effect is against) and The Estimate-to-Action Gap (turning an estimate into an act): three standing checks between a mechanism/estimate and a claim about what will actually happen to a person.
- It grounds the Layer-3 watch-for-compensation / net-effect-not-intended rule (CLAUDE.md): judge a complete, realistic strategy on its net effect, not its intended one.
- Method grounding is top-down from the meta-method corpus (realist mechanism vs the whole-system outcome) — cited up, not as a domain source.
Synthesis
It earns a page because the same correction — account for what the whole system does in response, which may subtract, redirect, or add — recurs across otherwise unrelated domains (activity, weight regulation, pharmacology, appetite) and changes the decision in each: it caps what a mechanism can promise, and it is the standing reason this wiki treats a plausible mechanism as a direction-only prior, not a magnitude. The generalization from “compensation subtracts” to “the response is signed” is what the alcohol instance adds to the exercise one — same structure, opposite sign, so the correction cannot be assumed to shrink an effect. Configurative, not aggregative: no shared number, a shared correction.