Source changed 2026-10-03 05:29:54 UTC · snapshot created 2026-10-03 05:30:57 UTC · last check 2026-10-03 06:00:20 UTC
I.2.8 - Causal-use and counterfactual-support repair
Case signal: “This policy would have prevented harm”, “this intervention caused the improvement”, “this fairness result is causal”, “this method is better on counterfactual outcomes”, or “these simulated counterfactuals prove the decision”.
Initial uncertainty: the reader may be seeing association, a metric disparity, temporal change, method execution, work-plan use, work occurrence, simulation output, deontic boundary language, or a real causal-use claim.
Tempting wrong pattern: use D.5 to treat metric fairness as causal fairness; use G.9 to compare methods across different causal rungs; use C.26 to hide a causal-support question under quantum-like wording; use C.27 to treat rate change as causal effect; use A.15 or A.3.2 to treat a sampling method, intervention procedure, or target-trial recipe as causal support by itself; use A.6 to turn causal evidence into a duty or release gate.
Disambiguating fact: the decisive question is not whether a causal-looking word appears. It is whether publication, choice, deployment, assurance, audit, benchmarking, or method dispatch relies on causal support governed by C.28: effect, intervention success, counterfactual comparison, causal fairness, policy optimality, causal evidence support, off-policy/causal-RL evaluation, or causal method superiority.
Recognition repair or question reclassification: if only a measured value is live, repair in C.16; if only rate, trend, or temporal adequacy is live, repair in C.27; if only method, work, or work-plan structure is live, repair in A.15 and A.3.2; if only boundary duty or agreement language is live, split with A.6; if residual QL modeling is the live question, use C.26 only when an exact contextual-model obstruction still changes inference or action after the applicable ordinary subject assertion and predicate are in place.
Actual governing FPF pattern body or projection role:C.28 carries causal-use question, causality-ladder rung, claim kind, causal estimand, identification, counterfactual sampling realizability, causal support components, one causal-use support result, supported use, and unsupported use. A.10 governs the relied-on evidence/provenance paths; B.3, D.5, G.5, and G.9 use the C.28 support result when the current downstream question is respectively assurance, ethical/fairness audit, method dispatch, or benchmark parity.
Admissible entry stop: a cheap downgrade sentence, a local CausalUseTriageRecord, a local or durable CausalUseEvidenceDesignRecord, a CausalUseSupportResult, or a named neighbor-pattern use that cites C.28 without claiming broader authority.
What not to infer: a randomized procedure is not automatically counterfactual support; a simulation is not realized counterfactual data; a target-trial phrase is not proof of identification; a fairness metric is not causal fairness; a method benchmark is not comparable if methods sit on different causal rungs or estimands; and a causal-use support result does not by itself create a duty, promise, commitment, release gate, or admissibility predicate.