authorship of a future evaluation is exclusive binding plus reason mediation, read as a derived predicate over the answer receipt and its payload

agent-decided, reversible. The authorship condition that enters the activation event of an evaluation occurrence is ReasonMediatedAuthorship.Authored: the binding event on every audited world is the principal's, and at a fixed principal-side policy the committed payload factors through the declared reason view (reasonMediated_iff_factor). Its content is blind_of_mediated: a reason view blind to the declared prohibited channels makes the payload blind to them, so the regret yardstick cannot be steered through such a channel (yardstick_invariant); an injective reason view makes the condition vacuous and a constant one recovers the isolated-principal reading, so the declared view carries the normative content and is external. Authorship is placed outside the generic protocol: Protocol.AnswerOK is evaluated at the strict prefix and cannot see the answer event, so Bind and Authored are a derived predicate over the receipt's event, the payload the authenticated history carries there, and the session's process receipts; exclusive binding is the existing Authority.warrant / Authorized typed as the principal role's binding warrant. The future evaluation is a partial object Ṽ on certified worlds; activated securities are completion-invariant (activated_completion_congr), activated regret is the activation mass times a conditional authoritative regret defined from Ṽ alone (regretU_eq_mass_mul_regretAuth), and every bounded completion's regret lies within D·η of activated regret in either direction (regretV_sub_regretU_abs_le; corrected by the 2026-09-08 consolidation entry — the one-sided interval first recorded here was false for world-dependent strategies). The consumer's Integrity hypothesis is the occurrence-local trace LocalLegit (OccurrenceLocalIntegrity.lean), a projection of PR #92's LegitimateForSegment (toLocalLegit); an unrelated occurrence's Integrity failure does not void an intact evaluation (Witness.unrelated_integrity_failure). Selection blindness is a separate condition on the advisor's whole session policy that authorship does not imply (Witness.leak). Rejected alternatives: similarity to a no-AI counterfactual judgment as the authorship criterion (it is the constant-reason-view end of the same family); placing Bind/Authored inside AnswerOK (not expressible without changing the generic field); the stochastic kernel form as the first theorem (a corollary of the per-seed deterministic form under seed privacy); literal no-preview as the admissible-domain condition (fixture H); global Evolution as the consumer's Integrity hypothesis (fixture J).