When the Human Is the Hidden Answer Key: Operationalizing Recovery Responsibility and Answer-Donation Contamination in Long-Horizon AI Evaluation

This paper develops Answer-Key Burden (AKB) into a narrower evaluation protocol for recovery responsibility and answer-donation validity; it does not claim conceptual novelty for the AKB framework itself. The protocol uses three controls. First, a minimal provenance receipt makes the freeze-before-observe order externally auditable by recording the committed rule state, frozen evidence state, completed Phase-A responsibility/action record, and first outcome exposure as ordered events. Second, absence of a nonhuman authoritative source can support HUMAN_AUTHORITY_REQUIRED only inside an authority-grounded locally closed source domain; outside such a domain, missing evidence remains UNRESOLVED and the action is HOLD_FOR_RESOLUTION. Third, the primary responsibility-effect outcomes are pre-attribution objective Phase-B events, timing, and counts; human semantic ratings are secondary or sensitivity-only unless assessors are genuinely blinded to condition-determining information. Phase C attributes candidate AKB only after those outcomes are observed. The paper specifies a protocol; it does not establish coding reliability or human-effect evidence.