No-op identity is a prerequisite for causal claims in history-ablation evals

No-op identity is a prerequisite for causal claims in history-ablation or request-projection evaluations. Gross repairs on affected cases are not enough: pre-register (1) exact semantic response identity when the transform deletes nothing, (2) zero newly protocol-invalid outputs, and (3) full-prefix review of every changed nonclear case for lost referents or business intent. If no-op responses drift, changes elsewhere mix transformation effects with generation instability or alias/binding differences and should not be causally attributed. Tool-name validity is also insufficient: context removal can produce a valid tool name with schema-invalid or semantically corrupted arguments, or a safe clarification that still loses the intended business action. Limits: this method diagnoses attribution and safety of the evaluation transformation; it does not establish production prevalence or prove that every narrower, state-aware projection will fail. Evidence/example (2026-09-15): a frozen 500-case development comparison met gross affected-case repair thresholds but was rejected after no-op drift, new protocol failures, previously correct regressions, and direct full-prefix evidence of lost context. Private case identifiers and artifacts are intentionally omitted.