legitimacy is judged locally, not by outcome survival

agent-decided, reversible. Whether an exercise is legitimate is now: did prior legitimate authority permit this exact edit, given this declared input and this evidence of authentic exercise. It is not: does the edit survive removal of the influence.

The case that decides it is ordinary persuasion. An agent with authority to revise is argued into a revision; remove the argument and the revision does not happen; challenge survival scores that as dependence. A legitimacy theory that cannot let an agent be argued into a revision is not describing the learning this programme exists to describe.

The cost is stated rather than absorbed: a counterfactual test can in principle notice a dependence the record does not declare, and a declared-input test cannot. The factorization hypothesis is what makes that safe, and it is falsified by a record whose effect reads an uncited settlement.

Rejected alternative: challenge survival, which the two previous passes carried as the central modal condition.

Record: projects/normativity/legitimacy/rounds/2026-08-25-legitimate-evolution/LEGITIMATE_EVOLUTION.md §7.