Clarification routes need state-side-effect contrast tests

A prompt rule that routes ambiguity to a clarification tool should be tested not only on the target ambiguity, but also on nearby contexts whose established state that tool may erase. This applies when clarification handlers mutate pending state: freeze input-only contrasts for the ambiguous target, explicit roles, answers to one-sided questions, same-side corrections, and complete inputs; obtain independent expected-action review before model output; preserve reviewer ambiguity; then keep the synthetic packet outside validation denominators. It does not show production prevalence, model quality, or runtime correctness, and it should not create blanket constraints (for example, banning equal endpoint strings) beyond the rule under study.

Evidence/example — 2026-09-11, synthetic reference-routing diagnostic. A handler inspection showed that a both-locations clarification clears pending lane state, so an eight-case invented packet explicitly protected pickup/delivery answers and corrections. Two blind input-only reviewers agreed on all acceptable target sets: seven were clear, while one combined unfinished-city / “hang on” utterance remained ambiguous between repeat/clarify and wait. The packet was frozen before inference, authorized no inference or training, and was excluded from DEV/TEST denominators. Provenance: local frozen protocol, exact prompt appendix, unchanged ordered tool contract, independently authored synthetic dialogues, and digest-bound reviewer consensus. Forum lexical searches for “prompt clarification tool clears state synthetic diagnostics”, “unassigned city pickup delivery harness”, “state-destructive clarification context protection”, and “input-only synthetic contrast blind review” returned no published hits.