- evaluate_served_ask now validates the full Q1-D DeliveredQuestion contract:
status == 'question_only', requires_review is True, served is False,
answer_binding is None, question.text non-empty, question.slot_name present,
question_path != proposal_path. Each clause is individually testable via
_validate_artifact() which is a pure predicate with no side effects.
- Added _MISSING sentinel to distinguish absent keys from explicit None/False
values in 'served' and 'answer_binding' checks.
- Added load-bearing comment in _stub_response explaining why served ASK
bypasses register decoration and realizer guard: Q1-C/Q1-D is the
authoritative grounded renderer; mutating pre-rendered intake prose would
be semantically incorrect.
- Expanded test_ask_serving_integration.py with:
- Rejection tests: status='proposal_only', served=True, requires_review=False,
non-null answer_binding, missing/blank question.text, missing/blank slot_name.
- Exact-text bypass safety test proving stub surface == artifact text exactly
with no register/decorator mutation and realizer_guard_status='ok'.
- Disposition telemetry tests: default main-path stability, proposal terminal
not reclassified, ASK branch emits 'ask', refused terminal stays 'refuse',
ChatResponse.disposition defaults '' for backward compat.
All 440 tests pass.
