1.8 KiB
REPAIR36 — bounded preparation submitted, no acceptance credit
Changes are limited to the S5 wrapper, new pure UID controls and these documents. Actual-Q AST helper loading, finite derived UID planning, import sealing, private-globals phase instrumentation, phase receipts and cleanup rechecks supplement the existing expected inventory. Historical selection, existing require gates, ownership/PID, engine, cache and formal mirror rules remain. The helper is loaded from real Q before controls construct synthetic refs.
Only py_compile and AST checks are executed and recorded in REPAIR36_STATIC_CHECKS.json. Wrapper fixtures, UID controls, check-only plans, Git negatives and every native diagnostic are unexecuted. No functional pass, full regression, 15+EV batch, original credit or completion is claimed. No Git, engine, Runner or other agent operations were performed. A shell heredoc edit failed before Python execution; edits then used apply_patch.
Context gap: Q runner API was read from build/owner-handoff/p04-after-p26-closure-r01/q35-uid-diagnostic/run_all_tests.py.reference; Root must confirm the actual Q runner matches that mirror (320ca668…) at runtime. Helper implementation was read from the adjacent P13 reference; runtime loading uses only the actual pinned Q, never either mirror.
Pending: S5 destruction/midnight/exit drivers; S1–S4/collector fresh Q;
derived 184 suites / 329 parse; truthful index; Root exact-candidate qualification.
Root order is py_compile → wrapper fixtures → UID controls with explicit scratch
→ fresh check-only plan → exactly one new-name diagnostic. The failed r35
diagnostic is preserved. All 7 domains + EV, formal 16/88, original 30/88,
P04 credit 0, all six D05 approvals pending, M c0ea009 and allowed_paths stay fixed.