P04-01: 同時発生の更新規約

2026-10-16 08:35:40 +09:00
parent 16207d353a
commit 8418560aaf
17 changed files with 2011 additions and 932 deletions
@@ -0,0 +1,82 @@
+# Repair22 — source-ready submission; NOT RUN
+
+This batch changes only `tools/check_p04_order_evidence.py` and this P04-01
+documentation subtree. No Git command, test/fixture, engine, Runner operation,
+agent, manual commit or integration was performed. The normal autosubmit owns
+the submitted revision; the contract base SHA is not a new evidence version.
+
+S3 B1 no longer supplies `power_stop`. Its full diagnostic checks remain under
+`dependency_closure_diagnostic`, including the actual dependency reason,
+closure/release, departure history and recovery sale. It approves no priority.
+Strict power-vs-closure order remains D05 pending/unknown. The existing approvals
+cover dependency hold and the18:00 shift interval only; A01 is not reopened.
+
+S4 is a new minimal single-case QA probe using the existing ActorHardening,
+physical-kitchen and guarded Writer/Reader pattern. It cuts the actual sole
+cafeteria edge at959, waits for normal evaluation/projection at960, retains the
+held state at961, reconnects that same edge, and waits normally through the
+source-derived hold until970. Source constants were re-read as10/10. No forced
+evaluation, hold bypass, direct power/progress/patience/capacity writes or
+product edits are introduced. Kitchen Facility.OPEN / network ACTIVE, positive
+mana, operating dependency and open operator distinguish it from B1.
+
+Three formal Writer points and separate-process Reader restores retain every
+typed Runtime field and all four SaveService members. Reconnect resend at970
+must leave the full Runtime unchanged; normal971 and Reader replay add no
+extra sale/completion. Full intermediate phase raw is preserved. All graph
+nodes, cut/restored edge sets, revisions, allocation and derived dependency
+restoration are checked. Normal post-completion guest redecision/despawn is
+retained, not forced or concealed; no exit accompanies the outage/hold.
+
+R now derives shift PID and sale IDs from the existing full raw joins. S1 joins
+the existing pre-travel reservation/task/staff cell to the boarded task/trip,
+and checks actual mana0 and elevator loss of power. The existing join fixtures
+are retained and extended with the requested semantic negatives and consistent
+dynamic-ID positive controls. Existing C25 and all other fixture sources remain
+in place. No fixture result is claimed: S4 positives await actual Root raw.
+
+ORDER_AUDIT.md and order-audit.json are one current seven-domain table with
+normative rules, source, C/S1/S2a/S2b/S3 evidence, pending S4, approved authority
+and pending D05. REPAIR17–21 remain untouched historical documents and outputs;
+their current-status claims are superseded by repair22. The current protocol,
+collection plan and this status agree: no final acceptance.
+
+Static validation is recorded in REPAIR22_STATIC_VALIDATION.json:
+
+- Seven Python files parsed with AST and compiled in memory; none imported or
+  executed. GDScript received static text/source review only; engine parse NOT RUN.
+- The entire frozen C region is byte-identical to the preserved full-stage
+  checker. Its one-final-LF body hash is
+  `33ae2af6afb98b48919ed679ab8519b7da7ddfda68c00f71c557d5cc9aad4a6c`, matching
+  the supplied repair21 C_FULL_BLOCK_EXACT receipt. Existing C checks are intact.
+- The complete15-condition registry matches that immutable source; EV, seven
+  domains and three safety outcomes are unchanged. Product1338 inputs and the
+ 316-file `.gd` parse roster were counted from the actual relocation manifest.
+- The five supplied public power bodies were compared to their actual relocated
+  source. Full bodies, hashes, timing and restore analysis are retained in
+  REPAIR22_SOURCE_FACTS.json.
+
+The exact finite Root plan is `s4/S4_ROOT_RUN.md`: one fresh own-B four-phase
+collection, with explicit arguments and full input/QA/source footprints. All
+old C/S1/S2a/S2b/S3 raw and all R17 sources remain accessible and are not rerun.
+S4 remains **missing/pending Root execution**. Future native/index/cross-gate
+work does not block this source submission. The updated collector/index
+references use a fresh r22 destination; no r17/r20 evidence is overwritten.
+
+Preserve the21615-check proof. Static full-map comparison found exactly the
+unchanged product1338 plus the old checker in its1339 map. The new checker hash
+differs; never claim1339 byte identity. The178 product suites do not invoke the
+checker and `.py` is excluded from the316 parse roster, so reuse requires the
+explicit QA-only source/config/input/environment applicability argument in
+EVIDENCE_COLLECTION.md. New-checker validation and S4 are separate, pending.
+
+Root must rebind support53cb through its normal registered task once the new
+source is known, without metadata bypass. Only if an exact final1339 full proof
+is required, perform one run after all fixes, S4 raw, local checks, eight suites
+and rebind are bundled; do not splice versions or perform a blanket full now.
+
+No required source context is missing for this preparation. Remaining D05
+choices are explicitly pending and not silently implemented. Formal validation,
+independent review, exact final-source applicability and normal integration
+remain pending. CurrentAlphaScope,30 packages/88 deliveries, original13/88 and
+P04 main9b are unchanged. P04 receives no credit; nothing is integrated.