P04-01: 同時発生の更新規約
This commit is contained in:
Agent Control Runner
@@ -0,0 +1,105 @@
|
||||
# RUN47 submission — bounded S4 admission repair
|
||||
|
||||
P04-01 remains **unaccepted**. This is the same original 30/88 unit, formal credit
|
||||
16, P04 credit 0. No completion, integration or native-success claim is made.
|
||||
All edits are under the contracted execution directory. Product code, registered
|
||||
spec, checker (including REQUIRED/REACH_REFS), denominator, caps and permissions
|
||||
are unchanged. No Git writes, other agents, Runner actions or engine launches.
|
||||
|
||||
## Changes and count/roster audit
|
||||
|
||||
- `s4/run_s4_native.py`, former line 301: removed current 316-script admission.
|
||||
`admission` compares the whole sorted source-derived `.gd` res:// list. It runs
|
||||
before check-only writes its plan and again before native writes parse-paths.
|
||||
Plans carry the exact list; missing/extra/duplicate/non-.gd/same-count changes fail.
|
||||
- Former line 240: removed selected/product 1339/1338 assumptions. Cache admission
|
||||
requires the exact current product map, with precisely the checker as its extra;
|
||||
the real product/selection difference allows no other input. Existing hashes,
|
||||
failed-cache rejection and no-adoption guards remain. Option B uses no cache.
|
||||
- `p04_native_series.py`, former lines 73/187/1030: removed literal source counts;
|
||||
retained whole source/manifest/input/source_after equalities and pin checks.
|
||||
Result count comes from the actual manifest. FreshSeries S4 also calls the
|
||||
admission function and compares plan and actual parse-paths.
|
||||
- `s4/receipt_lifecycle_fixtures.py`, former lines 81/92: split historical-pinned
|
||||
lifecycle from current-derived admission. It now explicitly loads the unchanged,
|
||||
hash-pinned r28 wrapper, retaining historical 1338/316 synthetic fixtures. The
|
||||
standard runner comes from the supplied retained M copy. Current cases live in
|
||||
`s4/admission_controls.py`; another valid derived roster is accepted as well.
|
||||
- Remaining 316 in current controls is only the deliberately stale negative and
|
||||
the historical AST/receipt fixture. The current wrapper has none. The 329 pin
|
||||
diagnoses this supplied M fixture only; production has no fixed roster size.
|
||||
- Resolver's 1721 members/71488772 bytes are the immutable historical relocation
|
||||
identity, already bound to the fixed manifest hash; retained. S4 phase-cut
|
||||
121/132, eleven kernel phases, three slots, four engine phases and case roster
|
||||
are scenario/receipt invariants rather than source counts; retained unchanged.
|
||||
- `S4_ROOT_RUN.md` now distinguishes the current derived rules from historical
|
||||
relocation/count descriptions. Its parse-path receipt row describes the new
|
||||
admission. Earlier 1338/316 observations remain historical. S2b/S3 normative
|
||||
fixed-product instructions are explicitly superseded. S1/S2 source-count
|
||||
observations describe earlier runs, not current admission. ORDER_AUDIT,
|
||||
EVIDENCE_COLLECTION, CURRENT_STATUS and REPAIR47_PLAN are synchronized.
|
||||
- `checker_series_fixtures.py` uses BUILD for disposable fixtures. No test was
|
||||
removed or disabled.
|
||||
|
||||
Before import is attempted (`phase is None`), cleanup requires the complete
|
||||
original source inventory with no generated UID. After import is attempted, the
|
||||
existing exact planned UID snapshot/seal still applies. Partial/unknown UID sets
|
||||
and changed source fail in both branches. The original exception stays in `error`;
|
||||
cleanup errors remain separate guards, and the pass predicate still requires a
|
||||
sealed import, all four successful phases and owned PID cleanup. Engine identity,
|
||||
fixed inputs, default-player protection, source checks and first-failure behavior
|
||||
are preserved.
|
||||
|
||||
## Q46 failure, preserved verbatim
|
||||
|
||||
```json
|
||||
{
|
||||
"passed": false,
|
||||
"error": "AssertionError: complete 316-script parse roster",
|
||||
"guard_errors": ["AssertionError: exact source and planned UID set"],
|
||||
"phases": [],
|
||||
"actual_captured_PIDs": []
|
||||
}
|
||||
```
|
||||
|
||||
No S4 game verification ran. REPAIR47_Q46_FAILURE_REFERENCE.json records the raw
|
||||
file hash and verbatim failure fields. The original raw and historical wrapper
|
||||
were not modified. Q46 successes elsewhere are reference only, never adopted.
|
||||
|
||||
## Engine-free verification
|
||||
|
||||
REPAIR47_VALIDATION.json includes commands, exit codes, raw-log hashes and full
|
||||
finite outputs. The final current admission program passed 34 controls: actual
|
||||
329 scripts, another derived roster, checker-only selection extra, every required
|
||||
roster rejection, actual check-only main admission and its six rejected mutations,
|
||||
the real FreshSeries roster block, the real cache set guard, historical 316 via
|
||||
pinned r28 AST, and pre-import/attempted-import UID/source rejection cases.
|
||||
Its check-only unit explicitly stubs historical template and engine identity;
|
||||
this is a synthetic control, not a fresh-Q provenance plan or native proof.
|
||||
|
||||
Affected series (62 controls), join (29 controls), and historical lifecycle (six
|
||||
cases) passed. Historical lifecycle was finally confirmed against retained M.
|
||||
Five changed Python files passed AST and compile checks; bytecode and scratch are
|
||||
in allocated BUILD. Initial control-development failures (two absent contract
|
||||
paths in the product-only M copy and one synthetic UID owner outside the existing
|
||||
plan's allowed selection) are recorded, not treated as passes. Their fixture
|
||||
references/owners were corrected; the final finite runs passed.
|
||||
|
||||
REPAIR47_PLAN.json records unchanged hashes against S46 for the eleven unaffected
|
||||
pure programs, their prior Q46 results, and unchanged top-level series function
|
||||
hashes. Those eleven programs were not rerun. This is reuse of unchanged pure
|
||||
checks only, not adoption of native receipts. The affected series/join checks were
|
||||
rerun. The checker file is byte-for-byte unchanged (SHA256
|
||||
`3bb0cde776bc295f6f18e74f1e23bf3d832d8e4f71aed260601c82d287c655fd`).
|
||||
The historical wrapper remains
|
||||
`2d3be39d2b310abac058e27336ac816c234043a0a1dd7b2c52d84994907dd31b`.
|
||||
Spec raw bytes remain unchanged; the existing one-final-LF dispatch identity
|
||||
protocol yields the contracted `9ec722e7…` hash (REPAIR43_PRECHECK records it).
|
||||
|
||||
Formal six-series check-only plans were not run by this worker: they require
|
||||
Root's clean final Q/provenance. No claim that W check-only passed, or failed only
|
||||
because of dirtiness, is made. No native/full/registered acceptance test ran.
|
||||
Root must follow the complete option-B sequence in REPAIR47_PLAN.json: freeze Q,
|
||||
affected pure, six clean plans, capacity/PID checks, one fresh native batch,
|
||||
readers/adapters, fresh all-15/EV/subclauses, full 184 once, index, verify-only,
|
||||
registered p04-acceptance before independent Opus review, then normal integration.
|
||||
Reference in New Issue
Block a user