Pawn first writer: serialized launch refused; trace coverage reviewed
The fourth record's actual first writer and a valid second native reply remain
unproved. No Pawn original, watchdog, input or box was created in this run.
No production change or emulator defect is claimed. Evidence is in ignored
scratch/runs/20261113-pawn-first-writer-1550/.
Verified all 169 seals from the 1445 modalguard parent and all 57 seals from
the 1520 writer analysis. Independently rehashed the whole original 100-byte
save in its existing location to
811659988ee98f19ca08615f1191431993f5d51646ea41efed5fdb6f0ff5cf77.
It was neither copied nor edited. Its e4/Nf6/d4 records and fourth
0,44,7,15,0,0 record retain exactly their parent's acceptance limits.
Concrete runtime blocker
LMarbles1550 announced its resource phase at 03:36:18Z and RESOURCE_DONE at
03:36:56Z. Prepared the qualified owner/modal/finite-drag helpers, unchanged
strict queue predicates, and bounded 0x402a50 trace with limit four. Frozen
Git was ef38e35c8eb3ed590fe3393def99bc6b034d691f; all 322 pins were computed
from actual current files, including newly enumerated source/library paths.
The actual module SHA was
6888cc5e4ec7693a8a317fe29147994c7f88fe43afe3109fe6ab0e3e349eec9f.
Git, relevant binary diff, run-helper patches and path/hash pins preserve
provenance without archiving source trees or copying media.
The single supervisor acquired /tmp/wine-heavy-runtime.lock, then its actual
process scan found Node test/run.js PID 1125248. It refused serialization
before reaching the floor/launch decision, arming its independent watchdog,
or creating Pawn. The scanned child's ownership and command beyond that match
are unknown: it was already absent at the follow-up read. Do not attribute it
to another worker or infer lock misuse from that incomplete identity.
The follow-up available-space sample was 1,498,624,000 bytes, also below the required 1,500,000,000 floor. No retry or independent recovery followed. Supervisor PID 1125257 actually waited 1, with separate stdout and stderr EOFs; its retained stderr contains the refusal assertion and PID. The outer tool completed exit 0 because it reports the supervisor's result. This is not native runtime success. Native/watchdog waits are N/A, not started.
First-writer evidence conditions and existing hook limits
Reviewed the existing hooks read-only while waiting; details and source hashes
are in trace-coverage-review.json.
test/run.js's trace-at compares returned EIP with the target, and the WAT breakpoint checks EIP at block dispatch.set_bpdoes not itself enable instruction stepping or split an interior instruction into a block entry. Native0x402a50follows the0x402a4cstore inside a straight-line sequence after return at0x402a22. Its prepared trace therefore has a coverage prerequisite: establish that this instruction is an actual block entry. A missing trace hit would not establish that the store never executed.- The ordinary watchpoint checks value changes at block boundaries. It can establish a changed value and predecessor block, but does not by itself establish the precise instruction or detect a same-value write.
- Existing
set_instruction_watchpointcan publish one instruction per decoded block. Existingnative_store_arm/getrecords overlapping-store epochs, last writer PC/address/width and host/invalid/overflow indicators. These are available primitives, not observers exercised against Pawn here. A last-writer receipt establishes the first writer only with a proven baseline and continuous scope, epoch delta exactly one, no invalidation, overflow or unknown writer; otherwise a complete ordered store journal is needed. Selected PV byte provenance needs its own continuous observation.
Thus a trace of static candidate 0x402a4c alone must not be promoted to the
first actual record-four writer. The known book/search branches and root-row
clearing remain hypotheses until observed on the original. This hook review
does not establish a Pawn API/interpreter fault, justify a failing gameplay
regression or warrant a production fix. No guest calls were forced and no
guest/save/position/health/AI/configuration/clock state was written.
Checks and next step
Both parent seal sets, whole-save hash and eleven helper syntax checks passed.
No unchanged control matrix was rerun. Postcheck found concurrent changes to
test/run.js; they are separately owned and captured read-only as a binary
diff. They were not part of any executed Pawn runtime. The other frozen pins
still matched. This documentation-only change claims no new build/test of
production code and no gameplay acceptance.
Next requires sustained disk headroom and zero actual runtime children under serialization, a new freeze of the then-current source/module, prearmed observable independent watchdog, and qualified observational granularity for the first write and selected-byte provenance. Preserve the whole native save, review e4/Nf6 before d4, and send no third move or replay. Box ID none, state NOT_CREATED, expiry null, stop receipt N/A.