Independent acceptance of native synthetic event-observer qualification

PASS, offline review of candidate f79ac8af5876ceb7356a07b16d47003719cac29c, tree 65cfcd8817b0a9e9a2080ed298ffc8cf61422b6c, based on ff594b75bc500a12200302e87dd963b277b5ac56. The producer executable and observer are unchanged. Acceptance reuses the producer's single native process and actual retained-module instantiation; this review performs no native execution, compilation, instantiation, game execution, download, deployment or box operation.

Independent findings

The 15,663-byte executable SHA-256 is 2159e84c0960679855b330adf813f94351cb59f4a5cc09fc9ef27e91b7ce303c. The execution-code identity, exact native argv, natural exit-zero terminal, raw output, parsed result and six raw records bind the reported native result. Pinned module, observer, runner, exports and scheduler identities pass unchanged preflight against the standing local inputs. Packaging reuses the coordinator's verification of all 20 sealed payloads; independent acceptance targets semantics. This is receipt-based acceptance, not an independent reproduction of execution.

OFF and ON use disjoint counter addresses and both record [0,0] before calls. Only EIP within the lane code span and EDI's callee address become relative to the lane base; all other fields remain absolute. Independently constructed expected states match all six rows in both lanes. Absolute native ON records also match the normalized rows and cache baseline/deltas. Independently packed stack/fdset/timeval bytes produce the retained scoped hashes. The scope excludes the rest of the arena and CALL's temporary stack slot below ESP. Cache equality is equality of measured store/clear/invalidation deltas, not cache contents. Starting registers, yield, breakpoint and counts are explicitly configured; pending returns precede execution, and the final pending interior call consumes the prior breakpoint skip state before the next lane. Matching measured transitions support equivalent fixture states without claiming every global reset.

Pending reason 8 produces halt 4 without dispatch. In the normal call, EBP changes to EAX, ESP returns to its initial value and EIP reaches head+4 with halt 5, consistent with the written MOV/CALL and native RET fixture. Entry counters increase before dispatch and remain explicitly unsuitable as execution proof. Invocation 3/batch 50002 observes native thunk words [4096,934]; unrelated id 935 does not create a select record. Head/interior occur at invocations 4/5, outside the old fixed batch window. All six records total 4,124 bytes.

The select stack, fdsets, timeval and thunk words were deliberately written synthetic metadata, read by actual native getters. The select API never runs. The pinned scheduler's checkMainYield calls noteMainSleep; its reason-8 path returns false without clearing yield. Only this controlled-owner path is covered; scheduler internal runSlice and cooperative-send dispatch remain unqualified. Source inspection confirms the native coercion fixture invokes both raw and wrapped Wasm functions with a valueOf that throws the same sentinel. Native PASS supports that assertion. Mock throwing/argument/receiver controls are separately identified and are not represented as native coercion evidence.

Cheap checks and rejecting controls

Unchanged candidate syntax and exact standing-argument --preflight pass under 256 MiB Node heap, CPU/wall 60 seconds and 131,072-byte output limit. Preflight also passes with WebAssembly compile/instantiate/Memory/Module/Instance replaced by throwing sentinels, demonstrating it stays offline. Its six mutations bind the production function source and reject reason-only attribution, an extra run, dropped arguments, swallowed exception, ignored record cap and wrong head event.

The adjacent independent review-controls.py accepts the retained receipts and rejects seven corruptions: false OFF/ON equality, a shared wrong normalized field, wrong native module identity, mock-only native instantiation claim, missing native event, changed executable and failed native terminal. These are receipt integrity and semantic checks; they do not manufacture new native execution evidence. Exact integrated-tree cheap checks and publication identity are retained in operational exact-tree.json and publication.json under the owned ops directory.

Resources, history and limits

Prepared sparse checkout reused; no clone, fetch, donor modification, new pack or broad materialization. Read-only alternate stores are standing dependencies. Initial complete projection was 53,121,528 / 67,108,864 bytes including full external launcher/session logs, conservative SQLite WAL charges, future growth, publication copies, narrow captures and 8 MiB reserve. Resource receipts recheck host disk floor 2,982,150,144 and MemAvailable 805,306,368 before commands. Owned ops retain manifests, bounded commands, terminal/reaping receipts and final projection. Report/evidence are committed; Git does not preserve external logs or ignored module bytes. Recovery uses the pinned standing inputs and producer receipts, never another scarce native attempt.

Historical oversized donor-fetch failure remains at /tmp/maelstrom-event-native-independent-integrate-oct24/ops/result.json: 25,165,824 truncated bytes and 81,868,568 projected peak exceeded its allocation. It was not a semantic rejection. Coordinator reused an already prepared sparse checkout and corrected alternate ordering without mutating donors. Earlier broad reads, unwrapped initial checks, late SQLite baseline and author identity correction remain explicit in producer/history receipts. This review's initial instruction/inventory reads had timeout but not full process/output supervision; SQLite baseline was late, so all extant WAL is conservatively charged. These deficiencies are not retroactively declared compliant.

The first publication staging check rejected trailing whitespace in this review's copied identity.log. That redundant publication copy was removed; the raw ops log and exact first failure remain. A compact identity receipt replaces it. This corrects review packaging only, without weakening any semantic oracle or changing the producer executable or evidence. The copied diagnostic itself quoted that whitespace, causing a second staging rejection. Both exact diagnostics are retained as JSON strings, preserving their bytes without introducing whitespace errors into a textual diff.

Coverage is invocation boundaries only; visits wholly within one call remain unobserved. Deadline is unavailable, API completion unobserved, API result unattributed. Runtime overhead, whole-arena equivalence, source-to-module rebuild provenance and original-game causality/gameplay are unmeasured. No new production fix was needed. No box is assigned: historical bx_qp99542b was reported stopped, bx_69aem736 is untouched, current state/expiry unqueried. Coordinator owns lifecycle. Root tracked checkout/index/refs and all donors remain read-only; only the append-only messageboard is updated. No workers spawned or refilled.