TuxRacer closure/publication independent acceptance, 2026-10-28

ACCEPT for static closure and ordinary-exception publication rollback of candidate a52929b9faf82c58b57d11c550120f6bd687e1d0. Production recipe SHA256: 567a716c33e97d9c8547a02c77ff10dc1839308afb45c3fe9848effb60a3134d. The recipe and both inherited tests remain unchanged. This commit adds only independent tests and this report.

Behavioral verification began with the unchanged historical stage reproducer and its exact pinned historical production recipe. Its exit 0 documents two historical REJECT findings; it is not candidate acceptance. Its fixed source identity assertion cannot run against the changed candidate. The independent suite explicitly repeats those stage concerns against the pinned candidate's actual main() before testing broader concerns. All synthetic stage calls use a declared fixture-only digest seam; a separate unpatched production CLI call rejects that fixture with the original SHA256 mismatch and creates no payload or report. No original archive is opened, copied or extracted here.

Six independent test methods cover the two historical counterexamples, malformed executable and bundled DLL, unresolved non-system import, missing deterministic literal Tcl source, absent and preexisting report pairs, complete static closure with a valid bundled DLL and literal source, two successful generations, failure at each report replacement, and a failed second generation preserving the first payload and report pair. Success inventories are checked against actual fixture and extracted-file digests. Validation/publication failures assert causal errors, empty success stdout, unchanged old/absent pair, no new payload, and no leaked staging directory. A deliberate mutation allowing unresolved DLLs is rejected with specific failed rejection assertions.

Reuse the coordinator-verified repair receipt at scratch/usage-watch/tuxracer-closure-publication-repair-oct28/receipt.json: 15 repair groups and the affected original PE group pass; unchanged focused groups and original171/prior190 evidence are inherited without historical packaging replay. The three canonical dispatch input hashes were independently verified at their actual donor paths and again against the sparse checkout.

Acceptance covers recognized static direct imports, the documented lexical Tcl subset, and ordinary synchronous exceptions. Dynamic Tcl/loading, original archive acceptance, gameplay, concurrent-reader visibility and power-loss atomicity remain unproved. Serialized integration may consume the candidate plus this independent test/report commit. No main, registry or deployment operation is authorized by this acceptance.

Canonical independent receipt: scratch/usage-watch/tuxracer-bounded48-pair-accept-oct28/receipt.json. It records commands/exits, distinct stdout/stderr, source/fixture pins, fresh aggregate disk and physical-memory admission, strict-SSH task publication, exact remote verification, recovery and cleanup. A failed initial checkout caused by nested Git alternates is retained; only the owned clone's alternates were flattened to existing read-only stores before checkout succeeded. Initial read-only terminal inspection exceeded the intended output gate; it is not claimed as controlled acceptance. Behavioral/Git operations use the bound supervisor and retain its admission and process cleanup records.

The 48MiB task allocation includes session/events and positive WAL growth once, prospective growth and an internal 8MiB reserve. Donor stores are standing, read-only inputs. The oversized standing board is not appended by this worker; the coordinator publishes the bounded task-owned publication note. No test box is assigned; no runtime lifecycle change is made. Stop after this finite source-only handoff; original/gameplay qualification needs a separate task.