Fresh UT memory Setup after prepared release helper

Evidence: scratch/runs/20261118-ut-memory-install-after-prepared-helper/. This single fresh original did not complete installation. Shared disk preemption stopped it after three ordinary Setup inputs, before Install. Finish/Run, installed-child gameplay and ordinary Quit remain unproved. Durable installation remains OPEN separately.

The actual original launch was 2026-11-18 11:40:53.646 UTC, with installation 1200 / total 2400 / final reserve 300 / cleanup 30 seconds fixed before launch. It used the unchanged CD1 ISO, default D: root, fresh memory C:, and unchanged default C:\UnrealTournament, with --spawn-processes --child-transport=memory. Neither old partial overlay was attached. Media preparation and preservation hashing preceded launch. The existing preload reconstructs the pinned Git graph in process startup; no new compilation or qualification was performed.

Runtime graph cc007fc738c2932bd5219de16fd26cbbad51b6df, clean module SHA-256 a1b157f7d9e800ac8739ddb0a31755f2738bc080d8f31146b2a31562663a8a64, and the three published helper files from ae551e2fa were pinned and verified. Shared held runtime changes were excluded by the existing immutable preload. Optional executing-child debugger attestation was not armed; no installed child was launched. No manual extraction, VFS/media export, config/registry/save/clock edit, callback injection or fabricated acknowledgement was used.

Welcome Next, I Agree and default-destination Next each returned an actual CONSUMED native dialog receipt and matching independent physical UP. Arms took 61 / 67 / 93 ms; confirmations about 132 / 106 / 110 ms; physical UP completed within 678 ms. The original one-second ARM, two-second confirmation and 5.5-second physical-UP bounds were preserved. READY remained unarmed. The selected whole frames show Welcome, license and default destination. The file named components.png still shows the destination page; it is not visual proof of the components page. No installation percentage or installed path-size-hash manifest was obtained in this scope. Native state fields and visually reviewed percentages are not interchangeable.

The independent 100 ms disk guard used a preselected 1,540,000,000-byte preemption threshold. Native cancellation records SIGTERM at 11:41:55.425; native exit zero, supervisor SIGTERM and owner exit one have actual waits with both output EOFs. The supervisor did not produce its containment receipt. The failed later unused standby was internally awaited before throwing, but its PID and EOF receipt were not durably retained. All seven recorded runtime PIDs were subsequently absent; this does not replace the missing receipts. No endpoint or replay followed the permanently closed scope.

Free disk subsequently sampled 1,123,250,176 bytes, below the owner's 1.5 GB floor, then recovered above the floor without this worker deleting foreign files. The final sealed result records the independent guard's actual minimum through drains and the overlapping seal sample. The burst writer is UNKNOWN. This native-only task created no browser profile or raw browser captures; the tmpfs worker was notified on the board. No UT engine defect is demonstrated.

The guard's retrieved minimum was 1,081,348,096 bytes. Its child wrote the summary and was absent at audit, but the enclosing tool exited 143 without a retained actual-parent wait or EOF receipt. The first seal attempt refused to publish because that receipt was missing. Board-watcher wait and both EOFs are retained. A fresh finite sampler covers final sealing after a short offline monitoring gap; no parent wait or EOF is fabricated.

Two routine local issues were handled offline. An owner wrapper started before immutable preparation finished and failed for missing module.json, before any native process or runtime clock existed; that error and actual wait are retained separately. After preparation completed, exactly one native original launched. After closure, an EOF bookkeeping error attempted to save the already retained unused-helper wait twice. The run-local controller now clears the retired handle before awaiting its replacement; syntax validation passed. No production code changed and the closed original was not rerun.

The ISO and both old partial overlays passed post-run size/inode/hash audits. Recovery uses Git plus the retained relevant binary diff and small unique run controls; no source-tree snapshots or media copies were created. No agents, workers, boxes, Python, Wine, deployment or held-title actions were used. Box ID, expiry and stop receipt are null; state is NOT_CREATED.