Conan reviewed-command gameplay investigation, 2026-11-06

The predecessor's missing reviewRef defect is corrected in the run-owned serialized command consumer. The single original successfully received ordinary reviewed New Game and Warrior inputs and reached Kordava loading. Player control and the actual failing first-draw owner were not established within the authorized window. Production emulator source, installed guest bytes, texture-residency repair 72768dcf8, and the strict singular-normal guard remain unchanged.

Evidence: scratch/runs/20261106-conan-reviewed-first-draw-gameplay/. Git is the source archive: source commit a58a3de6fad71f6104a20c84aaf9602594a30a61, relevant binary diffs, owned helper new-file patches and actual tested modules are retained. The genuine 37-file installation is referenced in place by path and SHA-256; no source archive or media copy was made. The source inventory includes the tracked shell manifest and fonts as well as the CPU, host and renderer closure.

The actual consumer parses the serialized JSON used by the original command queue. Before input it requires the latest capture label, unchanged PNG hash, personal review with matching session/hash, a live verified owner/module, and unchanged captured owner/endpoint/Worker identity. It validates the complete input envelope before dispatch and always releases a pressed key. Recoverable command mistakes retain the same browser with their errors recorded.

Before the original, a real production shell launched a generated control PE. Its capture was personally inspected. The same consumer accepted reviewed key and click commands, producing four trusted browser events in balanced pairs. Missing/stale references, a closed session, changed endpoint identity and an unverified identity all rejected before additional input. The original checks the qualified consumer's exact SHA-256 before launch. No original command guard was relaxed or bypassed.

Current-source qualification passed the locked standard/compatibility build, 20 public producer cells, 16 owning-Worker SEH/pager cells, four lifetime cells, bounded transport/history controls, fixture ABI checks, 24 native multiply/compare cases and 12 native envelope cases. Both-module baseline/diagnostic comparisons preserve saved GPRs, EFLAGS, x87 environment/register images, thread state and outputs. Generated strict singular-normal failures remain enforced. Module import/export contracts are identical. Diagnostic SHA-256 values are bad3ff2d6b88f2f46040ecc16a569874ab5d542152ba52844155f2d52f490cda (standard) and 9d2e5215ac083a330520c68b935431210716afb8e39f48477aa25f43c80add5f (compatibility).

The observer retains the predecessor's 16 resident generations sharing 2,048 calls and 8,192 detail rows. Observed destruction/reconstruction/explicit eviction releases budgets. Missing constructors, evicted generations, incomplete native tier boundaries, missing backing/instruction maps, mismatched PC/SP/thread/parent returns and transport loss remain rejection conditions. A count of 103 or a reused address cannot establish the draw owner. A palette window establishes ordering only; actual matrix/store forwarding is required for draw contribution.

The coordinator granted 15:25:08.406582–16:10:08.406582 UTC, including cleanup, after Sparse 8783caf9d and AoW 1bf538bf4 FINAL releases. Own read-only cleanup verified the three named predecessors' 25/7/43 absent PIDs, their recorded ports and profiles, four free locks and more than 2 GB free before runtime. The owned lock holder acquired the three runtime locks at 15:26. No boxes were assigned or used; box ID/state/expiry are not applicable.

The original ran 15:30:32–16:07:24 UTC and closed normally at its immutable cleanup reserve, with Chrome exit 0. All four original Enter commands carried current personally inspected review references; no original command errors or relaunch occurred. New Game and Warrior were visibly selected. Every later capture, including last-before-close.png, remains Kordava Loading. Names such as scene-initialized are anticipated checkpoint labels, not scene evidence. No cinematic, player control, movement or distinct second gameplay response was established. This is a loading-window timeout, not a gameplay pass.

The final streaming audit contains 5,770 segments, 41,801 rows, 2,052 matched returns, zero open calls and zero history failures. Temporary constructor histories were retained with generation identity and matched destructor returns; there were no actual target palette links or zero-world draw. High water was 141 calls and 2,110 detail rows. No actual singular failure recurred, so no actual failing contribution can be related to a target lifetime. Native x87 controls alone do not demonstrate an emulator discrepancy; no speculative CPU, timer, Win32 or renderer change was made.

Final offline verification matched all 395 pinned source files, 249 complete served response/range identities, seven distinct installed-media paths, and four actual Worker module observations (CPU/render Workers, including the loading thread) to retained module bytes. Full served response hashes and module identity records are retained. Cleanup independently verifies all owned processes and profiles absent, ports refused, four locks free and more than 2 GB free. The holder receipt records signal 15 at 15:47:28 UTC; consequently continuous runtime-lock ownership through the later original interval is not claimed. No other heavy runtime was observed at final cleanup.

The canonical candidate manifest contains no Conan candidate ID. The genuine local conan_original_local registration is retained; result.json uses null rather than inventing an unrelated canonical ID, matching the predecessor's explicit registration caveat. Next work requires a separately authorized window long enough to cross the loading boundary, retaining these reviewed commands and bounded owner histories. This run does not justify an emulator repair.