MDK2 honest menu review and Medium release reserve

Input is 00397b3a7a9d6f8cc9c845db4d72a318a7fcf8fa and mdk2-measured-native-boundary-gameplay-20261104.md. Evidence is scratch/runs/20261104-mdk2-honest-menu-medium-gameplay/ (RUN). One original attempt reached a personally reviewed New Game selection and the difficulty dialog with Medium highlighted. Medium's press and keyup were acknowledged, but the selector rejected its required rest before submission. Medium's native release consumption/callback, loading and gameplay remain unobserved. No second actual attempt followed the offline repair.

Before Chrome, 41 local checks, two generated-controller checks, the full locked build and the exact fresh-directory entry passed. The entry used an explicit plain browser mock and independent Chrome-spawn rejection, spawned zero Chrome and closed both HTTP servers. Exact intended next.py initial and next.py startup calls traversed the generated Python local transport and real ACK client: named startup-0, eight 64-step ACKs, named startup-512. The intended open/New Game/Medium field sets traversed the original author, operator, next, input preflight, approval validator and local transport with only the browser-facing selector boundary mocked. These were explicit offline fixtures, not visual reviews or historical screenshot rehearsal.

The assignment and release parser now bind this task to GLtron's actual gltron-low-latency-webgl-gameplay-nov04 heavy release at 21:41:19Z. Actual launch used that real board line, not a fixture release. Browser/bootstrap/ command/cleanup bounds remain 720/60/40/180 seconds, with absolute 23:50Z/ 23:53Z cutoffs and a long-lived external /tmp/wine-local-runtime.lock owner. Box ID, state and expiry are NA; the coordinator retains shutdown ownership.

Each fresh startup/title/menu image was personally viewed. At 3072 steps the title had legal text and no controller. Ordinary Escape passed its original native positive/rest readbacks. Captures 55/63 showed the opening with legal text; 66 no longer showed legal text; 69 showed emerging dim menu icons. Capture 72 at 3184 steps already showed a red highlighted controller and readable New Game. Its review correctly used main-menu / New Game, and its original native readiness check passed. The next capture, 75 at 3248 steps, showed the difficulty dialog with orange highlighted Medium. Its separate review correctly used difficulty / Medium; preapproval and native Enter release checks passed. No preceding visual classification was reused without reassessment. An early image-read request preceded completion of capture 2560; it issued no runtime command, and that capture was viewed before continuing.

The Medium press required twenty 16-step poll quanta before native Enter raw/ previous became one. The original menu was still visible in those native state readbacks. The aggregate selector ledger reached 496 steps: 176 preceding menu steps plus 320 press-poll steps. Keyup command 127 succeeded and balanced the trusted DOM event history. The first required 32-step rest would reach 528, so the selector's original 512-step guard rejected it before queueing. This is the first actual fault. There was no failed guest step, native trap or observer exhaustion. The keyup ACK does not prove native release consumption.

The continuous native observer retained twenty checkpoints, 346000 boundaries and 21 sample records, all with the same fixed deadline and no fatal report. The final terminal Worker report retains that history with terminal-cleanup status. There are no demand/finalizer/consume/poll-return/script-buffer/LuaLevel construction records in this actual history. Accepted prior native decoder and construction proofs remain separately identified as offline evidence. The 50-million-boundary, 360-second, 512-record contract and byte/nesting/module/ thread/owner/input/ACK/frozen-state guards were not changed or reset.

The causal repair is offline only. The selector retains its 512-step press ceiling and gains a fixed 32-step tail after Medium's acknowledged, balanced keyup, capped at 544 aggregate steps. Other phases cannot use the tail, and Medium's existing one-shot ledger prevents renewal. Native readiness, callback, release, input, ACK, 113-command and 240-second menu checks remain. The accepted 16ms adaptation is now part of the generated controller, so the composed entry tests the same selector route; it removes only the prior host tick20 operation. Guest instructions, configuration and pacing were not written.

The regression executes the actual selector over this run's command/native prefix with explicit plain fixture inspection bytes; it never reads or rehearses the original PNG. The old selector rejects before the required rest, while the repair reaches that exact 32-step submission. A sentinel stops the offline boundary before claiming any subsequent native execution. Additional cases retain the press ceiling and exhaust the fixed 544-step tail using explicitly hypothetical unsettled readbacks. These do not establish that the real game would finish its callback within the tail. The repaired controller also passes the fresh complete-entry zero-Chrome test. Final focused validation is 42 local checks and the full integration suite, plus the required locked full build.

Quit command 128 requested normal closure. Chrome PID 1224193/start ticks 187625254 exited 0 with null signal at 21:54:58.661Z. Node PID 1224055 and external flock PID 1224054, both start ticks 187625048, exited 0 and are absent. HTTP ports 38311/35881 and DevTools 34383 refused connections; the exact profile was removed after process absence and the external flock was verified free. The assignment-bound heavy release was posted at 21:55:49Z; Stronghold took the next local assignment. No foreign process was signalled or box stopped.

RUN records the exact launch Git, relevant launch binary diff, final binary diff and new-note patch. Original runtime Git remains 1329b1ba6eb00144ff77d4cd92a8513d4fc953ad; both module hashes remain cffca869b541e6eff12ae3a6b74c4476977e0a7e23cd43d8ec5ba8faecd4bcee and 580f49f3c78d26e6855efa0beb37046d03ea24690ff98cfcb3e94ce5e6cb1ee3. The preceding run's 3213 sealed files / 36472345 bytes were verified unchanged. Existing media/modules are referenced by path and SHA; no source snapshot, tar, media/module copy, npm install, boat or subworker was used.

Any further original attempt requires a separate assignment and authorization. The immediate unknown is Medium's native release/callback after the newly admitted rest, then loading and visible gameplay. This assignment stops after tested repair, publication and path release.