GLTron prepared-controller attempt refused before allocation

Two reviewed ordinary player turns and a natural round outcome remain unproved. The initial actual df -B1 /home observation returned 1,493,991,424 bytes available, below the required exclusive floor of 1,500,000,000 bytes. This attempt stopped before allocation. There was no holding loop, retry, browser, native input, or newly demonstrated controller/route/runtime blocker.

Evidence: scratch/runs/20261112-gltron-final-two-turn-floor-0045/. The exact current Git SHA is recorded in source-commit.txt; the relevant tracked-source binary diff is empty. No runtime source or helpers changed. Git and the retained parent helper patch remain the source archive; no source tree or media archive was created.

Read-only SHA-256 and size checks passed for all 58 sealed artifacts from the parent run, all 737 current closure inputs, and all 113 media-reference entries (112 retained media pins plus the unchanged calc fixture). The original GLTron payload remains the parent's 92 media files. The prepared controller, build modules, focused caller/ChromeWorker evidence, complete 738-file remote provisioning receipts, and independent release/UNKNOWN rules are retained by exact hash. These checks reuse existing qualification; they do not qualify original gameplay. The 17-path matrix, old latency preflight, build, and browser tests were not rerun.

The production provider, provider tests, and custody helper match the exact published bytes at 5c2c410a5; final publication c1edb0402 was read. The provider was not invoked. No speculative compatibility instrumentation was added. The closed-cache worker's files, caches, reserve, vendor files, databases, and sessions were untouched.

New box ID: none. State: NOT_CREATED. Expiry: null. Stop receipt: not applicable. No lifecycle watchdog or lifecycle child was started because allocation was refused. Parent box bx_5z3qkpz6 was not resumed; its retained, hash-verified receipt remains ARCHIVED with completed normal stop stop_430a5e0c662fcc97 and null expiry. This is retained evidence, not a fresh remote state observation.

Next requires actual capacity above the floor, followed by one fresh private box with an independently observable deadline watchdog armed before creation, joined complete provisioning and SHA verification before the browser, then fresh full scenes before and after each reviewed ordinary turn and the native outcome. Historical one-corner or motion/winner captures remain insufficient. No physical GPU FPS, audio/backend parity, or full 27-game claim is made.