Original GLTron complete boot and gameplay FPS — 2026-11-04
No paired gameplay FPS was measured. One software launch reached personally viewed gameplay, then its active-scene guard rejected marker arming. No marker was armed and no interval was collected. The pair stopped at this first fault; WebGL was not launched and software was not retried.
Evidence: scratch/runs/20261104-gltron-complete-boot-gameplay-fps/.
This run follows the repaired boot preflight in
0be601865fc463affafc11a2940c4cc94eb2332d. The earlier two pregame
dependency omissions were runner failures. Their original logs remain intact.
Offline prerequisites
The accepted 71-file evidence manifest passes every size and SHA-256 check
(7,691,474 bytes). All 92 original registered media files pass, including
the original executable SHA-256
85fb5184262490f0b5967c2b12ee27c77da058364e594b0fc86be7b5814e40ec.
No original media, guest instructions, configuration or pacing was changed.
The preflight reads the actual shell script manifest, font list, boot system
data list, app files and DLL seeds. It runs production resolveDllGraph with
the production import detector and DLL registry. The resolved real DLL graph
is zlib, libpng, sdl_sound, shared msvcrt and app-local SDL. A conservative
production lib/src/font/icon/compiler superset seals lazy workers, worklets and
fallback source compilation too. Both current compiled modules are pinned by existing
path and SHA-256, without copying them into evidence.
Every one of the 365 prior served HTTP records matches the current path, size
and hash. All 1,010 offline HTTP route resolutions pass with the actual static
resolver and /binaries/ rewrite. This check starts neither a server nor Chrome.
Discriminating tests remove or stale each required production pin, reject
unknown historical paths, wrong hashes/sizes, missing routes and unpinned
responses. The exact original dt branch runs under all four chaining/uop
settings in both modules; existing marker/uop-boundary tests pass 50 checks
per module. These fixtures are not gameplay measurements.
Runtime contract
Only the explicit mdk2-repaired-original LOCAL HEAVY RELEASE grants the local
slot, with Stronghold's 19:15:07 release also required. Historical MDK/Stronghold
releases grant no allowance. One external /tmp/wine-local-runtime.lock covers
the sequential pair. The fresh combined budget is 360 browser seconds and
180 cleanup seconds, fully reserved before hard Chrome 22:15Z / cleanup
22:18Z cutoffs. Bootstrap and ordinary command work are bounded at 60 seconds;
normal closure uses the separate cleanup reserve.
The production marker at 0x4051a9 counts entered nonzero-dt gameplay updates.
It does not count completed updates, physics substeps or presentation. Live
acceptance requires personally viewed gameplay before and after the interval,
main thread 1, the owning module generation and translated active callback
table, unchanged execution tiers and zero pacing. Raw counts and monotonic
read brackets are retained. Actual renderer attribution comes from owning
bridge/context/backend observations, never the requested query option.
The measured rate includes marker dispatch, tick reads, the enforced block/uop boundary, initial decoded-cache recycle and periodic observations. Read brackets bound observation timing; they do not estimate uninstrumented performance. Callback-table sampling cannot see a scene transition wholly within a guest slice. No overhead percentage is established.
Actual attempt and first fault
The required MDK2 release arrived at 19:32:34Z. The fresh pair held one local
runtime flock, starting only after that release and Stronghold's 19:15:07Z
release. Prelaunch source identity and binary patches are in runtime-pins.json
and relevant-source.patch; subsequent diagnostic edits are recorded separately.
The software arm served 234 requests, all matching path, size and hash, including
all 92 original media files. The actual owning page-main generation 1 loaded
canonical module SHA-256
78a0b1d61b23428a1ae909d739440ae0de059bbbd84a69a2ad3e4c37471ad664.
Main thread ID was 1, chaining 0, uop 1 and debug-tier guard 0. The owning
Bridge.local was OpenGLHostBridge, its context was software: true, and
software enabled was 1. No WebGL context was observed. The unrelated visible
D3DIM GPU dropdown is not the active GL renderer.
Personally viewed software/startup.png, main-menu.png, game-menu.png,
arena.png and gameplay-before.png. The first was a black loading canvas;
the next two showed the original menus. The arena capture showed GAME IS
PAUSED and callbacks 0x42a860. After the third balanced native Return action,
the 19:35:20.451Z gameplay capture showed the moving yellow cycle and a new
trail, with current callbacks 0x42a200. Six trusted keyboard events were
paired and targeted at screen. Frozen was false and host steps stayed zero.
At approximately 19:35:32Z, the sole measurement request failed inside
installPageCounter with not active gameCallbacks, before the marker setter
or observation timer ran. The scene seen in the reviewed capture no longer
satisfied the live arming guard. This is an observation/scene-lifetime failure,
not a missing boot asset or evidence of a production counter defect.
The original runner retained neither the rejected replacement callback value nor an image at rejection. Its cause (including whether this was a pause or round outcome) is unknown. No post-interval gameplay review or counter window exists. The live dependency still missing is an unchanged owning active gameplay scene at arming and throughout a personally reviewed >=5-second interval. No software rate, WebGL rate or backend speed comparison is claimed.
Offline repair and closure
The diagnostic now retains the exact rejected guest VA, translated offset and callback value, expected table, module generation, thread, game pointer, counter, marker, pacing, tiers and monotonic brackets before throwing. The runner saves this rejection and the owning renderer/context state before closure. Tests reproduce three review-to-arm scene mismatches (zero, the paused table and an adjacent foreign table), verify the exact rejected translation, and prove the setter never armed and the lifetime count did not change. This repairs missing diagnostic evidence; it does not claim to repair the actual scene transition or retrospectively recover its lost value.
The runner also requires the current named resource releases, verifies the production boot graph and historical HTTP inventory before Chrome, and uses one absolute cleanup deadline across pending launch, Chrome and server closure. Normal-exit, abnormal-exit and unresolved-exit fixtures distinguish successful closure from a signal or timeout. Final focused validation passes 46 checks, including both current modules' original branches under all four tier settings. The separate 50 marker/uop-boundary checks pass per module; original registry, syntax/diff checks and the locked full build pass.
Chrome PID 999258 actually exited 0 with signal null. Node PID 999234 was absent
after its wrapper exited 0; HTTP port 43117 refused connections, the server
reported listening: false, and the correct runtime flock was available.
Total Chrome time including closure was 69.015 seconds. Explicit local heavy
release was posted at 19:36:11Z. The owned browser profile was removed after
closure verification. No second launch, boat/API/SSH runtime, installs,
allocations, guest writes, configuration changes, set-count, host steps,
debug-tier disabling or screenshot-derived rates were used.
Box ID/state/expiry: N/A. Coordinator owns future boxes and shutdown. Git plus relevant binary diffs and owned new-file patches is the source archive; media and modules are referenced by existing path and hash. No source/media/ module copies, tarballs or snapshots were made. Stop here: any further original measurement requires separate authorization and a fresh valid scene review.