GLTron visible fullscreen bootstrap, 2026-11-14
The normal production fullscreen button now has a separate, reviewed host
bootstrap. Code ee9c89feb84901561d1999d9d5a9ba3fcdbaa8be is published on main.
The sole fresh original successfully entered actual browser fullscreen with one
trusted host click, removed the consent strip and reached an unobscured rendered
GLTron title/menu. Its first game-capture attempt then refused a new held paint
before ticket creation and closed. No game binding or guest key input occurred.
Two intact-cycle corners, a natural round outcome, human identity and native Quit
remain unproved. The closed original was not retried or rebound.
Evidence: scratch/runs/20261114-gltron-visible-fullscreen-bootstrap/result.json.
The fullscreen title image is a raw failure diagnostic, not an admitted game
canvas or a permanent native menu binding.
Host setup and the visibility guard
The derivative controller takes one raw browser screenshot of the actual production strip and records its browser PID/start token, target and page owner identities. Personal review names the visible “Use browser fullscreen” button, PNG hash and one-use nonce. Immediately before its single trusted DOWN, the controller rechecks the exact button object, label, geometry, hit target, viewport and browser process. The host menu observation must be younger than 30 seconds. An independent child is armed before DOWN with a 1,750 ms fallback; the intended host pulse is 120 ms. Cancellation, UNKNOWN or any failed attempt closes the bootstrap. No replay or obscured-canvas fallback is available.
The production button handler performs the Fullscreen API request through that
trusted click. The helper never calls that handler or requestFullscreen,
manipulates fullscreen classes/CSS, grants permissions or supplies an evaluate
user-gesture override. It requires the actual fullscreen element to be the same
#screen-wrap, the strip to be gone and the complete-output visibility guard to
pass. Ordinary resize/repaint readiness still precedes the first game ticket.
Trusted document mouse target observations distinguish the button edges from
canvas input. The synthetic control additionally observes the real renderer's
mouse entry points and confirms zero guest mouse calls.
Actual Chrome exposed a separate guard false positive: after normal fullscreen, ordinary titlebar/toolbar boxes behind the browser's fullscreen top layer were still being treated as potential overlays. The helper now scopes that scan to the actual fullscreen subtree. It globally refuses open modal dialogs and popovers, including their possible backdrops. Inside fullscreen it still refuses painted intersecting DOM boxes, including pointer-events:none overlays. Outside fullscreen all preceding conditions remain in force. Hidden, clipped or translucent output, owner/output replacement and cancelled tickets still refuse. The guard now reports its first failed clause and element box. This remains a conservative light-DOM check, not a general shadow-DOM/CSS raster certificate.
Only the derivative binding/controller and their existing test were changed. Runtime, index, page viewport, renderer and guest settings were untouched. No source interpreter optimization, forced repaint/hold release, solver, tracking, action loop, media download, box or deployment was performed.
Validation and qualifications
Final real Chrome/WebGL2 control uses the production index, viewport handler, compositor and actual trusted DOWN/independent UP with a synthetic producer:
| Check | Observed result |
|---|---|
| Actual production strip and complete button | Personally reviewed before one normal trusted click |
| Actual fullscreen / strip hidden / output visible | Passed; no game ticket during host setup |
| Exact f693 predecessor helper on the same fullscreen scene | Visibility false; corrected helper true, without ticket/repaint |
| Fullscreen capture vs actual Chrome compositor pixels | Exactly equal; original display and later read timestamps retained |
| Overlay inside fullscreen, pointer events disabled | Refused |
| Modal and popover above fullscreen | Both refused |
| Held startup and natural ordinary producer end | No new ticket/input while waiting; first new ticket unheld at frame 20 |
| Idle current display | 42 ms; publication 20 unchanged; exact compositor pixels |
| Known late hold | NOT_SENT before ticket; no forced release |
| Actual startup timeout | 128 ms with 125 ms control bound |
| Actual Chrome 3-second UNKNOWN | Closed at 3,250 ms; late response cannot reopen or dispatch |
| Owner replacement, cancellation, no replay | Retained focused/Chrome refusal controls pass |
| Real blank Chrome key and mouse input | Trusted 1,500 ms pulses and independent 1,750 ms EOF UP pass |
The synthetic strip sits above a complete canvas and does not itself cause occlusion refusal. This is recorded honestly; the control does not claim to reproduce the predecessor original's unknown evaluated clause. Its causal red/ green concerns fullscreen top-layer visibility. Synthetic gradient/purple scenes and earlier failed controls are not game proof.
Development controls retained: v1 asserted an occlusion that its geometry did
not have; v2 proved real fullscreen but refused the titlebar behind it; v3
revealed that bare :modal also matches Chrome's fullscreen element, corrected
to dialog:modal; v4's personal review expired with no DOWN; v5 passed before
the additional exact-Git causal assertion; final v6 passed that assertion too.
Every control is a separate synthetic process with pins, relevant Git binary
patches, actual native/watchdog/supervisor/outer waits and both EOFs. No failed
control or original was reopened.
Final focused controller/loopback/actual-child tests pass, including stale, changed, disabled and occluded host button checks and mouse scheduled/EOF UP. Five focused renderer files pass (20 entries, including 16 endpoint cases). The final-source locked full build passed in 31.28 seconds. Explicit own-path commit and strict pull/push completed before the original. All key/menu freshness, finite pulse/nonces/no replay and mechanical 180-second Quit reserve remain.
Sole original and exact missing proof
The original followed StarCraft then Pegged's actual native/watchdog/supervisor/ outer waits, both EOFs and heavy release, in the root-confirmed order. Its prearmed independent deadline was 600+20 seconds, with default 1,000 blocks per slice, normal real clock and all 92 existing media pins. Exact Git SHA plus relevant uncommitted binary patches and 677 source/media hashes are recorded; there is no source/media/WASM archive. Actual recovered module hashes are in the state and pin artifacts.
Before the host click, the exact first visibility refusal was touch-cursor,
an intersecting painted canvas. The reviewed raw bootstrap scene was black with
that visible cursor and the complete consent button. It was not called an
initialized native game menu. This new observation does not retrospectively
establish the closed predecessor's unevaluated clause inventory.
The sole click's mousedown, mouseup and click were all trusted and targeted the
consent button; no canvas mouse event was observed. Independent UP acknowledged
145 ms after DOWN dispatch. The same wrapper became document.fullscreenElement,
the strip was hidden, output visibility passed, and ordinary complete readiness
was recorded at publication 83, displayed at 21:59:46.400 UTC. The prebind
readiness record still reports zero guest input and no permanent binding.
A new held/in-flight paint arrived between that readiness observation and the
atomic pre-ticket gate. At 21:59:46.668 UTC the gate returned
CAPTURE_PRECONDITION_REFUSED, NOT_SENT. That path creates no capture ticket;
the chain closed and the controller performed host teardown. A subsequent raw
failure diagnostic personally shows the complete fullscreen GLTron title/menu,
with GAME/VIDEO/AUDIO/QUIT and no consent strip or touch cursor. It was never
promoted to a game-canvas capture or used to bind/input. There is no admitted
first game scene, known human identity, corner, natural result or native Quit.
Browser close is host teardown.
Native controller exit 1, Chrome exit 0, watchdog exit 0 (unfired), outer child exit 0 and outer caller exit 0 all have actual waits and both EOFs. Supervisor zero means completed custody, not game success. Original elapsed time was 28.594 seconds. The heavy lock is released, recorded runtime/browser PIDs are absent, HTTP port is closed and profile removed. All 677 pins, including all 92 media files, remain unchanged. All three closed predecessors retain their exact names, sizes and hashes. No ancillary board watcher was started by this worker.
A future distinct task can address the readiness-to-ticket race by passively
waiting on positively known pre-ticket busy state while preserving terminal
UNKNOWN/cancellation. This run stays closed; no second original was authorized
or attempted within it. Selected PNGs and their hashes are named in result.json;
unselected synthetic copies are inventoried separately. Box ID/expiry are null,
state NOT_CREATED, stop receipt N/A. No full-27, FPS, audio, parity, win or
physical-GPU claim is made. This bounded worker stops after publication.