GLTron current GPS browser input prerequisite, 2026-11-17
Evidence: scratch/runs/20261117-gltron-current-gps-native-browser-gameplay/result.json.
The first authored browser qualification stops before DOWN: the actual browser
host import scope has no guarded input authority and no native input actor
resolver. Existing input_consumed and check_input_token imports alone do
not establish a functioning native receipt path. CLI Get/Peek actor proofs
remain distinct from a browser-to-owning-guest qualification. No CDP return,
physical handler, capability flag or fixture image is promoted to native
consumption. No original is launched; moving trails, two turns, a round outcome
and native Quit remain unproved. Historical27 membership remains unknown.
Two helper defects are repaired without changing the renderer or guest:
- The observer now selects the current executor with
?gl-gps=softwareor?gl-gps=webgl. Executing the actual published selector shows that the old?gl-renderer=softwarequery defaults to WebGL. Actual GPS endpoint options, device counters and render Worker observations are retained separately from the requested arm. No legacy GL renderer is restored. - The complete-capture transaction records request, ticket, host and terminal phases locally. Its single 3,000 ms timer closes transport permanently. Controlled request/ticket/host stalls prove that late completion causes no dependent request or retained late pixels. Failure diagnostics never inspect the endpoint after UNKNOWN. Successful observations retain the complete source/output/host triplet and initialized GPS publication join.
An opt-in qualify-native authored mode passively retains the actual import
scopes created during launch and refuses a missing native authority before
input. It is a diagnostic prerequisite check, not a new input API. Capability
presence also cannot qualify delivery. The live mismatch, actor, focus, modal,
generation, late and UNKNOWN input matrix cannot proceed through an absent
browser authority; the offline checks do not substitute for that matrix.
The development WebGL fixture uses Git
e71caed166e49962b6fc9f893a5516ecdef10ebc plus the recorded helper patches.
It retains module
cdcbd50d245a9b3f6ff8c8ad9b61b52e24c9cf286c1f80dd045016cdbe42dd49
before clocks and checks both actual Worker bytecodes. The complete capture
takes 411 ms. Personal review of all three images shows the yellow triangle
on blue. Actual endpoint is gl-gps, WebGL, endpoint 1/epoch 1, device backend
webgl1. Its ANGLE Intel/Mesa string is browser API identity, not independent
physical-device attribution. Native route observation refuses with zero input.
Launch uses the unchanged published dedicated waiting owner, pinned to the actual task worker, and the GL process-parent supervisor. The owner survives the transient launcher; its independent deadline watchdog is armed before the outer, whose preparation/runtime watchdogs precede native execution. Actual native, browser, outer and watchdog waits and both EOFs are retained. Browser termination is host cleanup, never native Quit. Unchanged lifecycle proofs are reused by exact component identity, without claiming the changed full runtime module is the historical tested module.
The focused helper tests and canonical locked build pass. Published arm measurements and their exact source identity are recorded in the final evidence. The next required implementation is an actual browser native receipt bridge, followed by the same-path positive/negative input qualification and only then the authorized original arms. No boxes, other workers or public deployment.
Published measurements
Both fresh authored arms serve published Git
2380bf5f6f94701388f8a2b9d7ff831dd683086d, with every helper pin matching.
Both retain the same cdcbd50d…be42dd49 module before clocks and verify the
actual guest and render Worker modules. No foreign shared-tree WIP is served.
The software arm observes gl-gps, endpoint 1/epoch 1, both configured and
endpoint backend software, real software executor native-draw counters and
no observed WebGL contexts. Its complete capture finishes in 1,004 ms.
The independent literal source and output oracles both have zero pixel
mismatches. All three complete images were personally reviewed. It then
refuses the missing native authority before DOWN.
The WebGL arm observes gl-gps, endpoint 1/epoch 1, backend webgl, actual
render Worker WebGL context and webgl1 executor counters. Its complete-capture
transaction reaches permanent UNKNOWN in 3,001 ms, in read-ticket.
Several earlier read evaluations returned pending; the final evaluation is
still outstanding when transport closes. No host screenshot request is
sent for this observation. There is no fresh complete triplet to review or
substitute from the candidate. The repaired phase log distinguishes this
boundary from host screenshot completion; it does not establish the renderer
cause of the pending ticket or repair the intermittent timeout. Its earlier
passive native-route observation independently records the missing authority.
Both published arms have zero DOWNs, zero later CDP requests/HTTP after closure, no rebind, replay or retry. Actual native harness and outer exits are 1; preparation/runtime/owner watchdog exits are 0, and browser exits are 0. Native, browser, outer and watchdog actual waits and both EOFs are preserved. The dedicated owners remain alive through their outer joins with the actual worker alive. Owned browser scopes and profiles are absent after cleanup. These are host-cleanup outcomes, not native Quit.
The selector and diagnostic repairs are tested and published. The browser native receipt bridge and its physical-release/fault qualification remain the concrete input blocker. Original allowances remain unspent for both backends; there is no original gameplay or round evidence in this run.