ROCKROLL current published graph: natural WebGL motion, 2026-11-18
Evidence: scratch/runs/20261118-rockroll-current-webgl-natural-motion/.
One fresh registered scr_rockroll /s original produced a reviewed whole
1280×960 scene pair on current published main
5732320bb39c4b7f7327f51e30c6ed1d07e26074. The compiled module SHA-256 is
9862d12476ad04bd266de2e729dd7076f34951a5d1f85eaaf0843fda5c99cabd.
This is a new current-graph task, separate from the older 3ab982f2d /
06eaa397… results. No closed lifetime was queried or replayed. The later
historical WebGL pair in 20261118-dx-wormhole-scifi-paired-motion remains
historical evidence rather than evidence for this graph.
original-webgl/natural-before.png shows a large, mostly edge-on guitar ring
extending beyond the window, with dark distributed picks and a small central
star against the sunset gradient. natural-after.png shows a smaller, tilted
ring of face-on guitars, changed neck/body positions, and gold picks in changed
poses and locations. The author and coordinator reviewed both complete images. There are
403,241 changed whole-frame pixels; the visible geometry establishes substantial
natural scene motion rather than merely a cursor, overlay or counter change.
There was no guest input during the original.
Actual render-Worker GpuStreamWebGL completed DRAW_UP counts rose 3,939→8,371
and completed PRESENT counts 101→289. These are the last complete PRESENT
statistics sampled after each screenshot, not FPS or precisely simultaneous
pixel/command measurements. The executor reports webgl1, WebGL 1.0 Chromium,
and ANGLE (Intel, Mesa Intel(R) UHD Graphics 620 (KBL GT2), OpenGL 4.6 (Core Profile) Mesa 23.2.1). The launch's requested SwiftShader flags are not
substituted for that actual context identity. Software executor counts remain
empty. Native d3drm.dll and oleaut32.dll load/import receipts and all HTTP
response hashes are retained. Actual page, guest-Worker and render-Worker
module bytecode inventories all match the new module hash.
Preparation and finite controls
Source came from an immutable in-memory Git map; no source tree was copied.
Foreign worktree WIP was excluded and its binary diff recorded separately.
Eleven ordinary registered media paths and hashes were prepared before the
original timer; none was copied or changed. The single tested module is retained
once. Existing rm-startup-run.cjs and retained JavaScript durable supervision
were reused without modifying production code or capture helpers.
The existing authored 200-million-iteration Device2 workload reached the unchanged 15-second admission refusal and closed UNKNOWN with no later endpoint query. A separate 50-million-iteration positive control, matching the prior calibration, admitted in 4.508 seconds and produced the expected reviewed whole blue/white image, actual WebGL draws/presents and three matching module receipts. The existing deferred-inventory, error cleanup, separate presentation/composition and late-capture closure checks passed.
The original admitted in 3.405 seconds. Startup and motion ran without the V8
debugger; actual module inventory followed the pair, preserving repair
a373de9ba. The secondary inspector used defaultViewport:null. Ownership,
generation, 300 ms stable geometry, completed native presentation and composition
checks passed. Both captures preserved the exact baseline geometry. Deadlines
were declared before launch: 300-second initial startup, 15-second admission,
three-second whole-frame capture, 15-second delays before each selected frame,
390-second enclosing runtime and 410-second independent cleanup deadline.
No deadline was extended in flight.
prepared.json collected its source hash list after compilation. Verification
identified the compiler's existing Buffer-slice comment normalization mutating
that in-memory map. source-pin-correction.json reproduces every changed hash
and preserves canonical Git hashes; the original's pins-before.json records
the unmodified canonical served source. This is an evidence-metadata correction,
not a runtime source change. Recovery uses the pinned Git closure and normal
compiler; no modified comment bytes or foreign WIP are required.
Lifecycle and limits
The original browser close was awaited with exit 0; native command and independent watchdog exit 0, separate actual stdout/stderr EOFs, no supervisor termination, no watchdog firing, no poll error and no surviving tracked processes are verified. The earlier external-SIGTERM sender remains unknown; no external TERM recurred here. Browser teardown is harness cleanup, not ordinary guest Quit or guest ExitProcess; native saver quit support was not exercised by this helper.
No rendering/startup crash was encountered and no production fix was needed. The locked shared-tree build was attempted after the original and refused the concurrently owned UT CreateDialogParamA/CreateWindowExA/show_window_core callback producer hashes. That foreign WIP is absent from the tested production graph; its owners were already reconciling the gate. No gate or foreign file was changed here. This report-only publication does not claim a passing shared WIP build or a production code repair.
Recovery recipes, exact Git, excluded binary diff, passive observer suffixes,
module/media hashes, owned report patch and lifecycle verification are in the
run. result.json is written last. Home remained above 1.5 GB. Box ID null,
state NOT_CREATED, expiry null, stop receipt null; no workers, agents, boxes,
Python, Wine or public deployment were used. Backend parity, complete27 census,
FPS, audio, full saver cycles and Windows reference correctness remain open.