Software Tunnel and fresh software Flip3DTL — 17 November 2026
Both authorized fresh originals have reviewed moving 3D scenes, actual software GPS executor observations, and complete enclosing lifecycle evidence. This adds software Tunnel to the seven previously observed SDK scenes and supplies a fresh software Flip3DTL run with durable waits. The earlier Tunnel capture refusal and Flip3DTL missing waits remain unchanged in 20261117-dxsdk-four-examples-paired-rendering; neither old endpoint was queried or replayed.
Evidence: scratch/runs/20261117-dxsdk-software-tunnel-flip-rendering-finish/result.json.
| Fresh original | Personally reviewed early/later observations | Actual render Worker observations |
|---|---|---|
| Tunnel/software | Complete ordinary window, gray textured receding tunnel rings, yellow/black hazard signs changing position and orientation. | GpuStreamSoftware, completed PRESENT snapshots: draws 117 → 231, presents 39 → 77. |
| Flip3DTL/software | Complete Windows 95 textured cube rotates, changing silhouette and visible colored faces. | GpuStreamSoftware: DRAW_UP observations 696 → 1200, READBACK 116 → 200; CPU publication rather than PRESENT. |
Four selected whole-page PNGs retain the application content and surrounding page without host cropping. Tunnel's native window remains at (20,20), 320×320; Flip3DTL's native window remains at (0,0), 640×480. Both whole-page outputs are 1280×960. Geometry naturally meeting its own client edge is not a cropped host image. Complete image comparisons find 82,262 changed pixels for Tunnel over 17,003 ms and 175,014 for Flip3DTL over 17,217 ms. These differences support the reviewed motion; they are not cross-backend parity or FPS measurements.
The executed graph remains Git 70a680b54bf1f4f62243f6218314cb45610a1bd2, with the retained module SHA-256 cf674e8d1780249498a3b9797142d5d9f01e9daedcf2089c8fc880f515c753c0. All six debugger-observed page/guest/render Worker module receipts match it. Each fresh run's 1,503 source pins and 38 registered media pins exactly match its prior case. Media hashes remain unchanged; all 402 successful HTTP response hashes match the served graph. Source is served from Git into memory, excluding shared WIP. The module and media are referenced at their existing paths without copies or a new compilation. Chrome is 151.0.7922.108; guest endpoint 1 and render endpoint 1 identify d3dim-gps/software.
The existing qualified whole-page recipe is selected before each original. It preserves identity checks, native window/client mapping, display guards, and the fixed 3000 ms capture deadline. It does not claim immutable framebuffer publication. Retained authored Device2 controls cover all 307,200 source and 1,185,280 output pixels, plus eleven ownership/refusal controls on each backend. The prior windowed native GetPixel and whole-page capture control completed in 681 ms. Their exact evidence hashes are rechecked here. The focused test/test-boids-current-rendering.cjs tests pass for passive observations, literal pixels, client composition, source/owner replacement, lost/closed endpoints, deadlines and sticky UNKNOWN refusal with tracing off/on.
Only the two authorized local originals run, serially, with no physical inputs, forced guest calls, pumps, media/configuration/clock changes, additional originals, boxes or workers. Each uses prearmed independent preparation/runtime watchdogs (260/320 seconds), supervisor limits (240/300 seconds), and the existing detached durable launcher with a 600-second outer deadline and subreaper custody. Both native processes, Chromium closes, preparation/runtime watchdogs, enclosing commands and independent enclosing watchdogs have actual successful waits and separate pipe EOF evidence. Thirty adopted browser/helper children have actual waitpid records; some terminate with SIGTERM during owned host shutdown, explicitly retained rather than called native guest Quit. Both subreaper scopes end empty, known runtime PIDs are absent, and both listeners refuse connections. No deadline fires.
No native/render/lifecycle product defect is demonstrated and no production repair or canonical repair build is claimed. Changes are evidence and this report only. Exact current Git, excluded relevant shared diffs, owned run-local file patches, selected whole screenshots and artifact hash readbacks are retained; no source archive, media copy or frame series is created. /home remains above 1,500,000,000 available bytes; there is no RAM gate.
The earlier six complete app/backend cases and seven observed scenes are preserved. Across the prior evidence and these fresh cases, all eight named app/backend combinations now have moving-scene and enclosing lifecycle evidence, with the original failed/unknown records kept separately. This does not assert broader scene correctness, parity, audio, FPS, native Quit, the startup matrix, or full27 membership. No further original is needed for this task. Box ID null, state NOT_CREATED, expiry null, stop receipt N/A. Release and stop.