Recoil current Software Resume baseline, 2026-11-08
The one fresh original reached a personally reviewed current native Resume in the actual Software route. Its first Software baseline then refused stale evidence before Resume was clicked. No Resume, M or movement was dispatched. The concrete blocker is the original baseline's complete uncached guest WASM transport: 442 ms before durable retention, 547 ms through its completed proof, against the unchanged 250 ms limit. No replay or speculative repair followed.
Evidence R: scratch/runs/20261108-recoil-original-1702/.
Preparation Git SHA: 7f88f84951edf7ea5a308683da276a36986bcafd.
R/source-head.txt, R/relevant-code.patch and the small owned operational/input
patches record recovery; existing media remains referenced by path and SHA-256.
No source archive, repeated media bundle or peer WIP was collected.
Normal generation selects the preceding
20261108-recoil-software-ready-1650/repaired-ready-runtime-pins.json and adds
29 relocated physical helpers: 1,066 providers. The prior 1,037 rows and original
851 remain unchanged. The physical shared kernel is exactly published
2486507db; bounded endpoint remains 11683651 and persistent producer 68ea2f56.
Decoder remains 204938 bytes / f468ffce; compositor 227919 bytes / cb21daaf.
Cheap production resolver, nested closure and hermetic preloader passed before
allocation. Actual provisioning joined 1,067 uploads, content-verified every
provider, published the barrier, passed queue and parent admission, and
atomically consumed the sole original request. Previously accepted 2,894
references / 762,693,392 bytes were independently rehashed in place, without
rerunning unchanged fixtures.
The independent persistent producer was created before Chrome: PID6656/start 12752 versus Chrome6689/start13284, driver6645/start12523. Startup took 4.944 s; the first fresh challenged census took 3.211 ms. The proof binds the actual boot, net/pid/mnt/time namespaces, exact profile/DevTools endpoint and Chrome FD83 on port43667, plus driver FD22 on port39873. The challenged pre-baseline census also passed in 3.020 ms. This is current process proof, not Software admission.
Current native observations were personally reviewed:
- 17:04:44 Software radio black, Accelerator white, unobscured before native OK.
- Native Game > Start, MAIN MENU > NEW GAME, then unchanged
userand Medium. Pointer lock required reviewed relative movement. An initial absolute NEW GAME click left the menu displayed; this known positioning behavior was retained, not described as a new defect. The reviewed START was activated with an independently released 600 ms native left-button hold. - 17:08:49 Island Storm loading; 17:09:15 native briefing; 17:09:37 opening
camera; 17:09:58 transport/disembark camera. The capture named
hud-currentis opening footage without a HUD, despite its filename. - 17:10:20 living tank HUD, actual displayed health 100, timer 00:00:12, landing ramp ahead. No gameplay movement was sent. These values are current observations, not the historical Software196457/health100/timer09 image.
- 17:10:37 native Escape500 opened IN GAME MENU. 17:11:03 relative positioning
put the visible guest arrow at459,249 inside highlighted Resume380,247–504,279.
After installing the actual Software owner gate, the new 17:11:11
resume-armedcapture was personally reviewed again. PNG SHA-256:c27474f6f4f921deb70e6f325d98d54fd76738f0bb6dfda7021921b2b7ce16fc.
The sole qualified-map request entered its beforeResume baseline at
17:11:59.417. The read-only observer was prearmed for a possible M consequence;
it supplied no input. All three first-read response lanes completed and drained
through Promise.allSettled. Actual decoder and compositor sources matched the
pins in current page context1. Guest target C3A1919A47DCA5363C5F3B43293BEAC4,
script23, endpoint1/slot0, supplied all 2,182,663 WASM bytes with exact
6cf120d0 SHA-256. The matching complete response envelopes are retained as
source-response57/58/59, not substituted with hashes or cached content.
Guest source request began 17:11:59.506; response reached the driver
17:11:59.948 (442 ms); the retained, hashed and instance-rechecked guest proof
completed 17:12:00.053 (547 ms). Decoder proof took104 ms; compositor247 ms.
The baseline refused Software evidence expired (<=250ms required) at
17:12:00.159. Only the entry phase exists: the first owner read did not become a
certificate, the second read was never started, and the required six-response
baseline was not achieved. Earlier ordinary guest-only reads were 88–149 ms;
the record does not isolate why concurrent Software qualification was slower.
It does not justify changing clocks, deadlines, proof timestamps, response
completeness, or supervisor scope. No causal code repair is claimed.
Resume's mouse DOWN occurs after the failed baseline in compound-v4, so zero Resume/M/C/arrow DOWNs were issued by this command. Numbered transport retirement is sticky UNKNOWN. There is no argument-free post-Resume proof, changed-timer living pair, qualified Software before/after, M consequence or bounded movement. The five-second window, <=250 ms final PNG through decode/save, all six complete uncached responses, current contexts/pipeline owners and retirement checks were preserved. Loading, opening, menu and active HUD observations are separate; no exact simulation elapsed time is inferred from wall intervals.
Normal shared-kernel drain passed: exit0, all552 owned tasks reaped, no remaining
tasks, independently all553 recorded task/supervisor PID/start identities absent.
Adopted descendants, both ports, profile, runtime cmdline/FD writers and remote
runtime lock drained. No held gameplay input remained. Chrome's ordinary close
needed the existing fallback and ended SIGKILL/codeNULL; this is not claimed
as Chrome exit0. Kernel drain is separately exit0. One unresolved autoattach
lineage PID6960/start16859 remains explicit, with
kernelForkLineageComplete:false, untracedCloneFlagsAudited:false;
CLONE_UNTRACED and nonleader-exec limits remain unresolved.
All283 remote output references /445,105,416 bytes were retrieved or referenced in place and independently reopened/hash-verified before delegated normal API stop. Accepted identical modules and already fetched captures were reused. The box is bx_q4mxe3ax, ARCHIVED, current expiryNULL, initial expiry 2026-11-08T17:28:00.591Z. Stop receipt stop_a220b23a2d53f53e completed 2026-11-08T17:13:11.382Z, forcedFALSE/errorNULL. Independent observable 25-minute lifecycle roles were armed before the UUID-idempotent allocation; ID/expiry were durable before runtime. Browser deadline17:25:30.362 and outer stop deadline17:27:00.362 were immutable. Remote lock covered startup through owned drain. The local heavy-runtime lock was never needed or held. Lifecycle roles exited0 and supervisor recorded no remaining children.
This docs-only result records actual native progress and the first concrete baseline blocker; no production code or pins were changed after the attempted generation. Next work is to isolate the measured current uncached-transport latency before any separately authorized fresh original. No old resume/delete, public deployment, private guest writes, config/clock/seed changes, replay or subworkers were used. Full27, parity, audio and FPS remain open. RELEASE and stop.