Q3 readiness delivery observed; visual continuation stopped

Instrumentation is published on main as bfe8fc1746fa43071dc953d27bc3f51459ea8c7c. Evidence: scratch/runs/20261109-q3-readiness-wire-cause-2335/. The original pending request 347 remains unexplained. No causal production defect was demonstrated, so no production repair or original-game run followed.

The actual generated caller now has a separate opt-in readiness lane, independent of the existing recorder's navigation cutoff and exhausted startup capacity. It records selected startup/worker identity/readiness requests, expression hashes, raw send hashes and sizes, send return/throw, public MessageEvent receipt, consumer delivery, boolean/status results, protocol errors and settlement. Send return is not browser receipt. Public receipt is Node event entry, not browser ingress. Missing records cannot establish missing execution.

The cancellation fixture's generated runtime explicitly enables this lane. Default observers are inert. Limits are 2,048 wire records, 240 page progress events, 64 error events and 256 other existing Runtime/target events, with bounded details and 32 tracked requests. Before navigation, the existing init script arms a finite passive publisher for page/guest presence, generated fixture ready/error state, existing slice/log state and page errors. No additional CDP probe, provider change, timer widening, retry or post-failure diagnostic request was added.

Before allocation, focused checks passed: the actual sealed generated caller's delayed false/true replies, separate protocol-error and missing-reply terminal controls, unchanged real 20-second timeout, zero later protocol/input/capture/quit, inert defaults and record bounds. Controlled startup services in this whole-caller test are an explicit limitation. Separately, the exact composed fixture stage, evaluate/event functions and real startup/readiness/review service passed delayed attachment/readiness, missing attachment and interrupted readiness controls. Existing capture boundary checks, 27 closure refusals, seven service/eight native revocation controls, six observer controls, three actual wrapper/reaper controls, and the locked build passed. These offline controls do not prove native execution.

One native fixture used the sealed 946-provider composition and unchanged guards. native-analysis.json independently checks all 29 observed readiness-lane requests had send, public receipt, matching-session consumer delivery and resolution. There were 375 lane records, including 201 passive events and 119 page progress samples; no wire cap was reached. The original readiness predicate's SHA-256 is 89fbb53dad370301d6253f14f9f7d7ed776868b7e1def396f43925255860baf7.

Request UTC entry, 2026-11-09 Delivered result Public receipt after native pre-send
43 23:39:53.096 startup released 1.873 ms
44 23:39:53.099 false 1.277 ms
45 23:39:53.203 false 1.073 ms
46 23:39:53.306 false 1.354 ms
47 23:39:53.409 true 1.134 ms

Consumer entry followed public receipt by 0.107–0.141 ms for those readiness predicate replies. These are differences on one Node monotonic clock. Screenshot request 100 also received a matching-session response and resolved. The separately downloaded, SHA/size-verified first image was model-reviewed: a blurred, stair-stepped red shape occupied the small black canvas on a white page. It was judged insufficiently reliable for continuation. This is a visual stop decision, not proof of a rendering defect. The review gate was not published. The full fixture did not pass, and there was no input, second screenshot, Browser.close, retry, original launch, or movement/fire/arena claim.

The model stop was recorded at 23:40:22.159 UTC. Owning native cancellation was published at 23:40:22.251448 and SIGTERM sent at 23:40:22.252041. The complete await trace contains no protocol request after the model stop. All 162 owned processes were reaped. Three native parent waits and the local caller wait retain separate stdout/stderr EOFs.

Fifty-three unique artifacts were independently retrieved and SHA/size verified before exactly one delegated normal unforced stop. The settled inventory reader initially expected a filename that recovery had compressed; its failed receipt is retained, and the read-only reader was corrected to resolve the preserved compressed name. The first local audit encountered the still-open allocator stdout gzip; the final audit after its EOF passes, with no gzip/PNG CRC errors. Neither evidence-reader issue triggered browser work or a native retry.

Box bx_jrgbknzy, UUID 652d9f4d-4fa5-40a0-900a-5da11682b32f, is archived. Initial expiry was 2026-11-10 00:03:31.430 UTC; the immutable independent 25-minute watchdog deadline was 00:03:32.191665 UTC. Its configuration was independently rechecked unchanged, and its actual wait was zero with separate EOFs. Stop stop_cfa986bb3bcb6363 completed unforced at 23:41:23.796 UTC; the exact receipt and archived state were independently reread. No owned box remains live. Exact Git plus small patches preserve inputs; no source/media archive was made. /home retains 5.6 GiB free.

Next requires separately authorized visual qualification before fixture input and original movement/fire. The earlier request-347 cause remains unknown; this successful readiness delivery does not retroactively explain that stall.