Tux Racer title poll stall: causal observation blocked by provider transport
Historical source last updated, 2026-10-30. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
No production repair is justified. The single new launch request did not return an acknowledgement before its55-second client wall guard. A separate read-only observation, delegated snapshot-stop, and cleanup-state request each failed with HTTP525. Whether the new original guest executed is unknown. There is no new frame, input-consumption result, distinguishing runtime control, or gameplay claim. The launch was not replayed and no second Boat was created.
What was actually established
The complete coordinator root union plus this launcher/session rollout passed physical preflight with full8MiB work and separate8MiB publication reserves. Pre-runtime physical49598464, forecast66375680, cap67108864 bytes. Original allocation/access denials and compressed closed staging were preserved. RAM admission was off; disk and process restrictions remained. A655360-byte generated file passed the12MiB FSIZE guard, separately from a rejecting540672-byte output control under512KiB per-stream capture. Closed output-control data was losslessly gzip-preserved after byte comparison; it is still accounted. No full index or blanket checkout was used. Only diagnostic/report files are published.
The authorized finiteTTL1800 fork bx_6qfkmrgj was created09:59:14.818UTC, expiry10:29:14.775UTC, cutoff10:19:14.818, handback10:22:14.818, shutdown10:24:14.818. Independent watchdog armed before API smoke. API write/read smoke passed. The actual historical runner/module hashes matched runtime.json. Canonical validateLocalManifestSync enumerated intended files, validated their mounts, and the exact Nodev24.22.1 executable with512MiB old-space flag started successfully under the intended child configuration. Enumeration and inherited limits remain remotely in capture-causal-oct30/startup.stdout; local node-mount-startup.stdout contains the passing API summary, not that full remote receipt. No compilation or module/assets transfer occurred.
Retained boundary and source interpretation
The parent55-second run ended at main EIP0x00a69449, inside SDL.dll loaded at 0x00a47000 (original VA0x10022449). Its worker T1 was reported active at0x919ab2, yield0, waitH0xe0005. Parent native Enter/mouse acknowledgements are not dequeue or SDL consumption. mainPolls remained2 while interleaved run/return counts rose.
The exact retained runner checks named-event/render/thread-send/sleep/suspension in mainExecutionSuspended, but its cooperative interleave then admits run() while yieldReason1 persists. The observer calls this decision runnable. That label means admission, not executed guest instructions or successful wait completion. lib/thread-manager.js checkMainYield separately resolves main waits; it reads wait handle, handles pointer, timeout and stack-byte count, queries wait objects, and completes or retains the wait. An admitted call that returns does not itself prove a broken scheduler. Observer source and selected scheduler/wait source excerpts were inspected through the supported API; no semantic changes were made.
The planned single run disabled both observers and used read-only getter/stack samples. Native Enter was conditional on an actual PNG matching the known title hash, rather than a fixed early delay. observe-supervisor.py is the proposed instrument, AST-checked only: execution, control acknowledgements, capture, producer-thread state and its distinguishing control are unverified. No exact last API, current main wait object/type, current guest clock, new frame, or causal production diagnosis can be asserted from this attempt.
Cleanup blocker and next boundary
Next useful boundary is a read-only main wait snapshot and its producer: EIP, previous EIP, wait handle/type/timeout/stack, instruction/API counters and clock state, alongside producer-thread execution. Resolve the SDL original VA above against the exact retained DLL. Repeat only after transport availability changes and explicit runtime assignment; preserve the original runner first. A minimal repair needs causal evidence, a distinguishing control and actual original-game retest. Historical Main Menu captures do not supply those missing observations.