Maelstrom: repeated select returns within the first RunFrame Sync
Historical source last updated, 2026-10-24. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
One bounded original-program guest disproves a single parked WSOCK32 select. The genuine SDL_net select-return block counted0,1,3 at batches36165,36175,36185 (two consecutive ten-batch windows,200ms per batch); it counted972 at41999. PollEvent return remained3, actual Sync entry counted1, and first RunFrame entry counted1 throughout. The first RunFrame has reached Sync; the prior interior-address zero counter did not establish pending input drain. No production repair is justified or made.
| Batch | PollEvent return40287d | select return7a3362 | Sync40e000 | select thunk entries/reentries | last imported tick | yield |
|---|---|---|---|---|---|---|
| 36165 | 3 | 0 | 1 | 2 | 7232805 | 8 |
| 36175 | 3 | 1 | 1 | 42 | 7234800 | 8 |
| 36185 | 3 | 3 | 1 | 82 | 7236800 | 8 |
| 41999 | 3 | 972 | 1 | 23338 | 8399600 | 8 |
The after-NextWave trace fired at36163. The observer's first select checkpoint is phase-gated by the real406378 counter, so a menu select cannot consume it. Each checkpoint reads counters, EIP075040f8, ESP074ffb18, yield8, clock state and pointed bytes synchronously while the main guest is stopped between batches, before the harness clears network yield. Frame words are [007a3362,53000001,074ffb50,0,0,074ffb48]. The timeval pointed bytes decode{1,0}; the read fd_set has count1 and socket53000000. These identical arguments were reused across completed calls. The observed select-return increase is the decisive distinction. The final printed counter23341 is later than the41999 checkpoint23338; these are intentionally separate sampling moments.
tick.lastTick is the latest actual batch-clock getTicks result, copied without invoking a clock export/import. It advances1995ms then2000ms; the nominal batch bases advance2000ms each. CallsInBatch0 reflects the harness resetting its per-batch counter; it does not mean the last imported tick is the current nominal base. The diagnostic did not read the separate WASM tick_count global. Historical batch-clock.js (2726B SHA06d08aba14f550954f13d0efc9049394bb9c4ead8e9c4b44317365e218fa817a) and guest-clock.js (3204B SHA f2866f172126e35e0ae17555c32d2fb08aaccd6092e8fa9fe12f0ea09a2afc91), source d8cd8c54c8d63c107696f1e4ef7a1e229dcadc1e, define this state. The exact module exposes no select-specific waiting/deadline/readiness getters, so none was inferred from private offsets. This capture isolates repeated socket polling inside the first RunFrame after actual Sync entry; it does not establish why Sync continues polling, a permanent hang, a socket defect, or gameplay acceptance. No later PollEvent return occurred during the observed window. Event-producer/mutex/input-drain claims are unsupported.
The raw6947B log (historical artifact: maelstrom-select-return-capture-20261024/output.log), synchronized analysis (historical artifact: maelstrom-select-return-capture-20261024/analysis.json), actual command (historical artifact: maelstrom-select-return-capture-20261024/command.json) and terminal/cleanup (historical artifact: maelstrom-select-return-capture-20261024/terminal.json) retain the primary evidence. The observation lines are below8192B; startup/runtime output was guarded at131072B with RLIMIT_FSIZE and a live size guard, no silent truncation. One guest only,42000 batches,9.920s guest loop,10.110s supervisor wall, natural exit0. No additional continuation or second guest. Three PNG actions were preserved from the exact original schedule; they are retained externally, unreviewed by this diagnostic and make no gameplay claim.
The observer helper (historical artifact: ../../tools/maelstrom-select-observer.cjs) is read-only. A hash-gated host preload inserts exactly one call before canonical COM-yield handling; it reads existing exports/memory/clock state and writes bounded diagnostic text. It does not change WASM bytes, guest memory, clock, input, API results or socket readiness. Canonical harness bytes on disk remain unchanged. The historical counter checks EIP at block entry. Original CALL return PCs40287d/7a3362 and actual Sync call target40e000 were byte-checked before the guest; all had positive real counts. The observer's local and remote reachable fixture passed phase gating, completed-return progression, clock synchronization and unchanged-memory checks; invalid pointer and wrong authentic-return controls reject. Missing required exports would block before guest execution. Actual canonical trace-at/trace-from and dereference syntax were exercised before execution; routine fixture errors and their raw failure receipt are retained. Prior source-anchored interior-PC counterexample and its six rejecting controls remain authoritative at donor48bf87b2a159a0a332ca5040483a1cb237d05c87; no previous false causal attribution is reused.
Exact argv (historical artifact: maelstrom-select-return-capture-20261024/guest-args.json) retains P31000/31200, left35000/36000, up35200/36200, space35500/36500, PNG30000/34000/39000, tick200ms and42000 bound. Each first-match flag was replaced once. The canonical parser and supported tree mount enumerate all48 authentic filenames, including Docs/Ambrosia_Software,_Inc._FAQ.txt; only maelstrom subtree is mounted, CWD C:\maelstrom, MAELSTROM_LIB=., HOME=.. Broken comma per-file mounts, wrong root and missing-file controls reject. No network/options/input schedule variation.
Maelstrom original has no candidate-corpus registration. Metadata uses the retained investigation ID maelstrom-original-win32-3.0.6 with registeredCandidate=false, without attaching to another candidate. Owned run scratch/runs/20261024T083020Z-maelstrom-select-return-capture passes the containment checker with no errors/missing/invalid records; root dashboard files were not written. PNGs stay in that ignored run and external recovery, excluded from the committed compact record. The final local gate also rejects missing get_count and a wrong module identity.