Maelstrom: single CheckSockets return capture remains unresolved

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.

Exactly one original guest ran on bx_rfd49qx6, using the corrected single trace PC 0040e12a. The actual owned argv passed tools/maelstrom-startup-divergence.py --check-argv after checkout and again immediately before guest launch: startup BATCH_SIZE=1000. The unchanged 42,000-batch ceiling stopped observation while the guest was parked inside select. No CheckSockets return record was captured. Its EAX, current-frame missing slots, packet metadata and selected branch remain unknown. Neither timeout nor positive readiness nor SDL_net's internal errno retry is selected from missing records. No second guest or observation extension occurred.

Observed stopping boundary

Observation Value
Historical startup-path counter 406378 1
Sync entry counter 40e000 1
RunFrame counter 4019a0 1
PollEvent return counter 40287d 3
Select return counter 7a3362 972
Select thunk counter 075040f8 22,370
Requested 40e12a trace records 0 of at most 16
Final EIP / ESP / EAX 075040f8 / 074ffb18 / 53000001
Last observer batch / yield 41999 / 8

The unchanged passive observer retained four parked-select samples at batches 36165, 36175, 36185 and 41999. The parked stack begins with the authentic select CALL return 007a3362, followed by 53000001, a read-set pointer, two zero pointers and a timeval pointer. Its timeval is (1,0); its read set has count one and socket 53000000. These are pending-call observations, not completed select results. Final EAX is not a CheckSockets return value.

The positive startup counter reestablishes startup reachability after the prior five-PC startup-budget failure. It does not establish packet acceptance. The 972 return hits do not establish 972 Sync iterations. As documented in the accepted Sync diagnosis, CheckSockets can retry select on errno 4; that errno is not captured here. The precise unresolved predicate remains the return from CheckSockets and the current-frame barrier state at that return. The actual finite observation ends at the parked select thunk; no production repair or permanent-hang claim follows.

Exact configuration and evidence

The only trace address is 40e12a, the true return after the original CALL at 40e125, before ADD ESP,8; TEST EAX,EAX; JNE 40e19f. The intended zero successor is 40e131. None is claimed observed by a missing trace. The six historical count PCs remain counters, not added trace/break/watch PCs. The trace is armed from batch 36163, limited to 16 records and memory esp:384,0x41f9d8:128,0x420658:12 (at most 8,384 raw snapshot bytes; actual zero). The 42,000 batches, 200ms clock and complete P/arrow/space/PNG input schedule remain unchanged. No forced network/readiness/input change occurred.

Ordinary UID1000, Node24.22.1 and pinned pngjs7 passed the retained preamble, font, codec/pixel identity and rejecting PNG controls. The passive observer fixture, resistant-descendant TERM/KILL/reap control and 131,072-byte output guard control passed. Full stdout/stderr was retained without truncation. All three PNGs are retained and visually inspected: the menu is visible at 30000; the later images are black/starfield frames. They do not establish gameplay. The analysis (historical artifact: maelstrom-single-return-sync-capture-20261024/analysis.json) keeps unknown return/packet fields explicitly null. The evidence seal list binds the complete raw output, terminal, command, PNGs and compact gate receipts.