Corrected independent Maelstrom native acceptance — 2026-10-24

Accepted candidate cbac94c066710c18a74b6802b1f50b0b475a6603 without changing candidate tools, tests or assertions. Reviewer f64fc160b209ac283e8a7686ac3f520901a554e5 remains a failed, incomplete acceptance: its same-address counter rearm retained two hits. That was reviewer fixture contamination, not a candidate defect. This task used disjoint reviewer addresses 0x7a3c62/0x7a3c66; candidate addresses remained 0x7a3362/0x7a3366.

One corrected native execution passed, with bounded live JSON rows emitted before assertions. Pending yield at the exact breakpoint halted 5 before counting; skip-once then counted one entry and halted 4 without execution or decoding; the next entry halted 5 again. Consuming its second skip left two hits only in the disjoint reviewer fixture. Candidate original expectations remained unchanged.

The original 972-entry pending-yield loop passed: head count 972, interior count 0, EIP 0x7a3362, halt 4, EBP unchanged, cache stores unchanged, zero target snapshot reads. After clearing yield, the original MOV EBP,EAX / CALL EDI prefix with synthetic RET reached halt 5 at 0x7a3366, head count 973, interior count 0, EBP updated. Exact retained DLL bytes 8b e8 ff d7 83 38 04 establish that this target follows CALL EDI. This synthetic RET does not identify the historical EDI callee or errno.

Historical actual zero-record cause, historical EDI target and errno remain UNKNOWN. A count of 972 does not establish 972 native calls. The synthetic interior example does not establish an interior-only cause for the game's missing records. No production defect or gameplay success is proven.

Unchanged focused provenance/rejection suite and boundary suite passed before native execution. Both are also required on the reconciled committed tree before atomic main/task publication. Native evidence is reused on that tree only after verifying all relevant candidate paths unchanged; a second native execution is neither needed nor performed.

Native input: standing 1879746-byte module SHA-256 8fac39936b6df2418ad86cab118cef19a65a727f649fbc14c412b37818904686. Fixed shared linear memory 512 MiB, Node heap 512 MiB, CPU 90 s, wall 60 s, output 131072 bytes, per-file 25165824 bytes. Fresh memory gate and in-probe memory gate passed. Cheap tests used heap 256 MiB and CPU/wall 60 s. Disk ledger includes checkout/Git/evidence, full external session/launcher, positive SQLite/history growth, future growth and 8 MiB reserve within 64 MiB, plus other worker's 64 MiB reservation and host floor. See adjacent admissions and receipts. All tested child groups exited naturally and were reaped. No box assigned; coordinator retains lifecycle ownership. No source build, toy WAT, guest, acquisition or deployment occurred.

Recovery: candidate files are immutable Git blobs at cbac94c. Run the unchanged focused/boundary commands from their terminal receipts. To reproduce independent native validation, restore the adjacent independent-native.cjs and supervise.py under TASK/ops with candidate checkout TASK/repo, restore a fresh task-specific admission ledger and apply fresh resource gates; the standing module must match the exact SHA. Never reuse historical capacity measurements as admission. Historical resource failures in prior tasks remain failed.