Pawn: first current move/reply, second reply stalled

Run scratch/runs/20261113-pawn-qualified-two-moves-0730/ launched one original Pawn using current04 and the six original media. It confirmed 1.e2–e4 Ng8–f6, then made a separately reviewed d2–d4 drag. The second native AI reply was not observed. The task's two-reply target is therefore incomplete; the new blocker is a stalled second native reply, rather than the previous capacity refusal or legacy-owner preparation gap.

Before any launch decision, the native descriptor, local supervisor and exact qualified parent0640 input/owner helpers were frozen at Git 2404d8b55fddbaa0b492501063b2e050487bc7c7. All 63 sealed parent evidence files verified by SHA-256; no consumer controls or GetDC matrices were rerun. All 319 source/module/media pins passed before launch and after shutdown. The module remains 04c534192ee3fe39e388536c2ce710c12c47a5cd2ca4d95c19f9d00b1e8f0a17. Git plus relevant binary diffs and owned helper patches preserve source; media is referenced by existing path/SHA, without copied media or archives.

The assigned backup0650 launcher PID926718's actual process identity, registry, board progress and completed backup receipts were observed and saved in dependency-outcome.json. Backup reported 69,382,144 allocated bytes reclaimed, with normal stopped snapshot and restore verification. Its files and NFS0700 files were untouched. Only after that actual resource outcome, this run measured 1,526,427,648 available bytes on /home, followed by a fresh 1,526,054,912-byte prelaunch sample. The independently observable 600-second watchdog was armed before native PID937365 launched. No RAM gate, disk loop, cleanup, resource mapping edit or new box was used.

I personally reviewed the full unobscured starting board. The unchanged guard accepted its actual legacy Bridge WorkerConsumer, Worker thread3, device 136454184, generation1, module instance, window and memory buffer. The first finite drag had independent mouseUP ACK and parent wait0 with separate EOFs. reply1-settling.png shows White e4 and Black Nf6 with yellow g8/f6 outlines; SHA-256 3e8ca2c358d75164c9ced1fa40b3f74b9efd81a3534eb8ba76ccb1d6a250e9f2. That full board was separately reviewed before the legal d2–d4 drag.

The second independent mouseUP also ACKed and waited0 with both EOFs. The displayed white pawn moved to d4, but Black remained on f6. An ordinary, separately guarded cursor move off the board did not change that result. Repeated read-only captures remained identical while batches advanced beyond 1.1 million. The final full-board capture second-reply-stalled.png hashes to 0203e66c1c092c445af825ae6c2f77399fc8f5f198e4c086d26794c2530a40ad. Its lone yellow c3 outline is not promoted to reply evidence. Read-only observations show mouse mask0, input queue0, yield reason0, no quit/error, unchanged accepted owner, and render queue submitted=consumed=completed=1578, entries0/pending0. Sampled EIP0x40ca3f follows the original's Sleep(10) call; static disassembly and its Sleep import mapping are retained. These establish a quiescent observed path, not the root cause of the native turn transition or proof that the second displayed move was internally committed.

Normal control quit ended the original after 310.014 seconds / 1,132,983 batches. Actual parent wait0 and separate stdout/stderr EOFs passed for both the native process and watchdog; both independent release processes also waited0 with both EOFs. All four owned PIDs are absent. Syntax checks and source/media/helper rechecks passed. No production change or locked build is claimed; no production defect was isolated. No forced Present/repaint, substitute board, guest memory/config/clock/AI write, automated player, guard weakening, original replay, children or public deployment occurred.

Next: diagnose the original's second-move input/turn transition from this bounded evidence before another guarded original. Do not count two replies. Box ID none, state NOT_CREATED, expiry null, stop receipt N/A.