Pawn native Save As reached; confirmation guard refused

Run scratch/runs/20261113-pawn-native-save-1405/ observed the live closed-outlaws-text-recovery-webgl-1400 resource phase without waiting for the whole WebGL task. Its two opaque historical text artifacts reclaimed 28,106,752 allocated bytes net at announcement. Both encoded SHA-256 values, external full gzip decoded sizes/raw hashes and absent original paths were independently verified. Recovery pointers, gzip files and restore dependencies remain untouched and must be retained.

After the WebGL fixture's actual parent wait/both EOFs, browser closure and profile removal, no heavy runtime child remained. The Pawn supervisor held /tmp/wine-heavy-runtime.lock across its decision, original and cleanup. Fresh prelaunch free space was 1,517,203,456 bytes. The independent 600-second watchdog was observably armed before the single original.

The actual current module, source and six media files matched the prepared 1330 route, including published dialog 56327d542, pointer release 8ab452913 and menu Escape fa008abea. Module SHA-256 is 6888cc5e4ec7693a8a317fe29147994c7f88fe43afe3109fe6ab0e3e349eec9f. All 321 pins were frozen and checked; Git SHA, relevant binary diff and untracked-source patch preserve provenance without source/media archives. 110 parent evidence seals, helper syntax and five actual stream/control cases passed before launch. Controls cover byte identity and EOF for exit 0 and exit 7, truncated gzip refusal, one-request ACK/malformed UNKNOWN and interruption before EOF remaining unsealed UNKNOWN.

Personally reviewed the complete initial board and ordinary e2–e4 followed by Black's Ng8–f6 reply. The same actual software Worker owner was healthy, mouse released and queues drained before the individually reviewed legal d2–d4 drag. Both moves used independently armed UP processes, each ACKed and actually waited 0 with both EOFs. The second capture displays White pawns on d4/e4 and Black's knight on f6. No third move or uncertain replay occurred. The second reply was not authenticated.

Ordinary Game → Save opened the native Save As dialog with the Pawn Games (*.gam) filter. An independently released click focused its File name edit; ordinary WM_CHAR keypresses entered pawn-1405-secondmove.gam. Personally reviewed filename.png shows that exact filename and C:\ directory. No guest/config/clock/AI state was edited, and no forced Present was used.

The concrete new blocker is the run-local UI guard's refusal before the Save confirmation input, not the disk floor. At save-confirm-before.json, the healthy same-owner guest had quit flag 0, mouse mask 0, render submitted/ consumed/completed all 1584 and no pending render entries. Its input queue held one WM_MOUSEMOVE (512) targeting hwnd 65545 at screen (553,338), the filename edit location; the guest post queue count was 3. ui.py requires an empty renderer input queue and immediately requested ordinary quit after that assertion failed. No Save-button DOWN or UP was issued. The guard does not distinguish pending mouse motion in a modal dialog from an unsafe pending edge. This snapshot does not establish why the event remained queued or prove a production emulator defect.

Consequently, the native filename-entry milestone is authenticated, but no guest-produced save file was retrieved, no native history witness was obtained and second-move commitment remains UNKNOWN. The attempted post-refusal read-only extraction met a closed control port; it provides no file-presence evidence. The original was not restarted and the move was not replayed. Next: qualify the modal input queue and confirmation guard before another original, then obtain ordinary Save/history plus authentic file bytes before any third move. Do not weaken the guard based only on this snapshot.

Native PID 1075072 and watchdog PID 1075071 both actually waited 0 with both EOFs; the supervisor also waited 0 with both EOFs. Native and supervisor logs were losslessly gzip-streamed and their complete decoded hashes verified. The owned runtime PIDs were absent and the heavy lock released promptly; LMarbles1410 was notified on the board. The board watcher was separately terminated and is absent, with both pipe EOFs and complete gzip decode/hash checks. Its outer tool session ended 143 before writing a parent-wait receipt; the watcher's actual parent wait is therefore UNKNOWN. Box ID none, state NOT_CREATED, expiry null, stop receipt N/A. No production files changed and no production build is claimed or required for this documentation-only change.