Alice second gnome to knife — 2026-11-03

Run: scratch/runs/20261103-alice-second-gnome-knife/ (ignored evidence). Result: ordinary NewGame/Easy reached the timber passage beyond the second gnome. Knife acquisition and first attack remain unverified. A malformed capture request ended this bounded run; its task-local admission guard was repaired and tested. No emulator/game source changed and no successor was launched.

Runtime and retained inputs

The accepted crossing is 126b3c131f0ed1dfc129bbdb9c21ac742d52ce5f. Its supported/jump checks are route prerequisites, not new acceptance claims. Thermal flags were clear; this run makes no wind-force claim.

The inherited archive was restored and verified locally and remotely: all 7,803 payloads, 140,322,555 bytes, SHA256 34e33a0a97a977eb5867b66366e031e0b408c350582b0d0fdacc8584a6e91046. All 6,024 inherited sealed inputs, 5,836 exact Git blobs and 43 original game/dependency pins matched. Complete unchanged compile inputs justified reuse of runtime 277f9a32c984562ef33b2d32a5c13df949b5f87a, module 2,125,777 bytes, SHA256 065267ce7c208cb369b1fa930425627b3a238c9e184e5aa46d8ef16143b8b5b8. Foreign MDK2/Jedi WIP was excluded; no build or compilation-container download.

Runtime began only after MDK2's explicit release. One fresh TTL3600 boat, bx_zg9n87ue, original expiry 2026-11-03T22:15:31.856Z, was recorded before use. Original PID6026/session 6caff4ff-d9ca-41ce-89a2-72a50b21b63a ran the software renderer, frozen cooperative mode, batch size100000. Both compilation and instantiation receipts match the exact module. Diagnostic flags were audited; input-state was prohibited because --trace-input-dispatch was absent. No guest writes, injected calls, cheats, save/config edits or public deployment.

Inspected onward progress

Ordinary controls replayed the opening, first gnome, plank jump and far shelf, then allowed the second natural sequence to finish. Personally inspected captures include second_done.png, onward-east.png, timber2.png, and jamb-right.png; image hashes/reviews are retained in visual-reviews.jsonl.

Native onward_start at batch6957 places the selected Player at (-4781.648,4854.562,-127.181). After ordinary eastward movement, timber_onward at7008 reaches (-4640.385,4854.544,-119.861), facing yaw0.011. Another forward segment leaves that origin unchanged at7048. The inspected image shows the timber post, with purple light down the passage. Original BSP brush sections identify a jamb at this approach and possible clearance to the south. An ordinary right strafe produced inspected jamb-right.png, showing Alice in the central opening. A further32-credit forward segment completed at7098, with release/settle through7102; its failed capture means clearance is not proved. No gameplay beyond that point is claimed.

Eight held native samples pass native-route-assessment.json: one selected Player object61777448, actual original module/PID/session, matching native and client command bytes, 56 relocated code windows/377 operands per sample, plus 8 same-held original code windows and allocation-header/tail checks. Across 107,375 allocator events, one matching allocation (2048-byte request, tag4, 2068-byte block) and no matching free/reuse or observer error establish the sampled lifetime. Conservative raw capture fields remain unmodified.

The original script/map plan continues east before turning south to the knife altar (-3978,4326,-168). knife_cat attaches a knife to a cinematic fake player; that alone would not prove controllable ownership or attack. Neither occurred in this run. Exact gnome subtitle text and target damage are also unproven.

Preserved failures and tested repairs

An early module-observer read caught a partially written receipt. Its first failure is retained; the reader now treats incomplete JSON as pending while still requiring exact PID/module identity. A premature capture fetch exposed a pending-receipt exit-code bug; only completed receipts now permit dependent retrieval. Focused positive/negative tests pass for both repairs.

The final clearpost plan used incorrect cmd fields for step/png and was rejected before dispatch. Corrected ordinary movement completed under fresh labels. Its clearpost2 PNG request omitted path; terminal response200 was png needs a path. The existing stop-on-first-failure protocol preserved first-error.json and stopped the original game cleanly (exit0). Subsequent capture/sample attempts were blocked by that retained failure; no request was resent and no replacement game was launched.

The task-local control-plan guard now rejects missing, empty, relative, non-string and traversal PNG paths before either dispatch entry point. Existing input/diagnostic guards and a valid absolute PNG control pass locally and in the same boat. This repair is tested offline after the exit; native retest is absent. Recovery tests also reject changed payloads, duplicate members and traversal; the first outdated recovery-test fixture failure is retained. All changes to helpers are confined to ignored owned scratch, sealed separately from the unchanged compile inputs. This documentation-only commit needs no source build.

Recovery and release

The original game and supervisor are absent; control8123 is closed. No final native fault-status was collected after the terminal helper error, so no final fault-clear claim is made.

Complete local recovery passed before STOP: 7,224 payloads, 137,677,192-byte archive, SHA256 9794e67a79ee6db80e7726b7a051f8de14d60fcf034d3912269cff1107c44009. Every archived payload was streamed for full size/SHA, including canonical and compatibility inputs/modules; all43 game pins remain in the recoverable seal. Canonical SHA is above; compatibility module is2,128,390 bytes, SHA256 e7c59d3d39d937f5bef162b9f052259075ff1040d49f5b1be40e8534c5e9fdfe. RECOVERY.md records restoration and validation. Local disk remained above2GB.

A completed verified snapshot was confirmed before the explicitly delegated normal stop. Actual state is STOPPED, normal/nonforced/no error, completed 2026-11-03T21:44:48.657Z; snapshot completed/verified 2026-11-03T21:44:48.591Z. Runtime was explicitly released on the messageboard. No foreign/predecessor boats were operated. Remaining work would be an ordinary post-jamb route to the knife and an inspected first attack; this task stops here.