Darkstone: two Giant Wasp victories beyond the fork, 2026-11-08
One fresh original session reloaded the newly saved grassy fork, followed the visible sandy route around the tree line, and reached a living Giant Wasp encounter. Personally reviewed live counters advanced from Victories 3+0 to 5+0, with LIFE 28/35 → 24/35, Gold 1532 unchanged, MANA 25/25 and Food 90 → 85. Native Save a Game → New Save preserved the earned result. No pickup or area transition is claimed. Production code was unchanged; no new concrete production defect was demonstrated.
Evidence is ignored scratch/runs/20261108-darkstone-fork-encounter/ (R).
The predecessor is the new-slot handoff,
commit 12e1af819. Its recovered/original/native-vfs-saves.json supplied the
untouched game-written inputs, checked before staging and again after execution:
| Input | Bytes | SHA-256 |
|---|---|---|
characters.lpl |
1,342,092 | dee39793c2f1a43bc5af205e428b9c604da1ba31d08e3e7f1fc4dddc9b6328ee |
game641d_6af03b8e.sav |
3,204,264 | 3ad043582acc9f02426ef5700693fd39331c9889cd732aeecd37b1bae0dfaa67 |
R/checkpoint-provenance.json records their source paths and original VFS paths.
Only these two files were appended through a run-local composed HTTP manifest;
registry and original media manifest were untouched. Ordinary manifest resolution,
merge/normalization, production WineAssembly.loadFiles, mounted hashes and native
*.sav enumeration passed locally before allocation. Original-session read-only
mounted recovery independently matched both hashes before native loading.
Reviewed route and encounter
Native Load a Game → sole selected onward slot → OK restored Ardyl's grassy fork beside a rock and trees. The selected preview was not counter proof. R/review/restored-inventory-settled.png establishes actual live Gold 1532, Victories 3+0, LIFE 28/35, MANA 25/25 and Food 90 after loading. Documented Space closed the panel. Panels were never treated as pause.
Six individually reviewed native route holds, each at most 2,000 ms with UP
independently armed before DOWN, followed clear sandy ground around the trees.
Every next movement followed a personally reviewed settled view. The targets
were canvas (0.82,0.36), (0.83,0.72), (0.25,0.74), (0.32,0.79),
(0.69,0.78) and (0.18,0.74). These selected visible route ground rather than
an unreviewed attack target. Before/action/settled PNGs and exact commands remain
in R/retrieval-verified.json's mapping.
A living winged insect appeared during traversal. The initially retrieved settled frame showed a pale body, blood and recent damage; its appearance was initially uncertain. Reviewing the retained action frames then established the Giant Wasp caption, living insect and health bar in R/review/open-track-action.png. R/review/encounter-inventory-settled.png showed actual Victories 4+0, LIFE 27/35, Food 87. No targeted attack is claimed for this first encounter; it arose during native movement/ordinary combat behavior between captures.
A second living wasp remained immediately above the warrior. LIFE fell to 26/35
while the panel was open, further demonstrating that it did not pause combat.
After closing the panel and retrieving a fresh world view, I personally reviewed
R/review/current-target-before.png. Only then was a bounded 1,000 ms native
left hold at canvas (0.505,0.38) admitted against that visible wasp.
R/recovered/original/reviewed-wasp-action.png and
R/review/reviewed-wasp-settled.png show the Giant Wasp caption/health bar and combat.
Passive R/review/wasp-followthrough.png shows no remaining standing wasp/bar.
R/review/final-counters-settled.png establishes Victories 5+0, LIFE 24/35,
Gold 1532, MANA 25/25, Food 85. No healing, forced scene/clock/configuration,
guest-memory write, fabricated counter or input replay was used.
Selected media hashes (all paths relative to R):
| PNG | SHA-256 |
|---|---|
review/open-track-action.png |
76ed12baba3988f75c6b0012eaa55951ac93fea0b0304d6ca4d9af41eaaf49b8 |
review/current-target-before.png |
38adb62e0b010054ecc3a24e380ea622d657ab9d4a6bf8048e9448e3f7eab344 |
recovered/original/reviewed-wasp-action.png |
0c0ffb51d3c57d9dd778927eef501f1f3875e985d356d3707532dc6173a71d58 |
review/wasp-followthrough.png |
375cda6e487b5bc0a06b97a7ecfa8bfc3107604320410e9e6f79321051817279 |
review/final-counters-settled.png |
e98c1171e188f5f4b7c593336c1452630b58a9230ebad91afe45001ab95d335b |
review/saved-encounter-world.png |
7d6c8ebddbb2fca36fa0e29d254e28ab7da86ef47de67f7420b9974c14df9a5c |
New native save and recovery
Native Escape, Save a Game and New Save wrote the following untouched outputs. Read-only VFS recovery at 09:33:31.320 UTC retained them before browser closure:
| Output | Bytes | SHA-256 |
|---|---|---|
characters.lpl |
1,342,100 | cad82ff5a0bfbd556db2d06fc078e2570ee1e3ae4d4efb884bd313e6550473bf |
game641d_6af041d7.sav |
3,204,264 | d977d538fac58afd16d2fed4eff7e3fa1f50c2b1e14a74f9cf791309e6ad8372 |
R/recovered/original/native-vfs-saves.json maps native paths to hash-named .bin
artifacts in that directory. R/new-save-recovery.json gives exact local paths,
lengths, hashes and the recovery recipe. Verify both bytes, append them to the
run-local composed manifest at c:\dsave\characters\characters.lpl and
c:\dsave\games\game641d_6af041d7.sav with loadMode:required and
httpRange:false, then use ordinary HTTP/loadFiles, mounted-hash/wildcard checks,
and native Load a Game → the new slot → OK. Inspect actual counters after loading.
This newest slot has not been reloaded. The loaded onward .sav stayed
byte-identical. R/review/saved-encounter-inventory.png confirms 5+0/24/35/85 after
saving; R/review/saved-encounter-world.png shows the warrior beside the encounter.
Runtime controls, evidence and shutdown
Runtime source was published main commit
24913adc084472f8ac79b222520a0da4f2b45d81 (also recorded in
R/checkout-head.txt and R/runtime-pins.json). All 432 tracked runtime inputs
matched that commit; peer WIP was excluded. All 527 staged input pins matched
before and after execution. The executing guest/render module was
6cf120d0d3b3db1823a8105eb26ef52a9093a7f871c37a46117475e213fda1c0.
R/source-recovery.json records exact Git plus relevant binary diff and owned
helper patches; R/media-refs.json names existing media paths and hashes. No source
or media archive was made.
Current actual production guest/render Worker readiness, identity/staleness,
trusted native input and independently released delayed/rejected/pending ACK
controls passed under /tmp/wine-runtime.lock. Observed/unobserved actual x86
ExitProcess(37), distinct UD2 trap, registers/EIP/ESP/yield/blocks and renderer
retirement matched. Observer receiver/arguments/return/exception/once-only
transparency passed. Both local control browsers retired with exit 0 and profiles
removed. UNKNOWN remained sticky; no original input was replayed.
Two run-local preparation mistakes were caught before Chrome: an absolute helper pin duplicated the root, then a predecessor composed-manifest hash was stale. Both retained failures preceded the passing controls and sole original session. The first immediate post-close custody assertion also failed without identifying its predicate. A subsequent distinct read-only diagnostic and settled inventory confirmed all owned PIDs/ports/profile/writer/input drained; its original failed observation is retained, with no invented failure cause or guest replay.
The original had 37 exact successful ACKs, zero UNKNOWN commands, 18 independently released input pulses and 47 trusted events (11 balanced mouse pairs, seven balanced key pairs and 11 moves). No original guest process exit, trapped terminal slice, Runtime exception or target crash was observed. Explicit finish-run requested ordinary Browser.close at 09:33:51.741 UTC; the registered app was still running immediately before closure. Chrome exited 0 / no signal at 09:33:53.255. This supplies no cause for any historical exit.
All 382 remote artifact references / 29,289,592 bytes, comprising 308 unique contents / 29,274,037 bytes, were retrieved and independently SHA-256 verified before normal stop. R/final-audit.json passes 37 PNG CRC/scanline decodes, exact ACKs, trusted balanced events, input releases, save hashes, helper syntax and owned containment. Darkstone remains unregistered in the candidate corpus; R/investigation-result.json attaches to no unrelated candidate.
Sole box bx_ppqd8ukq was created 09:23:44 UTC, initial expiry
09:49:44.221. Detached normal-stop and independent 25-minute watchdog children
were observable before allocation. Watchdog deadline was 09:48:44.381, browser
deadline 09:46:43.962. After verified retrieval and cleanup, normal stop
stop_f0897dcd589f2a26 completed 09:35:45.820, forced:false, error null.
Final state is archived, stopped 09:35:45.793, expiry null; the final
completed receipt is R/stop-receipt.json. Both lifecycle children exited 0 and
the supervisor has no remaining children. No old box resume/delete, public
deployment or subworker was used. Disk stayed above the required 1.5 GB floor.
Only this note is committed. Next useful work is reloading the newly earned 5+0 slot and reviewing the next route/encounter. Full27, parity, audio and FPS remain open.