Diablo II demo: same-character combat, retained native save and normal Load, 2026-11-11
The registered original diablo2_demo completed the requested chain with SAMEHERO, one ordinarily created level-1 Barbarian. In the Blood Moor, two upright Zombies became corpses after native targeted attacks. SAMEHERO changed from XP 0 / life 66 of 66 to XP 66 / life 46 of 66. Native Save and Exit returned to the title. The same character's unedited save bytes were immediately exported and recovered locally. Ordinary Single Player → SAMEHERO → OK then loaded gameplay in the Rogue Encampment; the native character panel again showed SAMEHERO, XP 66, life 46 of 66.
Evidence root: scratch/runs/20261111-diablo2-same-character-combat-save-load-1115/ (ignored). This extends the parent combat/save run, whose FIELD combat bytes were lost and whose separate NATIVE load was unverified. No save mounting, game/state/script edits, generated media, or gameplay patch was used here.
Visible proof and retained bytes
Full reviewed images in review/:
baseline.png: SAMEHERO level 1, experience 0, life 66/66 in town.northeast25.png: native “Entering The Blood Moor” during ordinary movement through the town gateway.approach44.png: upright enemies and native Zombie nameplate.attack46.png,attack47.png,attack48.png,attack49.png: targeted combat, first corpse, second reduced health bar, then both corpses.statafter.png: SAMEHERO XP 66 and life 46/66, with both corpses beside him.freshsave.png,title.png,export.png: native Save and Exit menu, resulting title, immediate read-only export.loadsingle3.png,select.png,loadok.png,loading.png: existing SAMEHERO in the normal list, selected character, OK, then gameplay entering the Rogue Encampment.loadedstats.png,afterloadexport.png: SAMEHERO XP 66 and life 46/66 after normal Load.
The life row displays maximum 66 in its first numeric column and current 46 in the second. The initial panel showed 66 in both columns. Enemy identity is supported by the native Zombie nameplate; no quest-completion claim is made.
native-save/ retains the actual game-written bytes. Guest paths, byte counts and SHA-256 hashes are in native-save-manifest.json:
| File | Bytes | SHA-256 |
|---|---|---|
| samehero.d2s | 813 | 57de3f66eed5c2b5cb6903ae81232d32d9b2a4dec5cc563a00ea9eba82371f16 |
| samehero.key | 902 | 29c7012dd4dca077e162119cc4358acd7369f14ff5c32fac6fc1d20847dfbdc6 |
| samehero.map | 24 | 95e612334f4a2b74e6f5c4d300bfd29ddd5c29bc634bb113173c91215d04ae82 |
| samehero.ma0 | 5106 | 812162e10ccd0bb237451ee5f56d12d2c7fe3aaa15cd25c2109d3421922d4563 |
The exporter reads wine._helpCtx.vfs without mutation. Its exported JSON was downloaded and hash-verified before normal Load, then decoded without changes. A second export after Load was byte-identical for all four files (after-load-save-comparison.json). Load reused the native VFS files through shipped UI; no files were imported. This proves the native Save/Load chain within the running application and retains authentic bytes for future independent-launch recovery testing.
Controller and backend checks
Before runtime, the composed-controller regression exercised the actual browser execute capture/export paths, validated capture/click/key/text/export/quit call schemas, and proved missing waitMs is rejected before effects or queueing. The inherited exporter regression passed syntax, native save-path selection, complete high-byte preservation, unrelated-entry exclusion and no mutation. Both test logs are retained. The added scratch command validator is used before request queueing and again by the consumer.
Every dispatched gameplay input had a full current image review bound to the actual process, guest worker, module, render endpoint and canvas. The audit verified 93 exact-ID successful command acknowledgements, including 60 reviewed gameplay inputs; nine stale-scene attempts were explicitly refused before dispatch with zero effects, followed by new captures and new command IDs. There were no unknown input acknowledgements or replays. All 183 recorded native events were trusted and key/button presses and releases balanced. Long presses used prearmed releases and were followed by passive settling and full-image review.
The exact accepted production source remains Git 0db20330ccd6761a705eea70c5dd9bfc7704c147, with the observed recovered module SHA-256 ce57f0ad80254b974a0d8bcd4e2c9cbf396d2357cbdc7874e6084e980e1e68e1. runtime-pins.json, preflight-source-resolution.json, reuse.json, source patches and build-rehash.json preserve the matching source/module/media/control/build references. The reused passing locked-build log is 20261111-avp2-current-marine-encounter-0820/build-final.log, rechecked SHA-256 27d66f5d89a3df20dc3fe67fe0f32cafcc3c4aca928ca2563bf5efc49f2d5664. No production code changed. The original executable SHA-256 is d0aa0d30b55f8313e04026cca560ef0d178ee76b2ece6c3ccdc1fca4af46b3f1; media is referenced by canonical path/hash, without extra local media copies.
Backend evidence records isolated threaded browser execution and WebGL 2 through Chrome ANGLE/SwiftShader; this is not a hardware-GPU result. The first startup omitted the scratch worker observer from provisioning. That controller failed before gameplay input, exited with actual waits/EOF, and Chrome PID/profile/listener absence was confirmed before the corrected relaunch inside the same lease. This was a scratch provision-list error, not a demonstrated shared emulator defect.
Recovery and lifecycle
All 949 remote file references / 44,842,799 bytes were retrieved and hash-checked. A separately calculated post-retrieval remote inventory digest matched the original inventory, and every local recovered reference was rehashed. Audit validated CRCs and IDAT decompression for 84 unique full PNGs, the live module hash, authentic save bytes, reviews, acknowledgements and releases. The oversized first second-check command hit a local argument-length error before remote execution; a compact independent inventory-digest check completed the same verification. recovery-map.json, post-retrieval-rehash.json, retrieval-verified.json and evidence-audit.json retain the results.
lossless-map.json maps compacted large records/modules to byte-exact gzip files, including original sizes/hashes and restoration paths. Native save files and reviewed PNGs stay directly available. Git SHA plus relevant binary patches and owned-harness.patch supply source recovery; no source-tree archives were made.
Owned private box bx_uytzbr5k was created at 11:17:47.482 UTC, with initial expiry 12:02:47.496 UTC and an independently prearmed fixed watchdog deadline 12:02:47.200 UTC. After successful Load, Chrome and controller exited 0 with actual parent waits and both EOFs; independent inspection confirmed absent PIDs, profile and listener. Evidence/save retrieval and the independent hashcheck completed before the delegated normal stop at 11:45:33.574 UTC. Independent GET confirmed archived, active expiry null, stop stop_5999e81e29d87cd1, completed, forced: false, error: null. The unfired watchdog exited 0 with actual wait/EOF (normal-receipt.json, normal-archived.json, watchdog-wait.json).
The requested same-character combat → retained native save → normal Load proof is complete. A future independent-launch import/recovery test is separate from this demonstrated ordinary in-session Load and was not needed here.