_reports/worklog.md

# Game archive verification worklog

Bucket: https://huggingface.co/buckets/smodusermc/offline-games

## Scope
- Keep small scripts, manifests, logs, and verification evidence locally.
- Stage game assets on disk under `.cache/` (not the small tmpfs); remove after each game.
- Credentials are supplied through the process environment, never project files.
- Geometry Dash is excluded.

## Verification meanings
- **Storage audit**: bucket objects exist and their recorded sizes match. Not proof of completeness.
- **SHA-256 checked**: staged bytes match a recorded hash. A difference is recorded, not silently declared fixed.
- **Menu/intro observed**: visually reviewed browser evidence shows actual game output, not only an engine loader.
- **Input smoke passed**: a narrowly defined input assertion passed. This is not a complete playthrough.
- **Inconclusive/resource limit**: the 2 GiB test host could not safely finish. Not proof that a game is broken.

Browser tests use a local HTTP mirror with outside browser requests blocked. The repair phase can fetch missing resources server-side; the subsequent verification phase cannot. Routing absolute source URLs and supported runtime-library URLs to the local mirror happens in the HTTP response only. This is NOT a test of opening an unmodified HTML file from `file://`, nor a promise that the whole archive consists of self-contained HTML files.

## Initial findings
- Old regex-crawler error lists contain many false positives from bundled code and internal module names.
- Several games already reach menus with their captured assets.
- Shared Flash runtime dependencies were missing from some per-game inventories and are being added from observed browser requests.
- Eleven entry-page sizes differ from their old snapshot metadata. These existing bucket edits are being preserved, not overwritten to match older files.
- A console-emulator startup stalled the host. Further tests are isolated with RSS and elapsed-time watchdogs.

Progress and individual outcomes are recorded in `reports/verification-*.json`; final human-reviewed classifications will be in `reports/verification-summary.md`.


## Lazy-loaded audio repair and next eight entries

Restored Thirty Dollar Website sound library: 191 selectable WAVs plus three intros all decode. Preview and sampled sequence produce nonzero Web Audio output offline. Recovered 54 icons; restored checked checkbox SVG from upstream and fixed both CSS paths. Added 46 Cookie Clicker effects; all 75 declared jukebox sounds decode. Basket Random: 16/16 WebM files decode; one stale size declaration validated against current source SHA256. Nine audio replacements use documented upstream URLs. Added IDs 155,156,174,175,180,181,281,591; outcomes recorded without calling startup-only or resource-limited cases playable. Final presence/size audit: 81 manifests, zero missing recorded objects, zero size mismatches; no full SHA audit or playthrough implied. See LAZY-ASSETS.md for current scope. Temporary assets and bulky caches removed; credential-free tools/reports retained.


## Next eight entries and full declared-map/media audit

Added IDs 182,183,359,373,374,376,575,582. Also audited/repaired 180 and 181. All 149 declared temple-map paths are present. Recovered 94 OvO audio files and 113 maps from identified upstream releases/builds. 1,213 recorded SHA comparisons, 287 audio decodes, 587 image decodes, and 704 declared-path checks passed. No JSON parse failures or missing tileset JSON references. Cumulative presence/size audit: 89 manifests, zero missing recorded objects and zero mismatches. Not a full playthrough or proof of undiscoverable assets. See NEW-GAMES.md for runtime limitations and exact provenance.


## batch-18a03fb44e: larger batch

Default changed to 30 games (20–40 by size). Captured and checked all 30 selected entries. 800 recorded-hash comparisons, 377 audio checks, 180 image checks, 21 SWF checks. 205 declared gaps remain; runtime evidence kept separate from asset checks. 190 candidates remain queued. Stable evidence: _verification/batches/batch-18a03fb44e/. See LARGE-BATCH.md for caveats.


## 20-entry large-port batch and storage-growth investigation

Stable batch batch-f29ef445fb. 983 source-hash comparisons, 454 loose audio decodes, 66 loose image decodes. Approximate bucket growth 8.033 GB, including 275 checksum-matched Deltarune audio-path recoveries and four case-only aliases. Crashout loader repaired; Dead Tapes missing source binaries explicitly incomplete; source declaration discrepancies and runtime limits retained. See HEAVY-BATCH.md.