progress.md

Progress

Last updated: 2026-10-25.

Done

Disc and archive layer

• Read CloneCD raw images (2352-byte sectors).
• Read Alcohol images with 96-byte subchannel (2448-byte sectors).
• Read ISO 9660 images and stream files as io.ReaderAt.
• Identify the first-press disc: it is a voice-only disc.
• Crack the ARC directory: count + 36-byte records + additive key arcana.
• Verify the table against offsets and against known manifests.
• Implement the ARCANA LZSS codec.
• Implement the KWD1 wrapper.
• Unpack every archive from the DVD-ROM edition.
• Unpack the shop-demo archives and the arcana demo archives.

Runtime

• Go HTTP server with /api/script, /api/assets, /media/....
• CP932 script parser with full-width normalization and labels.
• React 19 + Vite + TypeScript frontend.
• 640×480 canvas stage.
• BMP image loading, LRU cache and next-image prefetch.
• Transitions: クロス, ワイプ, ワイプ２, 上下, 円１, 円２.
• Pause-aware clock.
• BGM playback and loop.
• Sound-effect and voice playback.
• Script commands: BMP, UPD, WAIT, MUS, SEV, JMP, END.
• Shop-demo loop (JMP デモスタート).
• cmd/extract CLI for directories and ISO images.
• CP932 dialogue parser with multi-line quotes and speaker names; 説 narration is normalized to an unnamed speaker.
• Text box, typewriter, click/Enter to advance, ￥ page breaks.
• Layered stage: background + L/R/C character stands, composited at natural size with a black color key (fixes stretched sprites).
• DVD pack selection and lazy LOD script loading.
• Opening yr1_00.txt plays to LOD YR1_01.
• Scene/chapter links in the URL (?pack=dvd&scene=yr1_02&cmd=123).
• Batched Google Translate sidecar for all DVD dialogue/choices; L/toolbar/Settings language switching with Japanese fallback.
• Line-level repair pass for partially translated Google output; only moans/onomatopoeia remain untranslated.
• scripts/repair_translations.py to rebuild partially translated pages.
• Save at choice points; toolbar remains usable above the choice overlay.
• Navigator Branches tab for scene-local choice targets and route changes.
• Chapter browser grouped by actual STS titles; branch jumps require confirmation.
• Global chapter-level story map at /api/storymap, with explicit route changes.
• Side panels share one tabbed dock (Archives / Navigator / Settings) instead of overflowing side-by-side.
• Ghidra helper scripts and durable reverse-engineering notes committed under tools/ghidra and docs/engine/reverse-engineering.md.
• Advancing dialogue stops the previous voice clip; SE channels remain independent.
• Gallery event/recollection cards show covers, heroine labels and a related Story link; event scenes are kept out of the story route.
• Removed the experimental story-interlude path after confirming e… scripts are CG/event recollection content.
• Gallery scene selection from the title screen clears title URL/state and starts the scene.
• Gallery Scenes is an event/CG recollection browser with cover thumbnails, heroine labels and 八尋編 / 朧編 grouping.
• 18+ in-story scene route/choice guide, Mitama market/cultural context, Tony/Ran and Itou Mai voice-actor notes, plus voice-production/ work-culture/marketing notes in docs/guides.
• Archive manager: Recent/Pinned/Chapters, search, notes, pinning, thumbnails, overwrite/delete, save current, legacy-slot migration; snapshots record the 八尋/朧 route and show it on archive cards.
• SQLite archive persistence (data/mitama-saves.db) with REST API, localStorage cache/fallback and offline delete queue.
• Page/chapter navigator: command scrubbing, dialogue previews, chapter grouping, filtering and direct scene entry.
• Archives and navigator docked as side windows beside the stage (including fullscreen) instead of centered modal overlays.
• Single-select choices (SLT and MSLT); SEND menu terminator.
• Case-insensitive LOD resolution (fixes YR1_02 → yr1_02.txt).
• Chapter transition from yr1_01 to YR1_02.
• KWD1 .XA audio decoder (pkg/xa), verified byte-for-byte against xadec.dll under Wine.
• Server-side .xa → WAV conversion at /api/audio with an in-memory cache.
• BGM and voice playback from the DVD data.
• kunoichi.arc extracted into the DVD pack (female voices).
• Scene URLs auto-start and rebuild stage/flags/BGM.
• -dev Go-server mode proxies the frontend to the Vite dev server while keeping /api and /media on Go.
• Resuming at a TEXT line replays the preceding SEV voice.
• Engine debug panel (D / ?debug=1) and /api/debug APIs.
• Canvas stage with a DOM dialogue overlay; selectable text and fit-to-viewport layout.
• Original title screen with new/continue, gallery, settings and quit entries; new game shows the original 八尋/朧 主人公選択 screen; scene links still auto-start. The Settings → Title entry returns to it and Continue resumes the current position.
• Space hides dialogue and HUD; click/Enter restores/advances.
• ? opens a keyboard-shortcut help overlay; the toolbar has no pause button and uses a text mute control.
• Fullscreen keeps a letterboxed 4:3 stage.
• Inline voice/SE annotations are hidden and played as page cues.
• Header/footer removed; controls live on the dialogue box and in ☰ Settings (pack, volume, debug).
• Background stack semantics: BMP BLACK 消去 reveals the background underneath instead of clearing it.
• Dialogue box is opaque and the DOM overlay updates per typewriter tick (no stale/accumulated box alpha).
• Adjustable dialogue opacity and immersive subtitle mode in Settings (persisted per browser); Settings opens as a docked side window.
• Typewriter modes: character, word and instant; fixed the first-frame slice(0, -1) flash caused by a negative RAF elapsed time.
• Hover-English tooltip for Japanese dialogue, with Settings sidebar and H shortcut.
• Optional HD asset overlay + waifu2x upscaling pipeline for smoother character stands and backgrounds, with 4× → 2× → original priority.
• Independent BGM/SE volumes and unified settings persistence (mitama.settings.v1).
• Replay-voice button beside the speaker name tag (current page/line), using a microphone icon; speaker abbreviations expanded to full names.
• Gallery: CG viewer, scene review, music player with official BGM titles/durations, seen tracking and CG-to-script story jumps.
• URL-addressable title, protagonist select, and Gallery tabs, including refresh-safe view/tab/from state.
• Refactor: App settings/archive/keyboard/battle modules and Player audio/dialogue/stage modules extracted.

In progress

• Full-game flags (IFT, FLG, EQU, INC, DEC, FCK, RND).
• Flags are included in save snapshots.
• MAP/PARTY state tracking, includes archive snapshots and resume.
• .pty / .chr battle definition parser and /api/battle endpoints.
• Playable simplified battle screen: shared party HP, enemy strikers, 剣/術/受 actions, victory/defeat, flag 20 branching.
• Battle BGM and idle/action sprites from .o / .spt / .mtn (color-keyed).
• Battle auto-resolve modes (win/lose/random) remain in Settings.
• Weighted enemy actions from :actionrate, special techniques from :special / :spname, and special attack motions.
• Motion se / c_se / se_attack, flash, effectem, movement offsets and parsed waitmov / attackend.
• :actionrate power range and optional :win/:lose jingles.
• Exact attack/hit/damage formula and map encounters.
• Interactive map movement and encounters (map scenes are currently script-driven through BMP MAPxxx + MSLT).
• sprite.arc / .spt / .mtn sprite animation.
• Server-side save files compatible with save%02d.dat.

Next milestone

Flesh out the battle screen with .o / .spt / .mtn sprites, battle BGM and the actual action-rate/special-move rules. Then implement map movement and encounters. Audio, chapter loading and story branching are working.

Notes / decisions

• Extracted files are ordinary .bmp/.wav/.txt where possible, so the player and debugging tools can use native browser support.
• The UDF DVD image is currently unpacked with 7-Zip. A small read-only UDF reader is a later task; it is not required for engine work.
• The shop-demo executable is the behavioral reference for the command VM; the DVD executable is the reference for the full command set and flag model.
• Assets are never committed. data/ and .cache/ are gitignored.