Current development milestone: 0.10.24 (DOS signed-wait correction and Play Mode trace audit)
An open-source Unity reimplementation of Realms of the Haunting that reads a legally obtained copy of the original game data. Original game assets are not distributed with this project.
Milestone 0.10.2.2 extends the narrative/runtime interpreter: DBASE100 choices are now selectable and execute their original ordered branch, consecutive DBASE400/500 dialogue is queued instead of overwriting itself, consecutive GDV cutscenes are queued with their own subtitles, 25 structurally safe random blocks are executable, and the retail DBASE300 DisplayTexture (IMG1) path is decoded and displayed. Conditional hidden choices preserve their original branch ordinal. Complex random blocks remain observed rather than guessed.
Validation against the English retail data: 29/29 choice actions have balanced option/branch counts; 31 random blocks were found (25 flat/safe, 6 complex/observed); the retail DisplayTexture decodes to 320x200 / 64,000 indexed pixels with an embedded 768-byte RGB palette.
ROTH Unity now includes a native experimental GDV cutscene pipeline and the first narrative database playback. The engine can parse the retail 8-bit Gremlin Digital Video streams, decode the main type-8 delta codec plus the type-5 LZ variant used by HAWK03A.GDV, update the original VGA palette, decode PCM/DPCM audio, and present video fullscreen without adding a Unity UI package dependency.
DBASE100 cutscene records are now parsed alongside global actions. DBASE400 text/subtitle records and DBASE500 DPCM voices are read directly from the original installation. Global opcode 5 can display the original line and play its voice; opcode 7 can resolve a DBASE100 cutscene name and launch its GDV; opcode 8 exposes its original choice string text while full branching choice execution remains under development. Cutscene subtitles are supported with an experimental 8-tick-per-second clock inferred from the retail timing data.
The 0.8 gameplay work remains: whole-game RAW loading, automatic DEMO*.DAS resolution, runtime texture mutation, inventory/flags, map warps, SFX and safe DBASE100 callbacks.
- Open the project with Unity 6000.6.0f1.
- Choose ROTH Unity > Quick Start GOG.
- The first path checked is
C:\Program Files (x86)\GOG Galaxy\Games\Realms of the Haunting. - Select a discovered retail map and click Create playable map test.
- Enter Play Mode. Use WASD + mouse, Space to jump, left/right click to test original interactions, Esc to release the mouse.
Mirror World X should remain OFF for the corrected original orientation.
- ROTH Unity > RAW Map Inspector
- ROTH Unity > DAS Texture Inspector
- ROTH Unity > RAW Command Inspector
- ROTH Unity > DBASE100 Inspector
- ROTH Unity > Quick Start GOG
- ROTH Unity > GDV Inspector
The development HUD in Play Mode shows current map, Sector ID, inventory count, the latest trigger and the latest command reached.
The English retail data used as ground truth contains 44 .RAW maps. The extended structural validator currently passes 44/44 maps. A separate command audit passes all 44 maps with 5,568 commands, 1,950 entry references, zero unresolved entry references and zero invalid next-command indices.
The map-resource resolver obtains 100% primary-DAS coverage for most retail maps. A few maps contain one special reference not present in the selected FAT and continue with a diagnostic fallback. See Documentation/RETAIL_MAP_RESOURCE_AUDIT.md.
Across all retail maps, 127 command-59 warps were found. Every named target resolves to an installed .RAW except a single MAS5 reference from MAUSO1EA; no alias is invented without evidence.
This remains a development engine, not yet a complete replacement executable. Open Door and other geometry-mutating commands are traced but not visually executed until their behavior is verified. Player Rotation remains observation-only while its packed angle representation is being confirmed. DBASE100 global actions now have a partial safe interpreter plus dialogue text/voice and GDV playback. Choice strings are visible but branching choice execution is not complete. Combat, final inventory UI, full choice/random blocks, complete puzzle execution and save compatibility remain future work.
See Documentation/MILESTONE_0_9.md for details.
This repository contains no original Realms of the Haunting copyrighted assets. A legally obtained installation is required.
RAW opcode 7 animates floor/ceiling heights via RothSectorMover; see Documentation/MILESTONE_0_10_3.md. Timing and player carry remain experimental.
RAW sector movers now optionally return to their initial height after the configured timeout. Requires retail verification; see Documentation/MILESTONE_0_10_4.md.
Prototype automatic door obstruction prevention using RAW sector polygon footprints. See Documentation/MILESTONE_0_10_5.md.
Sector-mover countdown begins after reaching the open position; re-triggers preserve current position. See Documentation/MILESTONE_0_10_6.md.
Experimental CharacterController transport on moving floors and capsule clearance guard for floor/ceiling motion. See Documentation/MILESTONE_0_10_7.md.
Experimental RAW opcode 9 horizontal movement (X/Z), isolated vertices, sector objects and grounded rider carry, separate floor/ceiling/platform texture-follow flags, auto-return and auto-repeat. Retail audit of 32 unique opcode 9 commands included as Tools/validate_raw9.py. See Documentation/MILESTONE_0_10_8.md for limitations. Unity build not verified.
Added overflow-atomic sector translations, position-preserving motion retriggers, rider rollback and regression tests. This is not a proven retail-faithful implementation. See Documentation/MILESTONE_0_10_9.md.
Fixes small-step rider obstruction detection and deterministic position rollback. Adds opt-in structured RAW9 trace logs for later DOS comparisons. Experimental and not yet Unity-playtested. See Documentation/MILESTONE_0_10_10.md.
Read-only Unity log parser with per-sector JSON summaries, anomaly diagnostics, and eight synthetic Python tests. Does not establish DOS fidelity. See Documentation/MILESTONE_0_10_11.md.
Analyzed an original user-owned ROTH.EXE (SHA-256 e2d54427cd0692798e2df457b1ea8d3bca8cf1383ec2d4ea782165fe0ba56a05): confirmed 80386 MZ+LE image, 81 page mappings, three objects and Watcom startup entry. Added Tools/inspect_roth_le.py and 10 synthetic tests. See Documentation/MILESTONE_0_10_13.md. The command-9 DOS handler is not yet located; no retail binary is committed.
Validated 14,968 LE fixup records from the user's own GOG executable and traced the 128-entry RAW command dispatcher to opcode 9 (object 1 + 0x22A99). Identified initial movement update candidates. CORRECTED in 0.10.15: the 6-bit fractional speed path belongs to opcode 7, not opcode 9. This does not yet prove DOS timing or the Unity translation algorithm, which remain experimental. See Documentation/MILESTONE_0_10_14.md and Documentation/ROTH_OPCODE9_ADDRESSES.json. No proprietary game bytes are committed.
Original LE animation runner at object1+0x247CC uses the same relocated function table with index offset +67. This proves opcode 9 update = 0x22BD9, opcode 7 update = 0x22D51 and assigns the 6-bit fractional path to opcode 7, correcting earlier claims. The frame delta derives from a 16-bit counter, but real-time tick frequency is not established. Audit and 12 tests updated; Unity movement remains experimental and unchanged. See Documentation/MILESTONE_0_10_15.md.
Examined the original GOG opcode-9 update and shared animation epilogue, verifying 30 exact x86 sites plus the LE relocation to the global frame delta. Reconstructed the signed 16-bit countdown and a branch-accurate completion truth table for timeout, repeat-bit and phase flag. The new Tools/audit_roth_opcode9_states.py checks the user's local ROTH.EXE; Tools/test_roth_opcode9_states.py contributes 18 synthetic tests. See Documentation/MILESTONE_0_10_16.md and state reference JSON. The Unity mover remains experimental and unchanged, because real DOS tick frequency and full scheduler behavior are not yet verified. No commercial game bytes are in the repository.
Using the user-owned GOG ROTH.EXE, verified IRQ0 PIT channel-0 programming (initial divisor 0x4242, running 0x4216, nominal ~70.527 Hz) and the LE-relocated 16-bit counter at object3+0x20FAC. Verified signed coordinate targets and clamp logic in the real opcode-9 update 0x22BD9. Added read-only audit, 18 synthetic tests, isolated non-wired C# arithmetic reference and 6 static source checks; 64 offline Python tests pass. Unity mover is unchanged; no Unity compilation or runtime fidelity claim. See milestone report. No retail executable or assets are committed.
The read-only Tools/audit_roth_pit_timer.py adds 15 independent instruction-site checks and 9 LE relocation checks for the IRQ0 → signed-16 timer → RAW9 delta chain. Tools/test_roth_pit_timer.py has 15 additional synthetic tests. RothDosPitClock.cs is a standalone, unconnected reference for fractional tick accumulation; it does not alter gameplay. See supplementary PIT audit. This complements rather than supersedes the 0.10.17 arithmetic milestone.
Read-only Unity Editor and Python audits identify the four original face/vertex references selected for command 9. The DOS evidence audit validates 17 instruction sites and four relocations; sixteen synthetic tests pass. See Documentation/MILESTONE_0_10_18.md. No Unity runtime geometry changes or original game assets are included.
Added a read-only RothRaw9TopologyInspector, extended the Unity Editor RAW9 vertex audit with topology warnings, and included a Python auditor for original RAW maps. It detects selected vertices referenced outside the sector, sister-face portal edges, distinct external vertices at the same coordinates, invalid sister pointers and non-four-face sectors. 94 offline Python tests pass, including 14 new synthetic/static checks. This does not change runtime movement, and no real retail RAW map or Unity Editor Play Mode was tested in this session. See Documentation/MILESTONE_0_10_19.md.
A new UseSelectedRaw9Vertices checkbox in RothHorizontalSectorMover activates an experimental geometry backend. It updates the four original RAW9 vertex coordinates, refreshes mesh/collider, refuses risky topologies or malformed quads, prevalidates bounds, and preserves the previous motion backend when the checkbox is off (the default). Ongoing motions lock their backend; map rebuilds invalidate them. Sector objects follow as in the Unity prototype, which is not yet established as DOS-correct.
This is NOT yet DOS-faithful movement timing. Motion interpolation still uses Time.deltaTime and the provisional end-start distance. Python simulation tests cover safe coordinate changes, but no Unity Editor compilation or retail-map Play Mode checks were performed. See the 0.10.20 technical report. The new geometry backend intentionally rejects some maps pending portal and collision analysis.
To try it, attach or find RothHorizontalSectorMover on the map GameObject, enable Use Selected Raw9 Vertices before triggering command 9 and test a topology-audited four-face sector. Keep it disabled for the existing prototype.
The opt-in four-vertex RAW9 backend now has a second opt-in timing mode, UseNominalDosTickPacing (OFF by default and only available with UseSelectedRaw9Vertices). RothDosTickClock.cs approximates the verified IRQ0 PIT divisor 0x4216 (~70.53 Hz), accumulating whole ticks and limiting movement allowance to 45 ticks per Unity frame. The byte speed is multiplied by this integer tick delta rather than by seconds; the legacy mover remains unchanged when the checkbox is OFF. Python model tests and source checks passed, but Unity compilation, Play Mode, real-map behavior and full DOS command-state fidelity remain unverified. See Documentation/MILESTONE_0_10_21.md.
Added the UseDosSingleShotAbsoluteBounds inspector toggle to RothHorizontalSectorMover, OFF by default and requiring both optional four-vertex mode and PIT tick pacing. For single-shot commands with zero timeout and no auto-repeat, it plans absolute DOS endpoints from the original four selected vertices, then advances via the signed-16 original RAW9 step calculation. It rejects unmodeled retriggers and opposite-direction targets; the old experimental modes are unchanged. The original retail GOG x86 opcode-9 branches were rechecked against the user-owned executable. 16 Python model tests passed; 20 committed-source contract checks passed. Unity compilation, Play Mode, collision geometry and retail command lifecycle are not validated. See Documentation/MILESTONE_0_10_22.md.
A new UseDosCommandStateTransitions inspector option (OFF by default, requiring the three earlier opt-in DOS/geometry switches) preserves the original command modifier bit 0x02 and per-command state across triggers. The mover now models the x86-completion branches for a signed 16-bit wait timer, automatic direction reversal, and valid active repeat/zero-timeout re-triggers. RothCommandMonitor supplies the original one-based RAW command index and initial modifier; RothDosRaw9CommandState.cs isolates the verified bit transitions. A new RAW9_DOS_STATE trace records command/animation bytes, countdown and activation status. 61 local Python tests passed and original GOG binary signatures were reverified, but Unity Play Mode, C# compilation, real-map collision fidelity and full DOS scheduler integration are not yet verified. See Documentation/MILESTONE_0_10_23.md.
Fixed the original opcode-9 signed 16-bit countdown to preserve x86 JG semantics on overflow; added defensive bounds/ownership checks before reversing an active sector. New read-only Tools/audit_roth_raw9_playmode.py joins RAW9_TRACE, ROTH_PIT_TRACE and RAW9_DOS_STATE from a Unity Play Mode log, reporting step direction, tick-speed violations, waiting-state movement and missing timer data. Locally tested 44 Python cases (25 state, 19 log integration) and reverified the original user-owned GOG executable audit; Unity C# compilation, Play Mode and DOSBox comparison were not performed. The DOS movement remains opt-in and OFF by default. See Documentation/MILESTONE_0_10_24.md.