Synthetic data by game

Inventory performed in the Firestaff repository on 2026-08-08. This document distinguishes between:

1. synthetic data used only by tests and negative contracts;
2. diagnostic worlds or fallback routes that must not be used in production;
3. placeholder, procedural, or AI-generated material that can resemble real game data.

A test fixture should not be replaced with a copied original file merely because that file exists. A fixture often tests a constrained format error or a negative branch. When the same production route already has an authenticated source, however, that route must not be presented or verified using fixture data.

Summary

DM1

Findings

• src/dm1v2/ contains compatibility models that could previously create host-invented champion values, weather, particles, shakes, logs, transitions, and similar presentation. They are now inert/no-draw or bound to tests, but must not count as original data.
• docs/source-lock/dm1_v22_finished_art_material_gate_pc34.md documents the V2.2 placeholder/procedural-art gate. That package is not an authentic DM1 graphics source.
• tests/fixtures/minimal.DAT, parity-evidence/fixtures/, and DM1 *_fixture/*_probe programs are synthetic or contract-bound. They test parser, no-draw, and negative branches and must not be used as pixel proof.
• probes/dm1/firestaff_dm1_v1_original_fakewall_view_collision_probe.c and src/dm1/dm1_v1_viewport_fakewall_pc34_compat.c contain a diagnostic fake-wall route. It must not take precedence over an authenticated PC34 view.

Real source

• Original PC/DOS data exists in ~/.firestaff/data/dm1, including the canonical Dungeon-Master_DOS_EN_Version-34.zip with DATA/ members. Native checks must read those members in memory rather than relying on an extracted directory.
• Original PC34 saves exist outside the repository under ~/.firestaff/saves/dm1/original-pc34/ and in the user’s Downloads corpus. They must not be replaced with generated saves; C13 still requires an authentic save that actually contains the C13 event.
• The FM Towns original is available as the direct original Dungeon-Master_FM-Towns_JA-EN.zip container under ~/.firestaff/data/dm1/; Firestaff reads its DATA/ records in memory.

Decision

DM1 V1 must continue to consume verified GRAPHICS.DAT/DUNGEON.DAT and save bytes, or leave the material empty. V2.2 placeholder art must not replace original PC34 records merely because it is visually complete. The remaining pixel-pair and C13-capture gates are evidence work, not permission to create synthetic saves or screenshots.

The V2 movement/viewport real-data check reads DATA/DUNGEON.DAT from the canonical PC 3.4 ZIP using the native ZIP reader. It no longer accepts an extracted DUNGEON.DAT as its primary corpus or uses a fixture as the positive dungeon decode proof. Its V1/V2 side-by-side seed is now constructed from that decoded original dungeon state; the former positive fixture composition is not used by this real-data regression. The standalone side-by-side seed gate uses the same ZIP-backed state for its positive framebuffer, RGBA, and region checks; absence of the local archive is an explicit skip, never a fixture fallback.

The companion V1/V2 directional parity gate also decodes the original member itself and checks all cardinal directions from the authentic initial position. The former built-in entry-state fixture and its data-free screenshot probes were removed: V1/V2 viewport verification now requires source-owned dungeon bytes.

DM2

Findings

• examples/dm2_hud_widget_synthetic/ is explicitly a synthetic 1x1 fixture for HUD-gate testing. Its manifest generator is synthetic_test; its files are not DM2 graphics.
• DM2 V2/V2.2 has documented placeholder/procedural-art gates and scratch fixtures. They describe gate behaviour, not finished Skullkeep materials.
• Dungeon-loader, weather, overlay, and startup fixtures can use small synthetic worlds or input. They are acceptable for deterministic tests but are not real-data runtime evidence.

Real source

Original DM2 data is available as direct DOS, Amiga, FM Towns, and Macintosh archives under ~/.firestaff/data/dm2. Production probes must select the verified edition’s GDAT/DUNGEON records directly from the container when available and must not promote synthetic HUD or dungeon bytes to production.

Decision

Keep the HUD fixture because it tests the state gate. It must not be installed as a real art pack. V2.2 must remain closed until real source-owned material records and pixel verification exist.

CSB

Findings

• CSB V1 loader/world fixtures use small synthetic dungeon buffers for lifecycle, rescan, and negative tests.
• parity-evidence/verification/csb_v1_experimental_launch_intent_fixture.json is explicitly experimental and must not count as authentic launch or gameplay evidence.
• CSB V2.2’s former procedural/material fixtures and generated v22_inplace_cache.bin are test material. Production does not open them; V1’s verified F0128 result remains byte-for-byte preserved until an original CSB decoder with palette and pixel-parity proof exists.

Real source

Original CSB data exists under ~/.firestaff/data/csb, including Atari STX, Amiga ADF, and FM Towns archives. Real-data probes must use the selected original platform’s GRAPHICS.DAT/DUNGEON.DAT.

Decision

Fixtures may remain for parser and rescan contracts, but must be explicitly fixture-only. Procedural V2.2 art and generated cache files must not replace real CSB records.

Nexus

Findings

• scripts/generate_nexus_v1_fixtures.py creates synthetic DGN, DMDF, and FNXS-save files. scripts/fixtures/nexus_v1_save_synthetic.dat is therefore never a real Nexus save.
• docs/source-lock/nexus_v1_phase7_verification_suite_H0357.md documents these fixtures and an older synthetic fallback for parser/round-trip probes.
• tests/fixtures/ and *_fixture targets are for deterministic testing, not for filling in a missing Saturn source.

Real source

Original Nexus data exists under ~/.firestaff/data/nexus, including the Japanese retail CUE/BIN source, LEV*.DGN, SLEV*.BIN, SNDLEV*.SAL, and *.MNS.

Decision

Keep synthetic DGN/DMDF/save data for parser and negative tests, but never present it as a playable original. Real-data probes must use hash- and format-verified files from the Nexus directory. Saturn pixel and runtime capture gates remain open until an authentic capture exists.

Theron

Findings

• src/theron/theron_v22_modern_assets_pc34.c and adjacent V2/V2.2 material describe generated, procedural, or AI art as fixture-only. It must not claim to be a real Track 02 installation.
• src/theron/theron_v1_viewport_runtime_noop.c and fixture startup routes are deliberately no-op/capture-gated; they do not create a replacement world.
• Track 02 parser, descriptor, and runtime fixtures can contain synthetic byte shapes for negative and shape-bound tests.

Real source

The current local corpus under ~/.firestaff/data/theron contains the authentic US Track 02 BIN (TQUS02.bin, MD5 f23601102138f87c33025877767ebf76), the authentic JP Rev. 1 Track 02 BIN (TQJP02.bin, MD5 b7afb338ad31be1025b53f9aff12d73a), and a combined US/JP RAR with regional CUE and ISO/audio members. Real-data CTest coverage confirms the JP raw BIN loads all seven source dungeons; the US and JP raw media also reach their bounded startup routes. The supplied JP TQJP02End.iso is only a 149-sector, zero-filled stub and is rejected as dungeon content. The US CloneCD ZIP and the authentic original-runtime transition/presentation captures are not present in this local corpus; tests requiring those external artifacts skip rather than falling back to fixtures. Production claims about levels, items, champions, and bitmaps must remain bound to authenticated Track 02 records.

Decision

Keep fixture and no-op routes as clearly labelled tests. Do not replace open semantic or visual Theron gates with procedural art, AI upscaling, or generated roster/level data. The next step is source lock and authentic capture, not more synthetic content.

Rule going forward

New data may count as real game data only when its original source, format, hash/provenance, and runtime owner are documented. If an equivalent original exists locally, production code must read it or refuse to draw/load. Synthetic files may appear only in tests/, probes/, examples/, or explicit fixture documentation and must never be used as positive real-data evidence.