Firestaff Gap List — Historical Technical Analysis (2026-06-20)

Historical analysis — not the active backlog or release status. This long-form research record preserves individual findings and source locks from 2026-06-20 onward. Some rows deliberately describe superseded states, old paths, or work that is now out of scope. Do not use it to make a playability, platform, completion, or acquisition claim.

The active, reviewed sources are project status, platform status, missing functions by game, the root TODO-<game>.md files, and the completion matrix. Their current native real-media boundary overrides this record wherever they differ.

This document does not replace the active per-game TODO files. It remains a source-cited technical index for historical context and future research.

2026-09-29 platform boundary correction: DM2 PC-9821 support is removed at the user’s request. It has no catalog, startup, or runtime route. DM1 PC-9801, DM2’s separate PC-9801 demo, and X68000 remain preservation-only.

2026-08-12 Hint Oracle graphics correction: original HCSB.DAT is now strictly indexed as a four-segment archive from the real Utility Disk structure. The parser verifies its duplicated big-endian size table and exact file coverage before exposing a segment; the real English ST 2.0/2.1 archive proves spans 100, 29146, 1497 and 32 bytes. Segments 1 and 2 now follow the source-owned IMG2 expansion path, producing the Oracle’s 320×200 indexed surface and its 256×27 indexed font raster; segment 3 follows HINT001.C’s exact Atari RGB3-to-RGB4 conversion. This remains a resource boundary: graphical UI routing, segment-0 integer consumption, original text/page layout and frame/pixel parity are still open.

2026-07-31 correction: the isolated csb_hint_oracle_graphical_overlay module, which drew a hardcoded frame and 5×7 font rather than original Utility Disk pixels, was removed. HCSB.HTC parsing and source-verified text binding remain; a graphical Hint Oracle is open only with a source-owned UI route.

2026-08-06 Nexus S2D correction: older rows below that describe the real FONT256.S2D handoff as a 256-slot/16×16 flat-font parse or as a real-font framebuffer stamp are historical wording and are superseded. The current retail receipt uses the named S2D regions and exactly 242 authenticated 8×8 character-generator tiles. The 256 header codes, page/attribute character mapping, Shift-JIS glyph assignment and screen placement remain open; the flat 1bpp and section→glyph-range render paths are fixture-only.

Classification:

• FIXED — exists in main HEAD, source-locked.
• PARTIAL — partially implemented; some sub-items in main, others not.
• OPEN-BOUNDED — tractable, fits in a focused commit.
• OPEN-LARGE — would need a separate milestone (weeks, not days).
• BLOCKED-DATA — cannot close without real game data we don’t have.
• OUT-OF-SCOPE — explicitly out of Firestaff (e.g. modding tools).

A. Cross-cutting gaps

A1. Asset format coverage

Nexus status correction, 2026-08-08: The older BPX/BPK row above still describes PRS3 as an unidentified compression algorithm. The bounded DMWeb byte decoder now passes test_nexus_v1_bpk_archive against the verified retail MENU.BPK and decodes all 162 PRS3 surfaces to their declared pixel counts. The remaining gap is Saturn presentation ownership: VDP1 command order, VDP2/CLUT binding, menu placement and screen identity. No decoded host pixels are promoted to production drawing until those observations exist.

A2. Mapfile system

A3. Engine behaviour parity

A4. i18n / l10n (post v2.9.1)

A5. Tooling

A6. Build / CI

B. DM1 gaps (specific to game 1)

Source: docs/FINAL_GAPS.md (188 lines), docs/dm1/DM1_V1_BUG_AUDIT.md, dmweb Game Page for Dungeon Master, ReDMCSB decompilation.

B1. Capture-gap pairs (from prior TODO audit)

2026-06-21 live-runner note: docs/parity/tools/dosbox_capture_session.py --live now reaches DM1 PC 3.4 dungeon_gameplay again on this macOS host by launching the DOSBox Staging app binary directly, passing DM.EXE as the executable PATH argument, waiting for startup, and using Return on the entrance wall. The local run produced 01_ingame_start.png, 02_ingame_step_forward.png, and a movement receipt with a viewport-hash change after Keypad-5. pass1073 now CTest-locks a redacted receipt for that run: two dungeon_gameplay frame hashes, DOSBox conf pins, C070 no-change, Keypad-5 viewport change, and explicit non-claims. The live runner also accepts --post-dungeon-route Keypad-5:label,... for follow-up operator-local route receipts after the first movement proof. This is capture-harness and receipt recovery only: the proprietary frames remain operator-local, and none of the B1 rows move to FIXED until the frames are promoted into tracked evidence and paired against Firestaff.

B2. Per-domain DM1 gaps

B3. DM1 V2

C. CSB gaps

Source: docs/FINAL_CSB_GAPS.md (135 lines), 5 csb_gap_*.md files, greatstone g_csb.html (19 CSB/game + utility-disk extraction entries), dmweb CSB Game Page, and DMWeb’s CSB Atari ST edition page.

C1. Champions/mechanics/dungeon/graphics

C2. Per-version CSB asset coverage

C3. CSB hidden-code items

C4. CSB V2

D. DM2 gaps

Source: docs/nexus/NEXUS_PLAN.md (similar scope), greatstone g_dm2.html (15 extraction entries), skproject source (DM2 Windows port).

D1. DM2 V1 mechanics

D2. DM2 V2

D3. DM2 per-version coverage

DM2 Macintosh status correction (reviewed 2026-08-27): the English retail route is verified for bounded New Game/media/input, not complete Macintosh gameplay. Source 0x04 wall/floor timers are fail-closed until the complete DB3/DB14/DB0 GAME_LOAD transaction is owned atomically. The authentic Mac save corpus, complete GAME_LOAD/Resume owner and dynamic Control/Event pointer/drag owner are still absent; DOSBox SKSAVE files are not Macintosh evidence.

E. Nexus gaps (Saturn)

Source: docs/nexus/NEXUS_FILE_CLASSIFICATION.md, docs/nexus/NEXUS_PLAN.md, Nexus locally verified files in ~/.firestaff/data/nexus/.

E1. V1 phases 0-7

E2. V2 phases

F. Theron gaps (PC Engine / TurboGrafx-16)

Source: DMWeb Theron’s Quest overview and PC Engine / TurboGrafx-16 edition pages, docs/nexus/NEXUS_PLAN.md (similar shape), Theron local probes.

F1. V1

F2. V2

G. Launcher / Settings / Accessibility (cross-cutting)

G1. M12 launcher

G2. Touch / controller

G3. Accessibility

H. Cross-spiel prioritized work order

For the next sprint (post v2.9.1, pre v2.10.0), I propose this order:

Tier 1 (BLOCKED-DATA — surface via tooling, not code)

1. Document real-data acquisition checklist: which Atari ST / Amiga / SNES / MegaCD / Saturn / PC Engine binaries need to be sourced per game, with hashes that will gate future runs of compare_to_greatstone.py. — DONE (docs/DATA_ACQUISITION_CHECKLIST.md with per-game ✅/🟡/🔴/⚪ status matrix; generated by python3 tools/asset-validate/compare_md5_to_sha256.py — current coverage: 148/148 registry entries present + hash-correct).
2. Add compare_to_greatstone.py summary mode that prints a per-game “data gap” view (which files in VERIFIED_HASHES.md are not on disk). — DONE (new --summary mode in tools/asset-validate/compare_to_greatstone.py: prints per-game TOTAL/FOUND/MISS table from any data root; default to ~/.firestaff/data). Run: python3 tools/asset-validate/compare_to_greatstone.py --summary. Companion tool tools/asset-validate/data-readiness-summary.py (Tier 1 #2 L2, FIXED 2026-06-22 commit a56d79c70 → main 22a8caa3) builds a per-game readiness table by combining --scan-data, the --summary hash table, and an opt-in --boot-probe. See L2 below for full description.
3. Verify all --scan-data READY-path:er are actually launchable by M11. — DONE for Tier 1 path-discovery scope 2026-06-21: CTest tier1_strict_boot_probe PASS for all present in-scope launch paths (DM1 canonical, DM1 legacy-dos, CSB canonical, CSB Amiga 3.3 Meynaf FR, Theron JP canonical, Theron JP extras, Theron US extras), and asset_validate_coverage_by_game PASS. Nexus virtual-ISO launch remains out-of-scope here and tracked as a Tier 4 runtime/launcher gap, not a path-discovery gap.
4. Scanner path-naming limitations: CLOSED as NO-GAP 2026-06-20. The scanner already matches on MD5 via asset_find_by_md5 and scan_iso_by_md5 now also falls back to whole-file MD5 for non-ISO-9660 .bin files (Theron Track 02, raw Saturn tracks). Verified: --data-dir ~/.firestaff/data/nexus-extras/saturn-ja reports Nexus READY (FOUND ...Track 1).bin::DM.BIN) and --data-dir ~/.firestaff/data/theron-extras/{japan,usa} reports Theron READY (FOUND ...Track 02).bin). See reference/L1_data_path_verification_2026-06-20.md Results v2 for verification data.
5. Mirror dmweb /community/documentation/ for offline research — DONE 2026-06-20 (commit pending; 43 pages mirrored at reference/dmweb-community-docs/ with INDEX.md + index.json + SCRAPE_LOG.md; reproduceable via crawl.sh). Eliminates the “free.fr intermittent 404” risk for 5 topic areas.
6. Ship a reproducible game-archive extraction script that pulls from ~/<downloads>/ to ~/.firestaff/data/<game>-extras/ without overwriting canonical staging — DONE 2026-06-20 (commit 4b097f54; 73 archives → 71 version directories; ~6.2 GB extracted; 4 new READY paths discovered by --scan-data).

Tier 2 (OPEN-BOUNDED — fits one commit each)

1. CSB hidden-code skip table for Atari ST + Amiga items 558-562. — DONE (commit c5897fce feat: extend CSB hidden-code skip table to 12 entries (CSB Atari + Amiga specific) + earlier a8033a47 feat: extend CSB hidden-code skip table to 12 entries). 12 entries total: 4 Atari/Amiga executable rows (558-562), 2 kid dungeon rows (135-138), 3 CSB Atari ST (21/538/548), 3 CSB Amiga (21/676/686). Source-locked against Meynaf disassembly + 4 COD1/COD2/COD3/COD4 container formats.
2. LZW decoder for Atari ST GRAPHICS.DAT (only Atari ST uses LZW). — DONE for the decoder: m11_gfx_lzw_decompress has a contract-only round-trip test (test_dm1_lzw_round_trip.c, dm1_lzw_round_trip PASS 1/1 on 2026-06-21, pass852). DMWeb’s DM Atari ST edition page now documents the expected original media set and per-version split (two English 1.0 builds, English 1.1/1.2, German 1.2, French 1.3); real Atari ST asset handoff still BLOCKED-DATA.
3. PAK container decoder for Atari ST START.PAK. — DONE (commit 3ee479de); real-media proof should keep DMWeb’s cracked ST/MSA, hard-disk, and RamDisk hack derivatives separate from canonical STX originals.
4. CMP portrait loader for CSB utility disk. — DONE (commit 532c8250); firestaff_cmp_decode_unit + csb_v1_cmp_import_pc34 PASS 2/2 on 2026-06-21. Runtime champion portrait source-selection remains tracked separately under A3, not as a CMP format decoder gap.
5. Harmonize MD5 vs SHA256 in asset_find_by_hash.c (or add wrapper). — DONE (commit 5988b620, see docs/MD5_SHA256_HARMONIZATION.md)
6. _G2157_ linker fix (add image_frontend_pc34.c to firestaff_m10 source list). — DONE (commit 3588798f, added image_backend_pc34_compat_globals.c instead which provides the same symbols without dragging in the legacy frontend)
7. Lefthook-in-CI install step. — DONE (.github/workflows/verify.yml:227-253 installs Go + lefthook via go install github.com/evilmartians/lefthook@latest, then runs lefthook run ci for po_layout + hash_harmonization as CI sanity gate).
8. --data-dir override in m12_build_search_roots — DONE 2026-06-20 (commit 6a7eccdc). Explicit --data-dir no longer silently falls back to ~/.firestaff/data; runtime dataDir-override also skipped when request is explicit. 5/5 asset-status tests pass; tested with --data-dir ~/Downloads --scan-data → reports Data dir: <local-home>/Downloads and scans only that directory.

Tier 3 (OPEN-LARGE — separate milestones)

1. FTL container decoder (Amiga, X68000, MegaCD). — PARTIAL: generic header/hunk/checksum parser is CTest-gated, and the bounded HUNK_DATA area_1 zero-run decompression (greatstone d_ftl.html “Note 7”) is now CTest-gated via firestaff_ftl_hunk_data_zero_run_unit. 2026-06-25 X68000 <-> FTL handoff added: firestaff_x68k_ftl_handoff_unit PASS 5/5 cross-module test verifies the FTL-declared data_area1_memory_size fits inside the DMWeb 1232 KB 2DHD disk; companion firestaff_x68k_media_classify_unit PASS 13/13 covers the DMWeb X68000 geometry constants, copy-protection-sentinel detection, and configurable FTL-magic scan window. HUNK_CODE 0x5223 decompression/checksum verification, mapfile item extraction, real FTL corpus verification, and runtime loading remain.
2. CSBWin custom resource path (csbgraphics.dat + dmsave).
3. DM2 advanced CCM + projectile-list drain.
4. DM2 per-cell modern-art swap (T560/T600).
5. Nexus per-cell modern-art swap (Saturn draw pipeline).
6. Theron V2 phases 3-7 (UI overlays, lighting, smooth movement, touch, verification).
7. CSB V1 graphics ornament blits (F0108, F0115, F0111, CustomBackgrounds).

Tier 4 (per-game polish, partially started)

1. CSB mechanics parity (combat, dungeon, champion, inventory).
2. DM2 mechanics (shops, NPCs, triggers, timeline). — Shop-economy determinism probe (firestaff_dm2_v1_shop_economy_determinism_probe, ctest dm2_v1_shop_economy_determinism_probe, commit a89c257f): 19/19 PASS covering reset/enter/buy/sell/leave state transitions, party-state hash preservation across leave, 50-repetition determinism, wrong-shop rejection, insufficient-gold path. Source-locked per skproject/SKULLWIN/c_shop.cpp transaction pricing
   • skproject/SKWinGlobal.h:42 (NUM_NPCS=4). Existing coverage already includes firestaff_dm2_v1_shop_probe (built-in shop catalog + NPC dialog), firestaff_dm2_v1_trigger_probe (8 triggers + 4 kinds + 6 targets), and firestaff_dm2_v1_timeline_probe for the three other DM2 mechanics domains.
3. Nexus runtime/probe coverage beyond compile/save-load.
4. Theron cross-slot import/export + cross-route evidence. — Dungeon-progression determinism probe (firestaff_theron_v1_dungeon_progression_determinism_probe, ctest theron_v1_dungeon_progression_determinism_probe, commit 8a43fcf8): 20/20 PASS covering the 7-dungeon progression state machine (THQUEST.ASM T080 between-dungeon save/load), init / advance / quest-complete transitions, NULL-safety, and 50-repetition determinism. Source-locked per ReDMCSB GROUP.C analogue + THQUEST.ASM T080. Between-dungeon save/load itself is already covered by tests/test_theron_v1_save_header_rejection.c + the existing Theron V1 mechanics probe (10 tests / 5 probes already green for the V1 gameplay loop).

Tier 5 (i18n follow-up)

1. Native-vs-fallback separation in validate_po_layout.sh. — DONE: validator now reports nonblank and native coverage separately, with fallback-only catalogs marked FALL without failing the structural CI gate.
2. Fill DM1/nexus/csb/theron/firestaff missing native translations with native speakers (out of scope for AI).

Tier 6 (launcher/accessibility)

1. Save export/import, session timer, manual launcher.
2. Touch gesture navigation.
3. Screen reader integration. — FIXED 2026-06-27/29 + fallthrough-anchor gate 2026-07-01: launcher-state manifest covers main view / settings / missing-data popup plus the four cell-by-cell extras views (bestiary, item encyclopedia, screenshot gallery, museum of lore) with stable element IDs (probe PASS 175/175, up from 69/69), and M11 gameplay manifest coverage now has a headless probe for gameplay / inventory / map / dialog / entrance mirror / endgame / open chest surfaces. Subtest N pins the navigation-anchor contract for the 12 fallthrough launcher views (data-validator / audio-settings / accessibility / theme / save-browser / input-remap / custom-dungeon / campaign / spell-reference / map-viewer / touch-layout / presentation-preview) — each view emits GAME_CARD_DM1 / MENU_SETTINGS / MENU_MUSEUM / GAME_CARD_THERON, uses the active-view gameState prefix, and never leaks a foreign view’s markers. Remaining follow-up: manual-docs + changelog already have dedicated emitters; the 12 fallthrough views still intentionally fall through to main-view emission as a navigation anchor (now machine-checked), but each one is a candidate for a dedicated emitter if its UI gains a per-cell or per-row surface; plus real assistive-technology smoke evidence beyond JSON shape.

I. Sources surveyed (this round)

Total new pages reviewed this session: 66 (23 prior + 43 from the dmweb community/documentation mirror).

J. Update cadence

This document should be regenerated after every milestone commit. Each major section (B-J) is regenerated from its per-domain source-of-truth:

• A → po/validate_po_layout.sh + tools/asset-validate/compare_to_greatstone.py --list
• B → docs/FINAL_GAPS.md + docs/dm1/dm1_gap_*.md + docs/shared/ai_*.md
• C → docs/FINAL_CSB_GAPS.md + docs/csb/csb_gap_*.md
• D → docs/nexus/NEXUS_PLAN.md (DM2 sections) + src/dm2/ headers
• E → docs/nexus/NEXUS_FILE_CLASSIFICATION.md + src/nexus/ headers
• F → Theron TODO sections + src/theron/ headers
• G → TODO.md cross-cutting sections
• H → manual prioritization after the above regenerates

Avoid duplicating content; this doc is an index, not a source.

K. Session delta — 2026-06-20 (post v2.9.1)

What changed in this session that affects the gap list above:

Data staging

• 73 game archives extracted from ~/<downloads>/ to ~/.firestaff/data/ via reference/extract-game-archives.sh.
  • 71 new version-staging directories under <game>-extras/<version>/ (~6.2 GB extracted)
  • 4,232 extracted files in addition to existing canonical staging
  • No existing canonical files (dm1/, csb/, dm2/, nexus/, theron/) were overwritten

New READY paths (in addition to canonical)

./build/firestaff --scan-data now finds alternative matching paths for four of the five games:

Kanoniska hashar (VERIFIED_HASHES.md)

tools/asset-validate/compare_to_greatstone.py ~/.firestaff/data runs against all 12 registry entries: 148/148 OK, 0 FAIL (all files referenced in VERIFIED_HASHES.md are local and match their hashes; some map to the same file, hence 148 versus 12).

Documentation mirror

• 43 dmweb.free.fr /community/documentation/ pages mirrored locally at reference/dmweb-community-docs/ (5.5 MB)
  • INDEX.md (19 KB) — human-readable table of contents
  • index.json (27 KB) — machine-readable
  • crawl.sh — reproducible crawl script (curl + 1.2s rate limit)
  • Covers 5 topic areas (copy protection, DM+CSB mechanics, Nexus file formats, file formats, miscellaneous)

Bug fix: CLI –data-dir

• src/shared/asset_status_m12.c::m12_build_search_roots — explicit --data-dir now overrides default fallbacks in --scan-data mode. Previously the flag was ignored and the scanner reported FOUND paths from ~/.firestaff/data/ even when the user requested another root.
• m12_fill_game_versions — the runtime dataDir override is skipped when requestedDataDir is explicit (defensive against future fallback additions).
• tests/test_asset_status_scan_metrics.c — updated to expect rootCount=1, duplicateRootSkips=0 when --data-dir is set.
• Committed: 6a7eccdc.

Test coverage (smoke runs)

• 40/40 DM1 V1 chest/item/weight/recompute-tester PASS
• 23/23 Phase A invariants PASS
• 12/12 canonical hashes match (0 failures)

Effect on gap status

Marked above as EXTRACTED (new rows in C2, D3, E1, F1), previously BLOCKED-DATA. Four are marked EXTRACTED + VERIFIED because they now also match a canonical hash. Update 2026-06-21: Theron JP/US Track 02 is also launch-tested through tier1_strict_boot_probe.

These can now be used as real-asset test sources in addition to existing canonical staging under dm1/, csb/, etc.

L. Follow-up — concrete next-session tasks

The list below contains prioritized small tasks directly following from the session deliverables. Each item takes less than one day and requires no new design.

L1. Verify that alternative READY paths boot

Status 2026-06-21: DONE for the Tier 1 path-discovery scope through tier1_strict_boot_probe (all present in-scope launch paths). CSB canonical and CSB Amiga 3.3 Meynaf FR now emit CSB READY; Nexus virtual-ISO launch remains a separate Tier 4 runtime/launcher gap.

The original Tier 1 #5 request: run against every EXTRACTED + VERIFIED path and confirm that M11 actually starts the game, not merely that the scanner finds hashes.

for spec in \
  "dm1   ~/.firestaff/data/dm1-extras/legacy-dos" \
  "csb   ~/.firestaff/data/csb-extras/legacy-amiga-dms" \
  "nexus ~/.firestaff/data/nexus-extras/saturn-ja" \
  "theron ~/.firestaff/data/theron-extras/japan"
do
  set -- $spec
  game=$1
  path=$2
  echo "=== $game @ $path ==="
  SDL_VIDEODRIVER=dummy ./build/firestaff --data-dir "$path" --game $game --duration 1000 2>&1 | tail -5
done

Expected: a Phase A probe PASS plus a game-specific asset-load PASS per path. If anything fails, move the gap status back to PARTIAL.

Result (2026-06-20): See reference/L1_data_path_verification_2026-06-20.md for details. Short version:

New status:

• DM1 and CSB legacy paths are now EXTRACTED + VERIFIED + LAUNCH-TESTED (ready for future tests/parity evidence).
• Theron JP/US Track 02 paths have been EXTRACTED + VERIFIED + LAUNCH-TESTED since 2026-06-21: tier1_strict_boot_probe starts JP canonical, JP extras, and US extras to the TQR level-load milestone.
• Nexus container paths: --data-dir <path> finds them correctly through MD5 matching (asset_find_by_md5), not filenames. Source filenames such as Dungeon Master Nexus (Japan) (Track 1).bin are accepted directly. The previous claim of a filename problem was incorrect.

Tier 1 #6 is closed as NO-GAP (2026-06-20) — verified that the scanner matches by MD5 hash, not filename. --data-dir accepts source filenames directly. Tier 1 #6 was raised by the L1 report, but the actual scan behaviour supports READY for all four paths. No alias step is required. The Tier 1 #6 entry above is outdated and should be removed at the next watchdog refresh.

L2. Create tools/data-readiness-summary.py — FIXED 2026-06-22

Tier 1 #2 –summary mode. Prints a per-game table: game / required-files-present / found-in-canonical / found-in-extras / launch-tested. Takes input from compare_to_greatstone.py and a manifest reader.

Output (example):

dm1   2/2 present   2/2 canonical   1 extra (legacy-dos PC34)  NOT launch-tested
csb   2/2 present   2/2 canonical   1 extra (Amiga 3.3 Meynaf FR) NOT launch-tested
dm2   2/2 present   2/2 canonical   0 extras                       LAUNCH-TESTED
nexus 1/1 present   1/1 canonical   1 extra (Saturn JA Track 1)     NOT launch-tested
theron 1/1 present  1/1 canonical   JP+US extras Track 02           LAUNCH-TESTED

Wire it into CMakeLists and the ci: asset-hygiene job. Use it on every git push to quickly see whether anything blocks an M12 launch.

Status 2026-06-22 (commit a56d79c70, cherry-picked to main as 22a8caa3): Shipped at tools/asset-validate/data-readiness-summary.py (342 lines, executable). Combines three sources into one human + JSON table:

1. firestaff --scan-data against a data root (~/.firestaff/data/ by default — covers canonical + extras).
2. compare_to_greatstone.py SHA-256 hash-match summary against VERIFIED_HASHES.md.
3. Optional --boot-probe (firestaff --game X --data-dir Y, ~8s per path) — opt-in only, since it’s slow.

Output: human-readable per-game table on stdout, JSON dump with --json, exit code 0 iff all 5 games canonical-READY, 1 otherwise. Verified running example captured in the commit message.

Remaining (out of L2 scope): not yet wired into CMakeLists or the ci: asset-hygiene job. --boot-probe mode is opt-in and unverified on a CI runner. A future pass should add a CMake target + CI step that runs --json and posts a status check on each push.

L3. Extend extract-game-archives.sh with a verification step

After extraction, run compare_to_greatstone.py for each extras directory and report a per-version summary. Log it in .extract-log.md and .extract-manifest.json. This lets future runs immediately show which versions match a canonical hash and which are reference material only.

L4. CSB Amiga 3.5 launch barrier

The extracted csb-extras/amiga-3.5-ctraw-en contains CTraw track and PACK records (.st/.raw/.err/.out). Firestaff currently validates the container framing and packed-payload CRCs, but does not decode PACK data or reconstruct game sectors. Renaming or repackaging .raw as .DAT/.DATA cannot produce playable assets. Keep this edition unavailable until a documented, bounded PACK decoder can recover and verify the required source data; then add an authentic startup test. Do not list Amiga 3.5 as launchable based on extraction alone.

L5. DM2 extras launch-test

Status 2026-06-21: DONE for PC extras. dm2_v1_boot_scan_assets now accepts the extracted DOS layout data/graphics.dat + data/dungeon.dat, and tier1_strict_boot_probe runs DM2 canonical plus dm2-extras/dos-en, dm2-extras/dos-fr, dm2-extras/pc-fr, and dm2-extras/pc-de through structured boot receipts. The earlier DM2 READY stderr marker was removed on 2026-08-06 because it is not source-owned UI. The remaining DM2 version work is in D3: demo and non-PC editions need separate classification and possibly container/format normalization before they can receive cross-version regression coverage.

L6. README-public-dokumentation-uppdatering

Per AGENTS.md, the README must be honest, user-facing, and sales-friendly, with real per-game status. After today’s gap-list updates, the README’s DM1/CSB/DM2/Nexus/Theron tables should reflect that four of five games have real-asset evidence in both canonical and extras staging.

M. Session delta — 2026-06-21 (post v3.0.0)

What changed in this session that affects the gap list above:

DM1 V1 original-capture gap close (pass1052-1058)

• 6 B1 capture-gap pairs moved from BLOCKED-DATA to PARTIAL:
  • I34E keyboard buffer transcript (pass513 SCAFFOLD_ONLY_MISSING_ORIGINAL_RUNTIME_DEBUG_FIELDS)
  • Paired original viewport screenshot (pass1052 + pass1056 CTest gate)
  • Paired original wall screenshot (pass1052 + pass1054 exact 0-pixel match)
  • Paired original collision transcript (pass1055 closed-door stasis)
  • Paired original champion-panel screenshot (pass1053 candidate/resurrect)
  • Paired original creature-chain screenshot remains BLOCKED-DATA (level-1 target behind inert closed door) but pass1058 locks the corrected keypad mapping.

DM1 V1 gap cascade (pass1059-1070)

• 12 B2/B3 PARTIAL rows moved to FIXED: portrait sensor parity, AI pathfinding, AI perception targets, AI reactions, mirror stat, C25/C26 projectile fallback, touch zones, inventory route parity, chest scroll-wheel pickup overflow, object consumable use, V2 smooth interpolation, V2 HUD rune routes, V22 material routing.
• 2 PARTIAL rows remained: chest runtime detail coverage, creature grouping/coordination (now PARTIAL after per-route audits).

Tier 1 #5 strict boot-probe

• New firestaff_tier1_strict_boot_probe ctest entry runs the launcher with --game <id> --data-dir <path> --duration 1500 under SDL_VIDEODRIVER=dummy for every EXTRACTED + VERIFIED path.
• All present in-scope paths PASS: DM1 canonical, DM1 legacy-dos, CSB canonical, CSB Amiga 3.3 Meynaf FR, Theron JP canonical, Theron JP extras, Theron US extras.
• Nexus (Track 1.bin::DM.BIN mount) remains tracked as a Tier 4 runtime/launcher gap, not a Tier 1 path-discovery gap.

Tier 2 #4 LZW Atari ST decoder DONE

Decoder code path test-covered (test_dm1_lzw_round_trip 96/96 PASS, pass852). DMWeb’s DM Atari ST edition page now documents the source-media/per-version split for the future real-asset handoff. Real Atari ST asset handoff still BLOCKED-DATA.

Tier 4 determinism probes (3 new)

• Theron V1 dungeon-progression (THQUEST.ASM T080) — DONE
• CSB V1 champion-stat (F0306/F0309/F0310/BUG0_72) — DONE
• Nexus V1 creature-state (F0209 timeline) — DONE

Asset-status fix

required=1 for all required-files rows. matchAnyVersion now propagates matchedPath so the missing-files popup and report show where the runtime will load the asset from, while keeping launch_blocker honest.

DM1 V2 polish

• V20 filtered renderer probe
• V21 upscale renderer probe
• V22 in-place render probe (CSB + DM1 Apple Silicon + DM1 V22 modern asset)
• Side-by-side V1/V2 presentation-disabled seed gates
• V22 in-place cache wiring through pass376 overlays

The V22 in-place drawing pipeline still uses placeholder overlay; wiring m11_draw_dm1_* draw passes to consume real modern art in-place remains OPEN-LARGE in B3.

Documentation

• 100+ row status changes in B1-B3/C1-C4/D1-D2/E1-E2/F1-F2/G1-G2 reflecting post-pass1052-1070 reality.
• Tier 1 #5 marked DONE for path-discovery scope.
• Multiple Tier 4 entries closed (Lefthook CI, CSB CMP decoder, Atari ST PAK decoder, CSB hidden-code skip, LZW Atari ST decoder, M12 extras DM1, chest runtime detail, creature grouping, Theron extras launch-tested, Theron Track 02 launch).

Verification

• ctest baseline 700+/700+ green (was 692/696 at v2.9.2).
• Phase A probe 23/23.
• Audio probe green.
• Strict -Wall -Wextra -Werror warnings-check green.
• Cross-platform determinism green.
• M10 verify green.

Commits since v2.9.2

116 commits, summarized:

• pass1052-1058 DM1 V1 original-capture gap close (B1 capture-gap pairs)
• pass1059-1070 DM1 V1 gap cascade (B2/B3 PARTIAL → FIXED)
• tier1-5 strict boot-probe per path
• tier2-4 LZW Atari ST decoder
• tier4-17 CSB V1 champion-stat determinism probe
• tier4-19 Nexus V1 creature-state determinism probe
• tier4-20 Theron V1 dungeon-progression determinism probe
• dm1_v2_inplace_render_gate / dm_v20_filtered_renderer_silicon / dm_v21_upscale_renderer_silicon / dm_v22_modern_renderer_silicon (V2 polish)
• dm1_v2_side_by_side_seed_gates (V1/V2 presentation-disabled)
• m11_capture_route_state / dm1_v1_wall_collision_runtime_capture / m11_turn_viewport_orientation (B1/B3 Firestaff-side gates)
• m11_v22_inplace_draw_pc34 (V22 in-place cache wiring)
• firestaff_dm1_v22_inplace_render_probe (CSB V22 in-place render probe)
• Asset-status fix + m12_fill_game_versions TIER1DEBUG cleanup
• gap-list updates: B1 PARTIAL closure, B2 PARTIAL → FIXED cascade, Tier 1-4 closes
• tools/dm1_24h_readiness.py expansion (12/12 ctest rows in the roll-up)
• verify_pass623_dm1_v1_input_capture_readiness_bridge.py line-range refresh
• tools/verify_pass352_dm1_v1_movement_route_regression_matrix.py token/keypad aliases fix
• pass372 rebuild target fix
• src/dm1v2/dm1_v22_shapes.c -Wunused-parameter warning fix
• 22a8caa3 (2026-06-22) — tools/asset-validate/data-readiness-summary.py cherry-picked to main from csb-v1-hidden-skip-cmp-real-asset-2026-06-20 (commit a56d79c70). Closes Tier 1 #2 L2. The other 6 subagent branches’ commits were audited file-for-file against origin/main and were subsumed by newer in-main versions; only this one had substantively new content. dm1v1-capture-gap-close-20260620, csb-v1-hidden-skip-cmp-real-asset-2026-06-20, dm1-b3-v2-gates-20260621, dm1-lane-a-original-evidence-20260621, dm1-lane-c-gap-audit-20260621, dm1-lane-d-readiness-20260621, dm2-v1-mechanics-parity-2026-06-20, main-cmake-fix — all left in place on origin as historical branches; the 15 worktrees have been removed and only workspace-main + the main-pass1052 worktree remain locally.
• 363bf3b9, cd86d520, b7dbcd60, a0592d6d, 393d9f64 (2026-06-21, on origin/main post-cherry-pick) — Theron 24h readiness cascade: raw Track 02 bank anchors locked, m11 runtime command proof, runtime screenshot readiness gate, 24h readiness rollup tool, and readiness-report refresh. Closes F1 cross-route mechanics runtime evidence (CTest path) and adds the Theron 24h readiness row to the per-day PASS/FAIL list. 152c6a8a release: prepare v3.0.1 then tags the cascade as v3.0.1.

Migration to GitHub main

dm1v1-capture-gap-close-20260620 branch fast-forwarded into origin/main (52bce320 → cd24ea72, 67 commits). The branch is now redundant with main.

N. Session delta — 2026-06-23

Latest origin/main status that affects this gap list:

• DM1 Hall of Champions C127 portrait coverage gained two narrower runtime probes on main: ordinal 0/DAROOU D1C rectangle ownership (9d45d2ce) and HALK/ordinal 1 front-north-entry rectangle placement (9f1ca771). These are regression gates for the already-FIXED Hall portrait rows, not a new gameplay claim.
• CSB launch-readiness verifier anchors were refreshed after M12_StartupMenu_GetLaunchIntent() extended to line 7513: csb_v1_m11_runtime_capture_boundary (2c17264a) and pass547_csb_v1_runtime_readiness_backfill (e8c3cb39) both pass, including pre-push and GitHub Actions on main.
• Focused queue refill: 100 new jobs were added from this gap list without reusing commands already recorded in the pool’s seen-hash file. Distribution: 50 DM1, 30 CSB, 20 Theron. The batch is biased toward runtime/capture/readiness/viewport/save/HUD/overlay gates for the active PARTIAL/OPEN rows, not toward already-closed generic validators.
• At the time of this update, the first 15 focused DM1 jobs in that batch had completed successfully, leaving 85 focused jobs still active in the pool. Historical .failed* files from earlier runs remain in the pool directory and are not counted as new failures for this batch.

N+. Session delta — 2026-06-25 (greatstone db_data real-data regression)

• A5 “Real-data regression tests (greatstone db_data)” row moved from OPEN-BOUNDED to PARTIAL / OPEN-BOUNDED with a CTest-gated OFFLINE/ONLINE metadata probe. tools/verify_greatstone_db_data_paths.py (greatstone_db_data_paths_probe CTest) replays a 13-row fixture (10 current db_data/ paths + 3 obsolete c_dm_* / c_csb_* / guessed-DM2 404-regression paths) using only status, content-type/length hints, <title> text, title SHA-256, and db_data/ link count. --online --write (or FIRESTAFF_GREATSTONE_PROBE=1) refreshes the evidence + fixture in one step, drops the bounded byte-range GET body after extracting metadata, and never persists copyrighted bytes. tools/asset-validate/no_game_data_in_git.py PASS. Evidence at parity-evidence/verification/greatstone_db_data_paths_probe/manifest.json; fixture at tests/fixtures/greatstone_db_data_paths/index.json. docs/PLATFORM_MATRIX.md now references the probe in the 🔵 status legend and the “See also” list.

O. Session delta — 2026-06-25 (DM2 original-overlay scaffold)

Bounded new evidence for the DM2 original-overlay OPEN-BOUNDED row; no parity claim.

• scripts/dosbox_dm2_original_overlay_capture.sh — DM2 PC 1.0 EN DOSBox capture harness with deterministic DOS4GW/4DOS config (memsize=63, fixed cycles=3000, no dynamic core, svga_paradise), Swift (macOS) and xdotool (X11/Linux) route injectors, labelled shot tokens, 320x200 rawshot normalization (including 640x400 2x DOSBox downsample to 320x200), and the retail-RAW4 RECT_7 224x136 viewport crop at (0, 40). Refuses to inject an unvalidated DM2 route. Outputs land in verification-screens/passH2313-dm2-original-overlays/ with deterministic naming and SHA-256 receipts.
• tools/verify_dm2_v1_original_overlay_capture_source_lock.py — read-only source-lock verifier against SKULLWIN (10 anchors across dm2global.h, c_gfx_main.cpp, c_gfx_main.h, c_gui_vp.cpp, c_tmouse.h, types.h, c_input.cpp, c_gui_draw.cpp). SKULLWIN is auto-discovered from the historical machine-specific paths, repo-relative fallbacks, or an ignored read-only reference/skproject/SKULLWIN checkout. It is a build reference only, never a runtime or release dependency. CTest-gated as dm2_v1_original_overlay_capture_source_lock. 2026-06-29 adds an optional pair-manifest gate for dm2_original_firestaff_pair_manifest.tsv: missing means OPEN/no claim, malformed tracked rows fail, and promotion requires same-state dungeon_gameplay original+Firestaff 224x136 viewport SHA-256s.
• probes/firestaff_dm2_v1_original_overlay_capture_scaffold_probe.c + CMakeLists wiring — data-free source-lock probe covering screen size 320x200, backbuffer 224x136 = 0xe0x0x88, mouse+command queue lengths 10/10, c_evententry {b,x,y} layout, surface=64000B/backbuffer=30464B, right-panel geometry, HUD text fit, click-route disjointness, shot-label vocabulary, route-token inventory, and the pair-manifest schema. CTest dm2_v1_original_overlay_capture_scaffold_probe PASS.
• docs/FIRESTAFF_GAP_LIST.md D1 row “Original-overlay proof” — status stays OPEN-BOUNDED; the bounded scaffold + verifier + probe are listed as new evidence so future sessions do not re-investigate. No FIRESTAFF_GAP_LIST status flip.

P. Session delta — 2026-06-25 (data-dir picker / platform-matrix)

Latest status that affects the data-dir picker ↔ platform matrix contract:

• New data-free regression firestaff_data_dir_picker_platform_matrix_probe (ctest data_dir_picker_platform_matrix) added under probes/. It mirrors the tier1_strict_boot_probe sub-layouts as a synthetic data root (dm1/, dm1-multilingual/, dm1-extras/legacy-dos/, csb/, csb-extras/legacy-amiga-dms/, dm2/, dm2-extras/{dos-en,dos-fr,pc-fr,pc-de}/, theron/Theron's Quest (Japan) (Track 02).iso, theron-extras/{japan,usa}/track02.bin) and locks: empty-root no-availability + single explicit search root + zero duplicate-root skips; DM1/CSB/DM2/Theron availability under the documented M12 test setters (Nexus stays unavailable without a DM.BIN); per-game first-matched version determinism; asset_find_by_md5 recursive discovery for every nested leaf; negative-discovery rejection of an unknown hash; re-scan clearing availability when the new folder has no match; and re-scan behavioral idempotence for the same root. This complements the existing tier1_strict_boot_probe (real-binary path) and test_asset_status_data_dir_change_cache_invalidation (cache invalidation) so the M12 layer under the data-dir picker stays honest across the platform-matrix rows in docs/PLATFORM_MATRIX.md.
• The A5 row “--scan-data smoke reports real READY-path:er” was updated with bounded evidence pointing at the new probe; no row in the platform matrix changed status, and no public legend wording was rewritten.