Firestaff

DM1/CSB ReDMCSB Behaviour Audit

Reviewed 2026-08-31 against ReDMCSB WIP 2021-02-06, Toolchains/Common/Source/, and the local Firestaff source tree. This is an audit of original behaviour, not a list of defects Firestaff should silently “improve” in preservation mode.

Reading the status

The previous version of this page called all listed ReDMCSB bugs “NOT FIXED”. That was inaccurate: it conflated missing frame captures with missing code and also treated historical behaviour as a Firestaff defect.

High-value ReDMCSB findings

Source item DM1 CSB Current Firestaff position Remaining proof
BUG0_02 24-bit timeline wrap affected affected Capture pending; no claim that long-session wrap is reproduced or corrected Original long-session trace/save near the 24-bit boundary
BUG0_03 VBlank palette race affected Atari ST fixed by CHANGE7_01_FIX CSB source modelled by the Atari-ST VBlank delivery path and nested-arrival regression; DM1 needs original ST capture before a timing-parity claim DM1 and CSB low-light frame traces under load
BUG0_04 / BUG7_01 creature palette quirks affected affected Source modelled palette-selection rules; preserve, do not correct Original map-specific RGB/frame captures, especially CSB Worm maps 0, 4 and 9
BUG0_05 portrait-sensor wall ordering affected affected Capture pending Original/custom-dungeon frame that exposes the ordering
BUG0_06 cropped flipped projectile/explosion affected affected Preserve, do not correct; render-plan coverage exists, but no source-owned edge capture validates final pixels Left-edge projectile and explosion captures for each selected edition
BUG0_07 cropped unflipped explosion DM 1.0 only; fixed in DM 1.1+ fixed Version-specific rule; do not apply a DM 1.0 defect to PC 3.4, Atari 1.1+, or CSB Version-labelled original edge capture
BUG0_08 square-first-Thing overflow affected affected Bounded Firestaff data handling fails safely rather than writing outside memory; this is intentional host safety, not preservation evidence A documented Modern safety policy if compatibility mode is ever exposed
BUG0_64 floor ornament/open-pit order affected affected Source draw order is recorded; capture pending Original frame/capture pair
BUG0_83 Thieves Eye/door relation affected n/a Capture pending Original animation capture
BUG0_86 custom-dungeon portrait resource condition affected affected Capture pending Original custom-dungeon corpus
PANEL.C F0352/F0353 eye hold/release lifecycle affected affected Source modelled — C071 records the held-eye state; mouse release restores F0347’s normal inventory panel, closes the transient eye-opened chest, and reopens an action-hand container through ordinary F0333 Original press-and-release capture for each selected edition

Specific corrections made by this audit

  1. CSB’s CHANGE7_01_FIX must not be reported as unimplemented. The source-modelled VBlank route is tested separately; frame-perfect parity is still open because the required original gameplay captures are absent.
  2. BUG0_07 is edition-specific. ReDMCSB marks it fixed in DM 1.1 and later, including DM1 PC 3.4 and the supported CSB families. CSB has no PC/DOS edition; it is wrong to label the defect as open for the supported retail routes.
  3. The CSB dialog engine-version surface is source modelled and covered by test_csb_v1_graphics_extras_pc34_compat and the boot/title handoff test; it is not an open graphics gap.
  4. No historical ReDMCSB bug is silently “fixed” in Original mode. Firestaff may safely reject malformed media or prevent host-memory corruption, but that safety boundary is documented separately from preservation parity.
  5. The DM1 D0/D1 wall and door source locks no longer mistake unavailable private media extractions or third-party source trees for renderer failures. They now require the named ReDMCSB anchors, Firestaff’s corresponding metadata, and the focused built viewport regression. This exposed and fixed a real test-fixture defect: a D2-only door-frame provider left newly added D1 graphic pointers uninitialised, allowing a D1 draw pass to dereference garbage. The provider is now fully zero-initialised before its D2 fixtures are installed. This is a deterministic test repair, not synthetic game data and not a claim of capture-level pixel parity.
  6. PANEL.C:F0352 is a held control, not a one-shot object-description command. ReDMCSB dispatches F0353_INVENTORY_DrawStopPressingEye on release; Firestaff now models that lifecycle before generic drag routing. This clears transient chest/action-hand state without changing preserved game-data behaviour.

Evidence locations