Files
Neutron/docs/tasks.md
T

5.2 KiB

Tasks

Completed milestones

  • Entire specification; original inventory/hashes; local-only Git and guidance.
  • Binary/compiler/API identification and address-based reports.
  • All GAM entities decoded and byte-round-tripped; C# corpus reader tested.
  • All indexed OMT meshes/materials/textures/RIFF resources decoded and audited.
  • ASE transform/morph-base UV/topology recovery and regression tests.
  • .NET10 + SDL3 + Vulkan1.3 shared source; real scene/model/texture rendering.
  • Original LV1B lab plus original Jimmy morph preview at GAM startpoint.
  • Correct normal atlas and null-material visibility from binary evidence.
  • Recovered impulse/gravity/speed/RNG arithmetic fragments with tests.
  • Native Linux Vulkan offscreen and SDL3 window/input/resize verification.
  • SDL3 recovered WAV stream consumption test (dummy device).
  • Self-contained Linux and Windows local packages, matching SDL3 libraries.
  • Packaged Linux runtime isolated from SDK/source/originals; renders successfully.
  • Preliminary Wine64 Windows-package render; exact fixed-time readback matches Linux.
  • Isolated copied-original Wine attempts; actual failures documented.
  • Correct preview floor overlap, identify still idle, add explicit animation inspection controls/tests.
  • User confirmed corrected Windows build loads and renders.
  • Controllable lab: recovered walking/backpedal/turn/no-gadget jump formulas, original animation clips, fixed60Hz simulation, collision and chase camera.
  • 11 collision and13 movement test groups; source lab and synthetic edge cases, render-rate independence, held/re-pressed jumping and ten-minute stability.
  • Linux240-step SDL key/jump/resize replay and GPU readback, zero validation errors.
  • Rebuilt both packages; packaged Linux window and preliminary Windows Wine64 window240-step walking/key/jump/resize pass, matching movement statistics.
  • Test this movement checkpoint natively on Windows (previous rendering confirmed).

Next implementation priorities

  1. Replace provisional capsule collision/pivot with recovered original behavior. Trace 0x479D20 past displacement integration, shared collision call via interface slot0x130 in 0x462520, and object collision caches. Retain invisible/null-material geometry; its exact collision masks remain unresolved. Recover original actor/model pivot, scale and spawn settling: the preview now uses a floor-derived model offset, which must be replaced with proven behavior. Recover stair/step handling, dynamic collider activation, and camera obstruction.
  2. Compare reconstructed no-gadget airborne behavior with the original. Walking, jump state table and zero-fuel held/re-press repeat path are now recovered in analysis/movement-checkpoint.md; transient state3 and takeoff phase remain approximations. Recover powered gadget states and original animation routing.
  3. Recover original frame cadence and phase order; current fixed60Hz preserves per-call speed increments independent of rendering. Find reverse timer+0x830 writer before implementing its >1.5 turnaround branch. Recover mouse steering.
  4. Execute LV1B STARTEXP/LABEXP intro. GAM parsing supplies records; implement task prerequisites, object activation/visibility, camera triggers and progression from code/observations. Then build a complete small scene (VR01 if useful).
  5. Recover original morph state selection, playback speed, smooth normals and transitions. Current interpolation is original; loop/speed are preview policy.
  6. Continue material/fog/light/blend/culling and render sorting mapping, then compare visuals to original; current unlit modulation is only a foundation.

Required validation and expansion

  • Usable original reference captures under local Windows/VM or working Wine.
  • Original walking/jump/collision/camera/input comparisons and first complete scene including intro/triggers/interactions (movement checkpoint implemented).
  • Physical audio output and original SFX/music timing, mixing/panning/positional audio.
  • Fullscreen, minimized-window, mouse/relative capture, display scaling and controller behavior.
  • Pickups, score/health, gadgets/inventory, combat/AI, all levels and transitions.
  • Original menus, cutscenes, dialogue/voice and state-dependent rendering.
  • Save/load schema and configuration persistence; NewGame reference is not save support.
  • Full native Windows input/audio/resize/driver testing (initial rendering is user-reported).
  • Fidelity regression references, remaining discrepancies and actual release candidate.

Repeatable commands

  • source scripts/env.sh; dotnet build
  • scripts/test.sh (20 Python tests, C# checks, native offscreen/audio, original hashes)
  • python3 tools/extractors/audit.py ../Original --report artifacts/extracted/audit.json
  • NEUTRON_WINDOW_TEST=1 scripts/test.sh with DISPLAY for input/resize smoke
  • scripts/package.sh for local investigation packages (not a finished game release)

Keep runtime, tests and docs coherent. No remotes/uploads/cloud CI. Do not modify originals or claim unobserved fidelity. All prior agent ownership has returned to primary; use explicit file ownership if further parallel work is started.