missing.md

Genuine CNA gaps relevant to the modern Racing Game route

Scope and baselines

This file intentionally lists framework gaps, not the work of porting a game. It uses:

• modern primary source /rv/tmp/RacingGame at d8092633e4e43e014ff168d8e913a9373538b851;
• ../cna at ac3aaaeb2a5ba27dbd9e22e782c7041e6e40947c;
• ../cnaintegration at 4ac696c748fb18eef7dd06cca82a0486549bcd5d only as newer evidence.

Recheck every row against the exact post-modularization CNA commit selected for implementation. “Validation gap” means not proven; it is not yet a defect.

Changes after the rds1983/RacingGame delta audit

Previous finding	New status
CNA needs a route for 57 legacy .x models	Superseded. The primary source has 57 checked-in GLBs and no .x; direct .x support is not a Racing requirement
A custom Racing model/material package is needed	Superseded. GLB + 57 .material sidecars + raw assets are the cleaner contract
Hierarchy/tangent/material metadata must be reconstructed	Substantially reduced. Data is present; CNA must preserve/expose it
Authentic generated XACT banks are missing	Superseded as acquisition risk. Real runtime-used version-46 banks are present; CNA validation remains
No runnable reference exists	Superseded. The user confirmed the modern FNA build runs on Linux
Rgba64 may not matter if rendering changes	Still relevant. Modern render code remains and requests it
Custom Effect support is important	More important and better evidenced. Modern FNA/MG code still relies on the full named Effect/technique/pass/parameter model
DDS cube issue	Still relevant. The normalization cube is still uncompressed RGB24

Confirmed framework gaps

1. No complete portable arbitrary custom-Effect runtime/asset path

CNA has Effect-related public types and stock effects, and it has backend-specific ShaderEffect facilities. It does not yet provide the complete reusable route Racing needs to populate an arbitrary custom Effect with:

• named parameters and texture values;
• named technique/pass collections;
• CurrentTechnique;
• ordered EffectPass::Apply() behavior;
• immutable program sharing plus material-instance/clone state;
• pass render-state requirements;
• backend program/shader-module binding.

FNA DX9 .efb, MonoGame MGFX, and general custom-effect XNB payloads do not provide a portable CNA solution. The framework gap is a CNA-owned portable Effect description/runtime, not a DX9 FX compiler.

Priority: highest, after modularization. Prove it with one normal/specular technique and one multi-pass post effect before bulk Racing shader work.

2. Current glTF/Model import does not preserve Racing's required semantics

CNA can parse GLB, buffers and indices. The inspected integration importer does not yet preserve the complete unskinned scene-node graph and instances, original node names/matrices/parents, mesh-to-ordered-parts grouping, material-name/index binding, or DigitalRise-equivalent bounds. It recognizes standard TANGENT, while all Racing GLBs use _TANGENT/_BINORMAL; the special sky cube uses VEC3 TEXCOORD_0 where the current path assumes VEC2.

This is a bounded general model-module gap, not a reason to port DigitalRiseModel or write a new model converter. Required proof assets: car, windmill, one alpha model, and sky cube.

3. Rgba64 render-target capability reporting and construction disagree

Modern Racing still requests Rgba64 in its render-to-texture path. In the audited CNA baseline, GraphicsAdapter.QueryRenderTargetFormat may select it while ordinary texture/render-target validation rejects it. Capability query and construction must agree.

The resolution may be either real backend support or an honest fallback/refusal. Android/Web additionally need a tested Color fallback and reduced-quality target policy at the game level. The contradictory framework answer is the CNA defect.

4. Loose DDS cube loading does not cover the supplied RGB24 normalization cube

SkyCubeMap.DDS is DXT1 and fits the current route. NormalizeCubeMap.DDS is an uncompressed 24-bit RGB cube and remains outside CNA's known loose-cube format path. Add a generic RGB24 cube decode/upload test and support, or an explicitly approved framework-level conversion to RGBA8. This is a small CNA content/texture gap.

Useful general additions, not proven blockers

5. A modular raw-asset provider/cache/lifetime boundary

Racing can implement narrow direct readers today, so this is not strictly missing. After CNA modularization, a general raw-asset boundary would be useful:

• canonical, case-sensitive-safe asset IDs;
• provider/stream abstraction for desktop files, Android packaged assets, and Web preload/cache/stream sources;
• typed cache keys including decode options;
• explicit RAII/unload/device-restoration behavior.

Do not copy AssetManagementBase's managed-object dictionary verbatim and do not build a parallel XNAssets clone. The Racing .material schema and track/combi/font parsers remain game-side.

6. Portable Effect shader-module sharing across the GL family

Newer integration evidence exposes OPENGL33, OPENGLES, and WEBGL2 through one GL-family implementation. Racing would benefit from one semantic Effect graph and a shared GLSL ES 3.00 body adapted to GLSL 3.30 core. Whether the final module API needs a new framework feature depends on the post-modularization design; do not freeze an ABI from the pre-modularized checkout.

Validation gaps — not yet CNA defects

7. Authentic Racing XACT banks have not been run through CNA end to end

The modern repository supplies real version-46 .xgs, .xsb, and .xwb files and uses them at runtime. CNA must be tested against these exact local inputs for:

• parsing and name/index resolution;
• every wave encoding present;
• cue/category/global and per-cue variables;
• RPC/variation/filter behavior used by motor/gear/brakes;
• fire-and-forget and long-lived cue lifetime;
• pause/resume/volume/device teardown.

Repeat first on Linux desktop, then independently on Android and Web. CNA's XACT demo is currently excluded on Android/Emscripten, and browser audio requires a user gesture; neither fact alone proves the library cannot work there. Do not file a framework defect until a focused supplied-bank test fails.

8. Racing render-target chains and custom layouts need an integrated proof

Existing focused CNA tests do not prove the complete shadow → scene → glow/blur/menu → UI chain with Racing's tangent layout, state transitions, cube textures, resize, readback, and resource restoration. Build this incrementally in the pinned harness.

9. OPENGL33 needs non-Linux qualification; OPENGLES/WEBGL2 need target qualification

The newer OPENGL33 profile has extensive real Linux/Mesa coverage, not Windows driver coverage. OPENGLES is the Android route and WEBGL2 builds through Emscripten, but neither has a complete Racing-class target proof. These are platform acceptance gaps, not automatically framework bugs.

10. Android sensor/touch and Web touch need game-shaped hardware tests

CNA already has TouchPanel, gamepad support, Android Back mapping, and an Android Accelerometer implementation with landscape-axis handling. Racing must test simultaneous steer/pedal touches, safe-area transforms, controller switching, sensor calibration/orientation, background/resume, and missing-sensor fallback. Any failure must be isolated before deciding whether it belongs to CNA or the game overlay.

11. Web persistent storage and audio-unlock behavior need a complete-game proof

CNA has an Emscripten main loop and browser input path. Racing still needs durable settings/replay/highscore storage and XACT/audio behavior after a browser user gesture, tab suspend/resume, and reload. A game-specific synchronization mistake or packaging choice is not a CNA gap; a missing reusable storage/audio primitive may be.

Port-side work that is not a CNA gap

• translating roughly 23.6k physical lines of C# to C++;
• typed parsing/validation of Racing .material JSON;
• binding Racing effect/technique/parameter names;
• track, combi, height, bitmap-font and replay/settings parsers;
• car/windmill behavior and alpha filename convention;
• desktop logical-control mapping;
• Android touch artwork/layout, left-handed/accessibility options and tilt filter;
• Android Gradle/APK packaging and landscape app policy;
• Web content grouping/download/progress/cache UX;
• platform quality presets, downscaled/compressed derivative assets and performance tuning;
• asset licensing/provenance decisions;
• pinning/reproducing the FNA oracle's sibling dependencies.

Verified capabilities that must not be reported as missing

• partial XNB support, including model/texture/cube readers;
• GLB container/accessor/index parsing;
• ordinary TGA/PNG Texture2D loading;
• DXT1 cube data route;
• vertex/index buffers, 32-bit indices and custom layouts on the GL path;
• render targets, blend/depth/rasterizer/sampler states and readback in principle;
• public Effect parameter/technique/pass types used by stock effects;
• keyboard, mouse, gamepad and multi-touch input;
• Android Back → Escape mapping;
• Android Accelerometer support (hardware validation still required);
• substantial XACT version-46 parser/playback behavior;
• Emscripten non-blocking Game loop and SDL input bridge.

Backend findings removed from the primary gap list

Old Bgfx/Vulkan/SDL-renderer-specific custom-effect limitations remain true or need their own audits, but they do not block the selected first route. The primary sequence is Linux OPENGL33, Windows OPENGL33, Android OPENGLES, then Web WEBGL2. Adding D3D11, Vulkan, Bgfx, WebGPU, SDL_RENDERER, or fixed-function backends is separate scope.

Priority after CNA modularization

1. Pin CNA and prove minimal OPENGL33 lifecycle/render/readback tests.
2. Freeze and implement the CNA-owned portable Effect runtime/description.
3. Add the bounded semantic GLB/model behavior and four-model proof.
4. Fix/test Rgba64 capability truth and RGB24 DDS cube loading.
5. Run the supplied Racing XACT banks through focused CNA tests.
6. Prove the integrated static scene, then the complete render-target chain.
7. After Linux completion, qualify Windows, Android and Web in that order.