Research: a detailed, licensed CAD model of a specific engine for graphics quality

[PouyanJay]

Every model in the app is generated in TypeScript. That keeps the repository self-contained and lets the separation and mechanism code know every piece, but the geometry is deliberately generic. The largest possible step in visual quality is a detailed, licensed model of one specific real engine. This issue is to investigate and decide, not to build yet.

What is needed

A written recommendation, added under docs/, that answers:

• Which engine to source, and from where. Candidates include manufacturer education programmes, university CAD repositories, and commercial marketplaces with explicit licences.
• Licence terms: whether redistribution in a public repository is allowed, attribution requirements, and whether derived works (our part naming, decimation, separation metadata) are permitted.
• Technical fit: units, coordinate conventions, level of detail, triangle count, and whether the model comes as a hierarchy with named parts or as a single mesh. The separation feature needs individual pieces with stable identities, so a single mesh is of little use.
• Delivery: whether the asset would live in the repository, be fetched at runtime, or ship as an optional download, and what that means for the current "no network needed" property.
• Effort: what the import pipeline would look like (glTF with a part manifest is the likely format) and how the existing catalog, picking, labels, cutaway and separation would map onto imported parts.

Success criteria

• A document in docs/ with a comparison of at least three candidate sources and a clear recommendation, or a clear recommendation not to proceed.
• The licence for the recommended asset has been read in full and the relevant clauses are quoted.
• A small proof of concept branch that loads a glTF file into the scene, registers its parts with the separation controller, and passes the identity and reassembly checks. This proves the pipeline before any purchase.
• A note in docs/ARCHITECTURE.md recording the source, licence, units, coordinate conventions, level of detail and hierarchy, as the existing Graphics section already asks for.

Suggested approach

• Add a glTF loading path to the scene behind a feature flag. Map each named node to a piece, and read a sidecar manifest that assigns nodes to component groups. The catalog in lib/engine/engine-data.ts gains an optional source: 'generated' | 'imported' field.
• Test the pipeline with a free, permissively licensed model first (for example a Creative Commons turbine from a public repository) so the code work is not blocked on procurement.
• Decimate and compress with the standard glTF tooling (Draco or Meshopt) and measure load time on a mid-range phone before deciding on delivery.
• Keep the generated models as the default. An imported model should be an additional entry in the engine library, not a replacement, so the educational reference set stays intact.

Out of scope

Buying anything, and replacing the generated engines.

Touchpoints

docs/, components/engine/engine-scene.tsx, lib/engine/engine-data.ts, lib/engine/explosion-controller.ts