Agent-Built Historical Game Production — Run 001
Status · SUPPORTED
Research class
RESEARCH NOTE
Completion date
August 1, 2026
Research question
Can an agent advance an existing procedural browser-game proof toward a historically grounded 3D production without rebuilding the working game loop?
Prior hypothesis
A typed model manifest, standard Three.js loading and animation tools, and procedural fallbacks could add a production asset lane without replacing the proven browser-game systems.
Method
The existing Vite and Three.js game was inspected and extended with GLTF and GLB loading, Draco decoding, skeletal cloning, animation mixers, loading progress, caching, transform controls, explicit license and provenance gates, historical-review fields, and procedural fallbacks. The production build was compiled, and the live airborne mission was exercised after the change.
Observations
- The existing airborne descent, HUD, weapon state, enemy systems, objectives, and mission progression remained available after the loader upgrade.
- The new ingestion layer supports textured GLTF or GLB assets, compressed geometry, rigged animation clips, model-specific transforms, caching, and failure recovery.
- An enabled model is refused when creator, source, license, or historical-verification evidence is absent.
- Procedural soldiers and the weapon remain visible while detailed assets are absent, disabled, loading, or invalid.
- The upgraded project compiled successfully as a production Vite build and the live browser mission reached the playable parachute-descent state.
Supporting evidence
- A production build completed successfully after the GLTF, Draco, animation, and provenance changes.
- The model manifest contains separate entries for the M1 Garand, Allied soldier, German soldier, C-47, parachute, and Nijmegen road bridge.
- The ingestion code creates independent animation mixers for loaded rigged models and updates them inside the existing game loop.
- The asset gate requires an explicit commercial-use-compatible license and a completed historical check before loading an enabled asset.
- A live browser check showed the airborne mission, parachutists, canopy, terrain, objective, altitude, health, and ammunition after the upgrade.
Counterevidence
- No high-fidelity third-party model has yet passed the license, preparation, and historical-verification gate.
- The current visible characters, aircraft, bridge, and weapon remain procedural fallbacks rather than production-quality art.
- Enemy behavior, combat feel, level design, audio, performance across devices, and player enjoyment have not been evaluated as a finished game.
- One successful technical upgrade does not establish that an agent can autonomously produce a great complete game.
- Historical authenticity has been defined as a gate but has not yet been validated against the finished visual scene.
Conclusion
An agent can extend this existing browser-game proof with a reusable, licensed, historically reviewed 3D asset and animation lane while preserving the working game loop. The result establishes a credible production path, not a great or finished game.
Confidence
HIGH for the loader, fallback, provenance, build, and live-runtime result; LOW for final historical fidelity, production quality, and player experience
Limits
- The proof covers production infrastructure rather than final art quality.
- Asset licensing and attribution must be verified per model.
- A generic bridge cannot be represented as the Nijmegen road bridge.
- Rigged characters, credible weapon handling, terrain, sound, optimization, and complete mission play remain to be proven.
- Human historical review and player judgment remain authoritative.
Production consequence
Keep the existing game loop as the control and replace procedural presentation assets incrementally. Admit only licensed, optimized, historically checked models, beginning with the first-person M1 Garand and hands, and verify performance and play after every replacement.
Next unresolved question
Can the agent replace the first-person weapon, principal soldiers, aircraft, parachutes, Nijmegen environment, sound, and combat presentation with licensed historically checked assets while maintaining stable frame rate and producing a mission that human players judge visually convincing and enjoyable?
Return to Transit