3D Browser Game

The coordinator applies library context. Inputs: player brief, existing project, target conditions, workspace and constraints. Choose and record unspecified creative details; ask only when a missing choice blocks progress. A bounded phase uses its prerequisites and returns reached work without claiming a finished game.

Use shared creative direction and one integration owner. Production staffing follows core execution rules; only the two playtests require fresh independent reviewers. Missing capabilities block dependent work.

Checkpoints

Use evidence for candidate identities and ready, needs change or unverified conclusions. Freeze each candidate until its independent playtest report completes. These are work gates, not user approval prompts.

Each checkpoint permits at most one necessary repair pass: reproduce material findings, fix shared causes, rebuild and replay affected situations plus start–play–outcome–retry. Recapture changed visuals and remeasure affected performance. Preserve the original report and identify the delivered revision; subsequent checks are maker verification, not renewed independent acceptance.

An unchanged ready candidate needs no repair, rebuild or replay. Pre-checkpoint QA/tuning is ordinary production. If a checkpoint remains needs change or unverified, stop dependent production and return the best runnable state, findings and precise next action.

1. Establish scope and capabilities

Record the player promise, requirements versus defaults, browser/device/input, session scope and observable completion criteria. Preserve existing intent and architecture; bound a larger release without silently dropping requirements. Select browser-game for design.

Probe install/build/preview, rendered browser input/captures and Blender source-to-GLB loading in Three.js under the test interface and library capability rules. Probe promised audio. Remove disposable probe content from the finished game. Continue only with a feasible ordinary-play route and recorded capability results.

2. Produce and test the core

Compare at least three materially different mechanics; choose rules, controls, tunable parameters and completion states. Before implementation, write experiment predictions and observable success/failure scenarios. Set load, frame-time and asset/content budget hypotheses for the agreed device.

Build a Three.js graybox under the test interface and Three.js reference. Complete the session with responsive controls, readable placeholders, real collision/progression, an outcome and ordinary-input restart. Verify rule behavior, interface and production boot before experiments; control defects must not distort them.

Run two focused mechanics experiments. Each predicts how one rule/parameter change affects decisions, compares observed baseline and changed behavior in the same scenario, and records the chosen setting and contrary evidence. Include alternate strategies, misuse and recovery. If meaningful decisions remain absent, report needs change before independent review.

The core handoff includes the runnable graybox, mechanics comparison/choice, scenario predictions, two observed before/after experiments, verified controls/interface and a complete ordinary session. Final assets wait for the core checkpoint.