Install the exact dependencies and preserve the resulting real lockfile. Run all tests and ensure the two Rapier tests execute rather than skip. Open the hosted application in a real WebGL-capable browser. Verify 1, 40, 163 and 1,000 bodies across scatter, gentle, handful, rain, pour, and independent removal. Inspect contact tunneling, initial interpenetration, energy gain, escaping/timeout policies, compound mass and inertia, correct finite bounds, wake-up after apparent rest, and rest-duration timing. Check multiple render rates and simulation speed settings without changing the final outcome on the same engine/runtime.
Capture actual physical manifests, frame traces, and screenshots. Do not substitute Canvas mathematical placement as evidence that those tests passed. Verify all documented Rapier calls against the installed binding and check GPU renderer initialization, draw calls, resize handling, picking, PNG capture, surface tilt, and graphics-resource cleanup. Resolve any failures before marking the adapter supported.
Select actual match and board specimens; measure dimensions, mass, mass distribution, and release conditions. Collect drop, sliding, restitution, and settling data with uncertainty. Fit physical parameters on one set and assess on independent held-out drops. Repeat at different heights and orientations. Evaluate aerodynamic effects and contact-law limitations. Do not label the current assumed damping coefficients as experimentally measured rolling resistance or airflow.
Add calibrated profiles, bounded uncertainty for their fitted parameters, and a physical bias-study interface. Analyze joint spatial/orientation structure and contacts, not only marginal histograms. Compare independent drops with batches and piles. Keep the ideal-theory interval separate from physical model uncertainty.
Implement and benchmark vectorized/GPU and multi-worker paths only after specifying a disjoint-stream or counter-based RNG contract. Differential-test their geometry against the scalar reference, including borderline endpoints and large counts. Retain exact integer reductions and bounded memory. Add measured device-dependent backend thresholds. Instrument actual worker, GPU, and collision cost rather than inventing utilization figures.
The requirement map identifies the unimplemented features: production photorealistic assets and optical effects, dense-line LOD, force/contact-vector inspection and frame replay, detailed physical heatmaps, multi-experiment side-by-side races, cinematic intro, empirical calibration workflow, imperial display units, Parquet, more historical material with verified sourcing, and optional variance-reduction / advanced geometric extensions.
Audit dependencies and licenses, preserve a genuine lockfile and software bill of materials, test supported OS/browser combinations, perform an accessibility audit, profile large UI studies and resource cleanup, add regression baselines and continuous integration, and define source-version/build provenance. Re-run all evidence-producing scripts on the exact release candidate. Update the validation report to distinguish newly tested paths from ones still skipped or unimplemented.