Flawless run and the most considerate code: spawn-overlap avoidance, exponential damping, floor-jitter friction.
Overview › Physics Lab › R3
Physics Lab · archive round
R3 · Bouncing-ball physics: gravity, friction, elastic collisions
Archive round — this round predates the blind-first layout, so answers appear with model names attached. Newer rounds keep the models anonymous until you reveal them.
Scores on this page are the original editorial 0–10 ratings from the 2026-07-17 write-up (writing / art-direction / runtime quality) — not blind rubric scores against a documented gold answer, which is how R31 onward are scored. The underlying answers are unchanged from that write-up.
Show task
A self-contained HTML physics sim, NO user input: balls with gravity, air friction, elastic collisions (walls + ball-to-ball). 8 balls at start, auto-spawn to 40, live counter + FPS.
Untouched model code, executed headless for the verdicts; the stills are the posters.
Answers
RUNS ✓spawn ✓ 11/40 @6sairborne physics ✓
RUNS ✓spawn ✓tightest file — 4.6 KB
Compact and correct: DPR-aware, framerate-independent, glowing render.
RUNS ✓trails + reactive glow
Works, with style — but shipped a dead FPS block, and the "spatial grid" secretly calls indexOf per neighbour. Flair, minus rigour. Disclosed.
RUNS — BUTall balls grounded @6s
Executes cleanly, but the energy budget is wrong: within seconds, a still life. "Elastic" was the brief.
RUNS — BUTballs nearly invisiblespawn stuck at 8
The R1+R2 leader stumbles: dark balls on a dark background, dead spawner. This is why we runtime-test.
SyntaxError: 'lastSpawnTime' has already been declared
One duplicate declaration, total loss — the script never parses.
renderer timed out twice (>45 s)
Locks the rendering thread. Worse than crashing: it takes bystanders with it.