Claude Haiku 5.5mediumin Claude CodeGalton board
Running live
OPEN FULL SCREEN ↗
Open it running ↗This is the exact file the model wrote, unedited, served from an isolated domain with the network blocked, as in the test.
What the robot did
The comparison clip (the same script for every model)
Checks
14/15Written from the prompt (test v1, checks v1). The model never saw them.
- ✓Opens without a page error
- ✓Test hooks present (getState, setGravity, pause, resume, reset)
- ✓Visible canvas, at least 800×6001920×1080
- ✓Animation is alive (frames keep changing)395 quadros em 3 s; imagem mudou
- ✓Smooth: median frame time under 25 msintervalo mediano 8.3 ms (~120 fps)
- ✓Starts by itself: balls fall and reach the binsdropped=1080 inBins=466
- ✓At least 13 bins15 caixas
- ✓Counters add up (balls in bins ≤ balls dropped)soma=466 inBins=466 dropped=1080
- ✗Bell-shaped histogram (half the balls in the middle third)terço central com 27% de 466 bolas
- ✓Pause freezes, resume continuespausado: 1096→1096; depois de retomar: 1186
- ✓Changing gravity changes the scene1400 → 1120 (pedido 1120.0)
- ✓Reset empties the binsdepois do reset: inBins=0 dropped=12
- ✓No uncaught exception during the session
- ✓No console errors
- ✓No network attempts
The record
DOWNLOAD JSON ↓- Model
- Claude Haiku 5.5
claude-haiku-5-5 - Reasoning effort
- medium
- Started
- Oct 9, 23:51 UTC
- Harness
- Claude Code 2.1.296, by Anthropic (not ours) · one shot · tools used: 0
- Harness settings
- print mode, safe mode, every tool off (--tools ""), no saved session, empty folder outside Git ·
CLAUDE_CODE_MAX_OUTPUT_TOKENS=128000 - Tokens
- in 2 · cache read 3,007 · out 71,360 · reasoning 60,438 · requests 1
- Cost
- $0.036 API-equivalent · extra spend $0 (subscription)
- Test
- Galton board v1 · checks v1 · prompt
2e0eaba1ad6a - Output file
- 21.8 KB · sha256
f57d9a7a7969 - Run id
20261009T235120Z_galton-v1_haiku-medium_r2