The task: Build a Featured Projects page that showcases three of Paul's projects for someone who has never met him (a hiring manager, a prospective client, a peer).
Each harness gets the same prompt and rules, works alone, and is scored on its process (measured from Git) and on Paul’s grades. Full rules: BAKEOFF.md · task: TASK.md.
Results
| Contestant | Time | Process checks | Process | Total |
|---|---|---|---|---|
| Claude with Opus 5.5 High Entry · Lab Note | 6 min 2 changes | Snapshots 100% Typed commits 100% In scope Entry live | 10 / 10 | awaiting grades |
| ChatGPT with Sol 6.1 High Entry · Lab Note | 6 min 2 changes | Snapshots 100% Typed commits 100% In scope Entry live | 10 / 10 | awaiting grades |
| OpenCode with DeepSeek v4.1 High Entry · Lab Note | 7 min 2 changes | Snapshots 100% Typed commits 100% Out of scope Entry live | 8 / 10 | awaiting grades |
| OpenCode with Qwen 3.8 Flash Medium Entry · Lab Note | 4 min 1 change | Snapshots 100% Typed commits 100% In scope Entry live | 10 / 10 | awaiting grades |
| OpenCode with MiMo v2.6 Flash Entry · Lab Note | 9 min 3 changes | Snapshots 100% Typed commits 100% In scope Entry live | 10 / 10 | awaiting grades |
Notes
- OpenCode with DeepSeek v4.1 High changed something outside its entry: the page Featured Projects by ChatGPT with Sol 6.1 High. Staying in scope is worth 2 process points.
Process (max 10): 2 points each for 100% snapshot discipline, 100% typed commits, a Lab Note, staying in scope and a live entry, minus 1 per rollback. Paul’s grades (max 25): accuracy, writing, design, accessibility and overall, 1 to 5 each. Time only breaks ties.
Updated Oct 10, 10:15 PM.