StyleSeed AI Builder Benchmark v1.0

Results

Agent	Bare	+Rules	+Rules+Gate	Gate vs bare
Codex	74.8	76.4	80.1	+5.3
Claude Code	74.1	70.4	79.4	+5.3

• 120/120 Claude Code and Codex cells scored; 0 failed.
• Neutral judge: Gemini 3.1 Pro Preview, fixed seed; five-run stability spread 2.
• Full prompt, code, HTML, screenshot, verdict, and revision evidence is attached.
• SHA-256: 7ecd5ca7bc54eb05f3e58e4594d66c455893d5d356492f9bcc68bed98e43767d
• v0, Bolt, and Lovable comparison cells remain blocked pending first-party access; no score was imputed.

See bench/v1/README.md, leaderboard.md, results.json, and validation.json in the repository.