Publish the 500-task shopping selection and historical agent scores
The current shopping comparison needs one versioned 500-task selection. V10 removes Ozon and Coupang equally, recalculates all scores, and preserves V9 plus every original outcome.
Browser Use Cloud shows 459/500, Dot shows 363/500, and Muse shows 306/500. The owner approves Dot as the display name for the historical OpenAI runner. Its original runner ID remains intact. Instinct remains partial. A later native message reports 292 provisional successes from 382 finished attempts on its original 513-task set. Updated task rows remain missing. Grok retains its separate 36/50 browsing study.
Per-task CAPTCHA report fields support the published counts. Original Muse failure categories distinguish website security blocks from browser errors. Later exclusions and the selected best Cloud run limit generalization. These are historical projections, not fresh V10 runs or infrastructure-only block rates.
Validation: six scoring tests pass. All 500 task IDs and domains are unique. Original outcome values remain unchanged. Public website downloads match the repository files.
The latest native Instinct report gives 292/513 provisional successes and a daily browser-time pause. It includes known partial results. The earlier archive snapshot remains separate.