DeepSWE-informed redesign of web/ (Svelte static): hero + stat strip (tasks/categories/models/trials); leaderboard table with pass@1 + Wilson CI + cost/task + tokens + duration and a cost-vs-pass scatter; task catalog cards from task metadata; methodology page (hidden verifiers, negative cases, clean-replay grading, admission gates); run/reproduce page with one-command CLI snippet + subset flag. Data source: the #7 export schema. Current 1-model board stays labeled provisional until the v1.0 run publishes. Taste bar: high (opus lane). Refs #10
DeepSWE-informed redesign of web/ (Svelte static): hero + stat strip (tasks/categories/models/trials); leaderboard table with pass@1 + Wilson CI + cost/task + tokens + duration and a cost-vs-pass scatter; task catalog cards from task metadata; methodology page (hidden verifiers, negative cases, clean-replay grading, admission gates); run/reproduce page with one-command CLI snippet + subset flag. Data source: the #7 export schema. Current 1-model board stays labeled provisional until the v1.0 run publishes. Taste bar: high (opus lane). Refs #10