{"slug":"wshobson-eval-harness-first","source_name":"wshobson/eval-harness-first","name":"Wshobson/Eval Harness First","description":"Build the evaluation harness that gates every fine-tuning run — golden sets, per-failure-mode graders, judge calibration, and base-model baselines. Use when starting a fine-tuning effort, when converting traces into an eval set, or when calibrating a judge against human labels.","version":1,"lift":{"pass_rate_delta_pts":37.5,"pass_rate_pct":91.7,"total_cases":24,"passed_cases":22,"tokens_delta_pct":33.4,"turns_delta_pct":0,"verdict":"mixed","benchmark_model":"gemini-3.6-flash","grading_method":"judged","completed_at":"2026-08-07T12:37:04.503843+00:00"},"skill_score":0.9167,"benchmark_models":[{"model":"gemini-3.6-flash","headline":true,"delta_pts":37.5,"with_pass_pct":91.7,"without_pass_pct":54.2,"tokens_delta_pct":33.4,"turns_delta_pct":0,"total_cases":24,"cases_aggregated":24,"verdict":"mixed","never_hurt":false,"completed_at":"2026-08-07T12:37:04.503843+00:00","run_id":"32c3e848-a13e-4266-99a6-104d7d5ef9f2","version_number":1,"is_latest_version":true,"gate":null}],"trust":{"skill_safety":"passed","safety_status":"clean","intent_verdict":"safe","content_status":"clean","indexable":true},"license":"MIT","install_count":0,"manifest_hash":"fab60b69f3c3448d093c101dfbc86b380edec4617572fab2846551b7bdf8bb6e","raw_url":"https://app.decimal.ai/s/wshobson-eval-harness-first/SKILL.md","scorecard_url":"https://app.decimal.ai/skills/wshobson-eval-harness-first"}