{ "approved_for_default": false, "artifact_type": "phase17-calibrated-release-selector", "baseline_router": "baseline-minilm", "candidate_router": "finetuned-embedding", "decision": "KEEP_BASELINE", "improvement_count": 0, "input_paths": { "baseline_results": "docs/demo/phase16-blind-validation/baseline-minilm/results.jsonl", "candidate_results": "docs/demo/phase16-blind-validation/finetuned-embedding/results.jsonl", "regression_summary": "docs/demo/phase16-blind-validation/regression-summary.json", "route_diffs": "docs/demo/phase16-blind-validation/route-diffs.jsonl", "task_root": "benchmarks/blind-migration-tasks" }, "metric_deltas": { "mrr": -0.03125, "ndcg_at_5": -0.023067, "negative_accepted_rate": 0.0625, "negative_hit_rate": 0.0625, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "phase": "Phase 17", "policy": { "max_negative_accepted_delta": 0.0, "max_negative_hit_delta": 0.0, "max_regressions": 0, "min_mrr_delta": 0.0, "min_ndcg_at_5_delta": 0.0, "min_recall_at_5_delta": 0.0 }, "reasons": [ "source guard_status is not PASS", "regression_count exceeds policy", "negative_hit_rate delta exceeds policy", "negative_accepted_rate delta exceeds policy", "mrr delta is below policy", "ndcg_at_5 delta is below policy" ], "regression_count": 2, "regression_flag_counts": { "mrr_decreased": 1, "ndcg_at_5_decreased": 1, "negative_accepted_rate_increased": 1, "negative_hit_rate_increased": 1, "new_negative_skill_selected": 1 }, "selected_router": "baseline-minilm", "source_guard_status": "REVIEW_REQUIRED", "source_phase": "Phase 16", "task_count": 16, "task_decisions": [ { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-browser-accessibility-tree" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-browser-form-wizard" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": false, "task_id": "blind-browser-smoke-console" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": false, "task_id": "blind-browser-visual-diff" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "KEEP_BASELINE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 1.0, "negative_hit_rate": 1.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [ "negative_hit_rate_increased", "negative_accepted_rate_increased", "new_negative_skill_selected" ], "selection_changed": true, "task_id": "blind-claude-mcp-routing" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": false, "task_id": "blind-claude-plan-session" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-claude-slash-command" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-claude-task-delegation" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-codex-apply-patch" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-codex-evidence-final" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": false, "task_id": "blind-codex-git-hygiene" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "KEEP_BASELINE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": -0.5, "ndcg_at_5": -0.36907, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [ "mrr_decreased", "ndcg_at_5_decreased" ], "selection_changed": true, "task_id": "blind-codex-worker-handoff" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-sp-debug-loop" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-sp-red-green" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-sp-verify-before-claim" }, { "aggregate_decision": "KEEP_BASELINE", "decision": "NO_CHANGE", "decision_scope": "task-level", "improvement_flags": [], "metric_deltas": { "mrr": 0.0, "ndcg_at_5": 0.0, "negative_accepted_rate": 0.0, "negative_hit_rate": 0.0, "recall_at_5": 0.0, "selection_rate_at_5": 0.0 }, "regression_flags": [], "selection_changed": true, "task_id": "blind-sp-worktree-isolation" } ] }