experiment-run

What problem does it solve? It turns completed background experiment jobs into disciplined, auditable pipeline decisions, preventing premature promotion, post-hoc threshold changes, and uncontrolled reruns in autonomous research workflows. ## Core Features & Use Cases - Result Collection: Verifies the plan, baseline, reserved scientific run id, and DSH job completion, then recomputes the declared metric and appends a factual record to research/RESULTS.jsonl. - Screening Routing: Routes passing runs to confirmation and failing runs back to mapping or literature with explicit outcomes like screen_pass or screen_fail. - Confirmation Routing: Promotes only frozen candidates that pass the full promotion rule, and rejects unstable or guardrail-violating results as not_confirmed. - Use Case: After experiment-bridge launches a screening run, use this skill to collect the finished job, record the metric delta, and decide whether the candidate advances to confirmation or returns to candidate mapping. ## Quick Start Collect the completed experiment run reserved by experiment-bridge and route the screening result to the next pipeline stage.