What problem does it solve? Autonomous research loops often jump from reading papers to changing code without a defensible link between prior work and the observed failure. This Skill converts deep-read literature and measured baseline results into a ranked pool of falsifiable adaptation candidates, so every code change is attributable to a specific mechanism. ## Core Features & Use Cases - Structured Candidate Mapping: Writes each candidate as a row in research/CONTRIBUTION_MAP.md covering source contribution, assumption, baseline match, hypothesis, falsification criterion, minimal adaptation, novelty risk, screen, and cost. - Guard-Allocated IDs: Runs transition_guard.py allocate-candidates to obtain stable candidate ids before writing rows, keeping the research state machine consistent. - Evidence-Based Ranking and Selection: Ranks candidates by mechanism match, source evidence strength, expected primary-metric effect, distinct contribution, and screening cost, then advances the mapping -> adaptation transition or returns to literature with an evidence_gap. - Use Case: After a baseline experiment fails and a literature refresh is complete, use this Skill to produce three to five ranked adaptation candidates with explicit falsification criteria before selecting the next experiment. ## Quick Start Map the current literature and baseline failure into ranked adaptation candidates and select the next one to test.
