What problem does it solve? After a screening round of environmental model calibration, some validation targets still miss their observations, and it is unclear why. This Skill runs Phase 3 (DIAGNOSIS) of the A2MC calibration workflow as the offline agent, systematically root-causing why the screening ensemble misses its targets and producing a structured diagnosis for the next phase. ## Core Features & Use Cases - Ensemble-level diagnosis: Analyzes the top-N leader cases as a set, grouping them by miss pattern rather than fixating on the single best case, using phase3 diagnostic tools (compare_targets, check_edge_parameters, carbon/nutrient/mortality analyzers). - Evidence-grounded reasoning: Pulls context from RAG/GraphRAG, the model knowledge base (five surfaces), and per-site Adaptive Memory before asserting any mechanism, and verifies every mechanism claim at file:line in the checked-out model source. - Structured handoff: Produces ranked root causes, severity-classified failing targets, parameter recommendations, selected base cases, and testable hypotheses, logged via PhaseLogger and handed to Phase 4. - Use Case: After Phase 2 screening of an EcoSIM or FATES calibration round, ask the agent to diagnose why the leaders miss standing carbon while matching production; it gathers diagnostics, checks Memory for failed directions, and returns a ranked root-cause diagnosis with base cases for Phase 4 experiments. ## Quick Start Ask the agent to run Phase 3 and diagnose why the current screening round's targets are not calibrating, using the existing ensemble outputs.
