What problem does it solve? Running ad-hoc parameter-sweep experiments on top of an A2MC Morris calibration ensemble often leads to duplicated plans, contaminated ensemble outputs, unverified parameter files, and uninterpretable results. This Skill codifies a reproducible 12-step pipeline so each hypothesis test is falsifiable, isolated, and feeds verified findings back into the knowledge base. ## Core Features & Use Cases - Prior-plan and literature grounding: Searches dev_logs/ana_logs for existing experiment plans and runs a focused literature plus satellite-data review before designing variants. - Falsifiable variant design: Builds variant matrices with a control (V0), a falsifiability variant, and pre-committed quantitative thresholds, then generates and verifies per-variant FATES parameter files. - Isolated HPC submission and analysis: Uses dedicated output directories, case-suffix naming, pre-flight validation, a V0 reproducibility gate, and a decision tree that injects confirmed results into the knowledge base. - Use Case: A user suspects the clumping_index parameter drives a GPP bias in an ELM-FATES Arctic site; the Skill designs an 8-variant sweep on Morris base case 1304, submits it on HPC, verifies V0 reproduces the baseline NRMSE, and writes a results ana_log with a KB-injection decision. ## Quick Start Ask the agent to design and launch an offline parameter-sweep experiment testing a specific FATES parameter hypothesis on a named Morris base case.
