What problem does it solve? Running EcoSIM calibration ensembles fails in silent, expensive ways: namelists exceeding a 4096-byte buffer, parameters routed to the wrong NetCDF surface, Saltelli designs analyzed as sequence designs, and truncated runs scored as complete. This Skill encodes the full run-and-test workflow so those traps are caught before compute is spent. ## Core Features & Use Cases - Multi-surface case materialization: Builds probe or ensemble cases across EcoSIM's four parameter-file surfaces (plant, management, microbial, soil/grid) with derived routing that raises on ambiguous or missing parameters. - Pre-submission validation and HPC submission: Validates every case's perturbed values against the sampling matrix, then submits as a SLURM job array respecting the 4,995-task shared queue cap, with heartbeat-based monitoring. - Target-driven scoring: Detects true completion via final restart files (not sacct COMPLETED), handles the real Gregorian calendar, and scores through each target's own reduce function. - Use Case: You need to run a 60,000-case Sobol sensitivity ensemble for an EcoSIM site. This Skill walks you from sourcing the correct config, through dry-run materialization and validation, to chunked array submission and scoring that matches the round's own targets. ## Quick Start Ask the agent to set up and validate an EcoSIM ensemble for your site, then submit it as a SLURM array and score the results against the round's targets.
