What problem does it solve? Long offline calibration campaigns for process-based environmental models (EcoSIM, PFLOTRAN, ATS, ELM-FATES) drift over many experiment cycles: monitors go unarmed after simulation launches, state files go unvalidated, commits pile up unpushed, and round summaries ship without a next-round plan. This Skill is the definition-of-done checklist that keeps every cycle and round consistent. ## Core Features & Use Cases - Per-cycle checklist: Verifies each phase is logged with its phase skill into log/{stem}.md plus a self-documenting phase_results/{stem}/ folder, that analysis is first-hand, that tape variables are reduced per their own temporal semantics, and that workflow_state is updated, validated, committed, and pushed after every phase. - Per-round checklist: Enforces the ordered round-close sequence (summarize-calibration-round, compare-calibration-rounds, write-report) and requires the round summary to propose a concrete next-round work plan covering parameter add/remove, bounds recentering, base updates, and residual routing. - Config-driven loop limits: Reads A2MC_MAX_SKIP_TESTING, A2MC_MAX_EXPERIMENTS, and A2MC_CONFIDENCE_THRESHOLD from the sourced machine config rather than hardcoded literals. - Use Case: While running a multi-round EcoSIM calibration on an HPC cluster, re-check this Skill at every phase transition to confirm monitors are armed after each Phase-5 launch, the rethink protocol is answered on 6-to-3 routings, and the cycle-end synthesis report exists. ## Quick Start Ask the agent to run the calibration-discipline checklist against the current experiment cycle to verify every phase was logged, validated, committed, and pushed before advancing the loop.
