What problem does it solve? Long-running stochastic compute jobs (bootstraps, permutation nulls, Markov batteries, MICE refits, large ETL) often launch with guessed parallelism, no checkpointing, and no wall-time budget, leading to stalled runs, wasted hours, and unrecoverable failures. This Skill produces a defensible resource plan before launch. ## Core Features & Use Cases - Production-Entry-Point Benchmarking: Sweeps worker counts 1→N against the real script at realistic scale, flagging sub-scale estimates as PROVISIONAL. - Structured Preflight Record: Emits a dated JSON record capturing data scale, candidate worker counts, selected strategy, heartbeat cadence, and wall-time estimates. - Platform-Specific Guidance: Covers WSL background process survival, Windows PowerShell runner pitfalls, and GIL/backend constraints for kernels like exact Wasserstein-2. - Use Case: Before launching a permutation-null battery with B=1000 at landmark count 5000, run the preflight to measure per-diagram timing, select the optimal loky worker count, and record a checkpoint/resume plan with a wall-time budget. ## Quick Start Ask the AI to run a resource preflight for your planned bootstrap or permutation job, including a worker-count sweep and a written wall-time estimate before launch.
