Scripts reference

Script	How to run	What it does
simulate.py	uv run python simulate.py params.json --slurm --wait	Submit or run one simulation; wait for results.json
launch.py	uv run python launch.py params.json	Set up a run directory and write a SLURM script, but do not submit
chain.py	uv run python chain.py config.yaml	Sweep one parameter across values via YAML, checkpoint-restarting between segments — how-to
sweep.py	uv run python sweep.py config.json	Sweep any combination of parameters from JSON (zip or cartesian) — how-to
sample.py	uv run python sample.py config.yaml	Space-filling batch of independent runs (LHS / random / grid / Sobol) — how-to
loop.py	uv run python loop.py config.yaml	Full multi-fidelity BO loop (DoE → train surrogate → suggest → repeat) — how-to
postprocess.py	uv run python postprocess.py runs/my_run/	Extract kLa, mixing time, vorticity, and shear-stress KPIs → results.json
collect_results.py	uv run python collect_results.py --sweep config/my_sweep.json	Aggregate all results.json files into a single CSV
plot_heatmaps.py	uv run python plot_heatmaps.py	Generate KPI heatmap figures from completed sweep results → experiments/figures/
render_videos.py	uv run python render_videos.py runs/my_run/	Convert raw frame dumps to MP4 (called automatically by SLURM jobs)
suggest.py	uv run python suggest.py exp_dir param_space.yaml	Print the next highest-EI parameter point to stdout
train_surrogate.py	uv run python train_surrogate.py exp_dir model.pkl	Train the multi-fidelity surrogate from existing run data
stage_segment.py	uv run python stage_segment.py <prev_run_id> <next_run_id>	Safely stage a chain-restart checkpoint — refuses unless the predecessor has a genuine results.json

See Project structure for the full annotated tree, including the diagnostic one-off scripts not listed here.