Python/R analysis, plotting, and reports

Downstream analysis starts from finalized stage/manifest.json. The AI may generate a Python or R script for the data and objective, but the executable object is first written to the current Job's reproducibility directory and then run in the isolated container.

Analysis contract

Check sample counts, groups, missingness, matrix orientation, metadata keys, and file formats first; record each choice, default, and limitation. Statistical reports should distinguish biological from technical replicates and report effect sizes, intervals, multiple-testing policy, and unsupported inferences. Pin R packages, Python environments, and random seeds.

Plotting and reports

Regenerate figures from retained data objects. Include axes, units, sample counts, groups, statistical annotations, and color-accessible legends. Retain the source-data summary, script, plotting parameters, generation time, and output manifest with HTML/PDF/PNG deliverables so every artifact is traceable to the same inputs.

Retention and boundary

Generated scripts stay under reproducibility/generated_scripts/, analysis results under analysis/, and logs under logs/ or the registry's log list. Downstream reads only read-only stage/; it cannot modify upstream files, access the external network, or read the authoring workspace. Save stdout/stderr and exit code on failure; do not mark an empty plot or partial report as successful.