Regenerating Publication Outputs

PQID-Bench v1.0.0 distributes the frozen numerical evidence and the scripts needed to recreate publication-facing figures and tables. It intentionally does not distribute rendered manuscript figures, captions, editable figure sources, copy-ready manuscript tables, the manuscript walkthrough notebook, or the undeployed interactive gateway.

This boundary separates scientific reproducibility inputs from submission products. Regenerated files are local working outputs and are not part of the frozen public manifest.

Environment

From the release root:

python -m pip install -r requirements.txt

Main Regeneration Commands

The following scripts derive visual outputs from the archived data and analysis files:

python ".\scripts\build_clean_vs_mutation_stress_schematic.py"
python ".\scripts\build_pqid_bench_result_panels.py"
python ".\scripts\build_pqid_bench_inferential_figures.py"
python ".\scripts\build_pqid_bench_diagnostic_panels.py"
python ".\scripts\build_supplemental_method_expansion_figures.py"
python ".\scripts\build_pqid_bench_stochastic_repeatability_panel.py"

Additional specialized builders under scripts/ regenerate retrieval, signature-sensitivity, circuit-exemplar, and release-readiness panels.

Several analysis scripts can also emit tabular derivatives under a local tables_copy_ready/ directory. The canonical public evidence remains the corresponding JSON, JSONL, CSV, TSV, and Markdown analysis files under artifacts/; copy-formatted derivatives are not authoritative inputs.

Verification Boundary

ARTIFACT_MANIFEST.tsv covers only the distributed release. Files generated by these commands are expected to appear as untracked, ignored working outputs. Their absence from the archive does not affect evaluation replay, metric reconstruction, statistical analyses, or the lossless benchmark splits.