gen-report-detailed

To generate a reproducible experiment report, provide the experiment name and root directory to collect training logs, CSV metrics, model configs, and git metadata, producing a timestamped markdown file with metric provenance and failure analysis.

Metric provenance in experiment reports records file paths, line numbers, and step indices alongside formal mathematical definitions, ensuring researchers can trace reported metrics back to their source data for full reproducibility.

Failure analysis on training anomalies uses collected CSV metrics, training logs, and artifact summaries to diagnose irregularities, producing a structured 13-section report with a reproducibility checklist and detailed findings.

Yes, the report generation process collects SLURM or run scripts and available PNG visualization files alongside model configurations and artifacts, integrating them into the final timestamped markdown report for comprehensive documentation.

The best way to prepare a reproducibility bundle is to collect training logs, CSV metrics, git metadata, and visualizations, enforcing metric provenance with formulas to produce a complete 13-section experiment report.

The generated experiment report is written to a timestamped markdown file, containing 13 sections with collected logs, metrics, artifact summaries, and a reproducibility checklist for post-run forensic analysis.