aris-analyze-results

Point the skill at your results directory and it locates all JSON and CSV files, parses them into structured data, and builds a comparison table organized by hyperparameters and metrics. It reports mean and standard deviation when multiple seeds are present.

The skill always computes delta versus baseline as part of its comparison table step. Results are organized by independent variables like model type and data config, with relative improvement shown for primary and secondary metrics.

It parses JSON and CSV result files, searching common output locations such as figures/, results/, or project-specific output directories. Files are parsed into structured data before statistical analysis begins.

Yes, the statistical analysis step flags outliers and suspicious results, checks reproducibility across seeds, and identifies trends in parameter sweeps such as monotonic, U-shaped, or plateau patterns.

Output includes a raw data table, numbered key findings with observation, interpretation, implication, and next step for each, plus suggested follow-up experiments. Significant findings can be drafted into project documentation updates.