Results Reporting

Outcome

Turn user-owned metrics and analysis artifacts into a report a stranger can trust. Every quantitative claim must be traceable to data, validation, and reproduction evidence. Separate observed results from interpretation, and separate predictive performance from explanation or causality.

A results report tells the story of one evaluation. A model card is the durable operating contract; an experiment log is the trial history.

When to use this skill

Use it for notebook summaries, research notes, README result sections, internal reports, public writeups, and final answers after an analysis. If there are no metrics or baselines yet, stop and produce the evaluation artifacts first.