skills/build-report/specifications/technical-report.md

# Methods-focused reports

Use this reference when the reader needs to audit measurement, statistics, experiments, models, or data quality. It is not a required outline or a separate runtime mode.

State the result at the level the evidence supports. Define unfamiliar metrics, populations, units, exclusions, and baselines before relying on them. Explain methods, assumptions, uncertainty, validation, and failure modes where they affect interpretation. A methods-first structure is appropriate when the method itself is the question.

For experiments, distinguish design, estimated effect, uncertainty, and limitations. For forecasts, separate observations from projections and disclose assumptions. For data quality, show the affected population, severity, evidence, and limits of the check. Keep substantial reproducible code, fitting, and diagnostics in the companion notebook or source artifacts unless they are part of the requested reading experience.

Check that another technical reader can follow the logic without guessing how values were measured or what a result establishes. Include negative results and sensitivity checks when they could change the conclusion. Do not invent a recommendation merely to end with an action section.