Skip to content

Structured Results

Summary

Structured results are machine-readable records that summarize statistical outputs in a consistent shape. PyFLASH uses them for report digests, pipeline manifests, and function describe coverage.

Where PyFLASH Uses It

  • Group-comparison, correlation, regression, linear-model, cosinor, and acrophase functions can emit report records when the collector is active.
  • Pipelines write manifest.json files that describe inputs, options, output paths, selected results, skipped results, and run metadata.
  • The function registry marks which functions have reviewed structured descriptions, which are exempt, and which remain unreviewed.
  • Report records describes the output-facing record format.

Inputs

Report records are built from the same objects used to create plots and tables: group summaries, pairwise test rows, correlation coefficients, model coefficients, adjusted means, and pipeline result summaries.

The report collector is opt-in. When it is not active, emit calls are inert and the analysis still runs normally.

Outputs

Structured output can include:

  • kind: the record type, such as group comparison, correlation, or linear model.
  • descriptive statistics and test statistics.
  • p-values, q-values, effect-size strings, and pairwise rows.
  • direction and headline text generated from numeric results.
  • pipeline manifests with saved paths, selected rows, skipped rows, and option settings.
  • digest text rendered from collected records.

Large tables are summarized before being stored in records. DataFrame payloads are converted to JSON-safe summaries with shape, columns, and a bounded set of leading rows.

Interpretation

Structured records are summaries of the actual tables and figures. Use them to trace what was run, compare options across runs, or generate a report digest. For detailed interpretation, open the corresponding CSV tables and read the specific statistics page for the method.

The describe registry is a coverage tool. Functions in the covered set have reviewed structured descriptions. Exempt functions are intentionally excluded from describe coverage. Unreviewed functions are known gaps rather than hidden methods.

Limitations

Report records are not a complete audit log of every intermediate calculation. They are intentionally compact and may truncate large DataFrame previews. A digest is generated text from structured fields; it should not replace checking sample sizes, filters, skipped rows, and raw output tables.

Manifests describe a run but do not validate that a statistical design is appropriate. They should be read with the method-specific pages and the original analysis plan.

See Also