Skip to content

Worked example: AOI robustness reporting bundle

This synthetic example packages the existing five-AOI disclosure-dwell sensitivity workflow into a deterministic manuscript/reviewer bundle. It adds no new estimator or public API.

Executable source: examples/aoi_reporting_bundle.py.

Run

Run examples/aoi_reporting_bundle.py with an output directory for the reporting bundle.

Bundle contents

  • analysis-plan.yml
  • perturbation-audit.csv
  • assignment-stability.csv
  • assignment-table.csv
  • assignment-frequency.csv
  • model-results.csv
  • inference-stability.csv
  • failures.csv
  • provenance.json
  • report.md
  • aoi-geometry.svg
  • aoi-assignment-stability.svg
  • aoi-coefficient-stability.svg
  • manifest.json

What the manifest records

For each evidence file, manifest.json records the relative path, evidence role, byte size, and SHA-256 hash.

The bundle preserves the observation level, overlap policy, analysis evidence, failures, and provenance required to audit the declared AOI sensitivity workflow.

Robustness frequencies are descriptive summaries of the declared perturbation set. They are not probabilities that a substantive conclusion is true.

Reviewer use

Review the declared analysis plan first, then the perturbation audit, assignment stability, model results, failures, provenance, and diagnostic figures.

This separates geometry uncertainty, assignment sensitivity, model sensitivity, and execution failures instead of collapsing them into one robustness percentage.

Limitations

The bundle documents robustness only to the declared AOI alternatives. It does not establish that the nominal AOI is objectively correct or that all plausible sources of measurement uncertainty have been evaluated.

This is a synthetic reporting template, not an empirical result.

Privacy and sharing

The synthetic example contains no private data. Real-data bundles should be reviewed for participant identifiers, trial identifiers, and other sensitive information before external sharing.

Continue to the reporting-bundle guide, worked AOI sensitivity example, failure clinic, and AOI perturbation API reference.