AOI Robustness Reporting Bundle

Purpose

A sensitivity result should be reviewable as an evidence package, not only as a narrative sentence. Keep the prespecified plan, geometry audit, assignment summaries, model outputs, failures, provenance, report text, and diagnostic figures together.

Recommended bundle

• analysis-plan.yml — declared perturbation/model/failure plan;
• perturbation-audit.csv — branch completion and geometry failures;
• assignment-stability.csv — measurement-level robustness;
• assignment-table.csv — observation-level branch assignments;
• model-results.csv — branch coefficients, intervals, convergence, and N;
• inference-stability.csv — term-level robustness summary;
• failures.csv — retained non-evaluable branches;
• provenance.txt or JSON — source/AOI hashes and analysis provenance;
• report.md — manuscript-oriented narrative;
• diagnostic figures and a file manifest.

Export from a result

out_dir <- "workflow-output/aoi-reporting-bundle"
dir.create(out_dir, recursive = TRUE, showWarnings = FALSE)

utils::write.csv(result$grid_result$audit,
                 file.path(out_dir, "perturbation-audit.csv"), row.names = FALSE)
utils::write.csv(result$stability$overall,
                 file.path(out_dir, "assignment-stability.csv"), row.names = FALSE)
utils::write.csv(result$assignment_table,
                 file.path(out_dir, "assignment-table.csv"), row.names = FALSE)
utils::write.csv(result$models,
                 file.path(out_dir, "model-results.csv"), row.names = FALSE)
utils::write.csv(
  assess_aoi_inference_stability(result, term = "condition"),
  file.path(out_dir, "inference-stability.csv"), row.names = FALSE
)
utils::write.csv(result$failures,
                 file.path(out_dir, "failures.csv"), row.names = FALSE)
writeLines(report_aoi_sensitivity(result), file.path(out_dir, "report.md"))

If a cryptographic manifest is required, create it with the repository or archival tooling used by the project and record the hashing algorithm. Do not add a package dependency solely to hash this example.

Reviewer-facing reading order

1. analysis plan;
2. perturbation/failure audits;
3. assignment stability and figures;
4. model results, convergence, intervals, and N;
5. provenance;
6. narrative report.

Interpretation

The bundle demonstrates what was run and what remained stable within the declared AOI alternatives. It does not prove that the nominal AOIs are scientifically correct or that the perturbation set captures every source of measurement uncertainty.

Privacy and limitations

Observation-level assignment tables can contain participant/trial identifiers. Review them before external release. Geometry sensitivity also does not replace calibration-quality, event-detector, missingness, or estimator-specific diagnostics.

API map

Use run_aoi_sensitivity_analysis() for the core result, assess_aoi_inference_stability() for model-level robustness, report_aoi_sensitivity() for the narrative draft, and the four AOI plotting functions for visual evidence.