Interpreting Outputs
This guide explains how to read Autonima run artifacts and turn them into review decisions.
For file locations and folder structure, see the Outputs guide.
Which File Should I Open First?
| Question | Open this file first | What to check |
|---|---|---|
| What did search return? | outputs/search_results.json |
studies list and metadata completeness (PMID/title/abstract). |
| Why was a study included/excluded at abstract stage? | outputs/abstract_screening_results.json |
decision, reason, and confidence/criteria IDs. |
| Did full-text retrieval work? | outputs/fulltext_retrieval_results.json |
fulltext_available, pmcid, and full_text_path. |
| Why was a study excluded at full-text stage? | outputs/fulltext_screening_results.json |
decision, reason, confidence, and applied criteria IDs. |
| What are the final included studies? | outputs/final_results.json |
studies section (final included only). |
| Which studies had no full text? | outputs/unavailable_fulltexts.txt |
PMIDs that still have no retrievable full text. |
| Which studies had incomplete full text? | outputs/incomplete_fulltext.txt |
PMIDs screened as fulltext_incomplete. |
| What did annotation produce? | outputs/annotation_results.json |
Per-analysis annotation_name, include, and reasoning. |
| Are coordinates parsed and ready for NiMADS/meta? | outputs/coordinate_parsing_results.json + outputs/nimads_studyset.json |
Parsed analyses/tables and generated NiMADS payloads. |
Status Values You Will See
Screening and study records use stage-specific statuses:
pending: not screened yetincluded_abstract: passed abstract screeningexcluded_abstract: excluded at abstract screeningincluded_fulltext: passed full-text screening (final included set)excluded_fulltext: excluded at full-text screeningfulltext_incomplete: full text was present but incomplete/truncated for eligibility decisionsretrieval_failed: retrieval failedscreening_failed: screening call failed
Interpretation rule:
- A study is final-included only when status is
included_fulltext.
How to Read the Pipeline End-to-End
- Start with
outputs/search_results.jsonto confirm the search scope is correct. - Move to
outputs/abstract_screening_results.jsonto review early exclusions and reasoning quality. - Use
outputs/fulltext_retrieval_results.jsonandoutputs/unavailable_fulltexts.txtto identify coverage gaps before full-text screening. - Review
outputs/fulltext_screening_results.jsonfor final screening logic and criteria application. - Use
outputs/final_results.jsonfor downstream analysis inputs and reporting.
Important Caveat About final_results.json
final_results.json is intentionally filtered to final included studies in its studies list.
If you need exclusion audit trails:
- use
abstract_screening_results.jsonfor abstract-stage exclusions - use
fulltext_screening_results.jsonfor full-text-stage exclusions
Interpreting Confidence and Criteria IDs
confidenceis model-reported certainty. Low confidence can be auto-excluded if a stage threshold is enabled.inclusion_criteria_appliedandexclusion_criteria_appliedcapture criterion IDs used in the decision.- If criterion mappings are configured, these IDs should match your mapping keys (for example
I1,E2, or namespaced IDs).
Meta-Analysis Readiness Checklist
Before running autonima meta, confirm:
outputs/nimads_studyset.jsonexistsoutputs/nimads_annotation.jsonexists- Included studies in
final_results.jsonalign with your expected selection
Then run:
autonima meta <output-folder>/outputs
Meta outputs are created under:
<output-folder>/outputs/meta_analysis_results/