What problem does it solve?
Screening hundreds of papers for a literature review often produces misclassifications when relevance criteria are vague or untested. This Skill builds a scoring rubric collaboratively, validates it against user-labeled test papers, and iterates until accuracy reaches at least 80% before bulk screening.
Core Features & Use Cases
- Collaborative Rubric Design: Brainstorms keywords, data types, paper types, and edge cases with the user, then encodes them as weighted scoring rules in screening-criteria.json.
- Test-Driven Validation: Fetches candidate abstracts via Scopus and OpenAlex, collects user judgments on 5-10 papers, scores them with the rubric, and reports false positives and false negatives.
- Bulk Screening and Re-screening: Applies the validated rubric to full result sets, caches abstracts, and re-scores all cached papers when criteria change, producing a diff of status changes.
- Use Case: A researcher screening 127 papers on tuberculosis drug resistance builds a rubric, fixes misclassifications around analog and derivative terminology, reaches 90% test accuracy, then screens the full set and re-screens after adding combination-therapy terms.
Quick Start
Help me build a screening rubric for my literature search on tuberculosis drug resistance and validate it on test papers before screening all results.
