Family systems · child anxiety · parenting
Coverage at a glance
Twenty-five scenario families each contain a baseline, an exception, extended family context, and a repair/progression branch. Balance shows design intent—not clinical validity.
Model benchmarking should not begin until each item receives two independent qualified-clinician signoffs and material disagreements are adjudicated.
Mechanism balance
Family-system domains
Contrast design
Linked branch types
Cross-coverage
Family-system domain × branch type
Each cell contains five linked draft decision points.
Participant and risk design
Role, safety, and difficulty mix
Dataset explorer
Decision points and linked contrasts
Filter the draft corpus, inspect the reference standard, and move between all four branches in a scenario family.
Content validation
Clinician signoff
Review item validity separately from scoring a model response. Records are saved locally through a mocked API adapter.
Clinician-efficient design
From scenario-family validation to response labeling
The current app validates benchmark content. The next-response scorer uses four dimensions, one critical gate, one disposition, and no more than two optional failure tags.
Co-design families
A clinician author and separate editor define each family, critical distinction, acceptable alternatives, and uncertainty boundary.
Draft structure implementedValidate contrasts
Two independent clinicians verify each branch and the logic connecting baseline, exception, extended context, and repair/progression.
Interface implementedScore efficiently
Use four anchored dimensions plus a non-compensable gate and disposition. Run pairwise preference as a separate, faster task.
Scoring prototype updatedPilot and release
Measure agreement, burden, discrimination, context sensitivity, and score stability; freeze public/private splits at the scenario-family level.
Planned after validationMock endpoint contract
API activity
GET /api/v2/scenario-familiesLoad linked branches without leaking private holdout familiesPOST /api/v2/validationsCreate an immutable clinician signoff recordPOST /api/v2/candidate-responsesAttach blinded outputs and generation provenancePOST /api/v2/annotationsSave four scores, one gate, disposition, and optional failuresPOST /api/v2/preferencesStore a separate pairwise preference taskPOST /api/v2/adjudicationsResolve material disagreement without deleting raw ratingsPrototype handoff
Clinician-efficient scorer
Open the updated response-scoring prototype. It keeps item validation and response ratings distinct and demonstrates the reduced workload.