How tukanzee/avt-testing works
Clinical Ground Truth Runner supports synthetic doctor–patient role-play, recording spoken facts, and exporting a ground-truth record. It also offers transcript-led AVT review and a separate Level 1 validator. The graph follows the documented workflows and sampled call/import relationships; the validator page and several of its internal stages were not excerpted, so their detailed wiring is not asserted.
Main entry points: TranscriptAvtValidator.tsx Transcript AVT workflow · App.tsx Application routing · ClaimReview.tsx Claim review · ValidatorPage.tsx Validator entry
Components 22
Role-play scenarios
- Application routing
src/App.tsx - Case catalogue
src/data/cases.ts - Clinical case scripts
src/data/cases/dka.ts
Ground-truth recording
- Spoken fact capture
src/App.tsx - JSON export
src/utils/json.ts - PDF export
src/utils/pdf.ts
Transcript-led validation
- Validator entry
src/validator/ValidatorPage.tsx - Transcript AVT workflow
src/validator/TranscriptAvtValidator.tsx - Transcript parsing
src/validator/transcript/parseTranscript.ts - Audio preparation
src/validator/components/AudioTranscriptPreparation.tsx - Speaker attribution
src/validator/speakerAttribution/localAttribution.ts - Local transcription
src/validator/transcription/localWhisper.ts - Evidence chunking
src/validator/transcript/createEvidenceChunks.ts - Atomic claim extraction
src/validator/claimExtraction/atomic.ts - Evidence matching
src/validator/matching/evidenceMatcher.ts - Local semantic models
src/validator/browserSemantic.ts - Claim review
src/validator/components/ClaimReview.tsx - Omission review
src/validator/components/OmissionReview.tsx - Findings review
src/validator/components/FindingsReview.tsx
Level 1 validation
- Level 1 analysis
src/validator/level1Engine.ts
Other components
- Doctor / actor
- Human reviewer
Connections 26
- Doctor / actor → Application routing: uses
- Application routing → Case catalogue: loads cases
src/App.tsx - Case catalogue → Clinical case scripts: catalogues
src/data/cases.ts - Application routing → Spoken fact capture: records spoken facts
src/App.tsx - Spoken fact capture → JSON export: exports JSON
src/App.tsx - Spoken fact capture → PDF export: exports PDF
src/App.tsx - Application routing → Validator entry: opens validator
src/App.tsx - Validator entry → Transcript AVT workflow: offers workflow
- Transcript AVT workflow → Transcript parsing: parses transcript
- Transcript AVT workflow → Audio preparation: prepares audio
- Audio preparation → Local transcription: transcribes locally
- Speaker attribution → Local transcription: uses transcription
src/validator/speakerAttribution/localAttribution.ts - Transcript AVT workflow → Speaker attribution: attributes speakers
- Transcript AVT workflow → Evidence chunking: builds evidence
src/validator/transcript/createEvidenceChunks.ts - Evidence chunking → Atomic claim extraction: extracts atomic claims
src/validator/transcript/createEvidenceChunks.ts - Transcript AVT workflow → Evidence matching: retrieves evidence
- Evidence matching → Local semantic models: checks entailment
- Transcript AVT workflow → Claim review: presents claims
- Claim review → Evidence chunking: shows context
src/validator/components/ClaimReview.tsx - Claim review → Omission review: reviews omissions
src/validator/components/OmissionReview.tsx - Transcript AVT workflow → Findings review: reviews findings
- Human reviewer → Claim review: reviews evidence
- Human reviewer → Omission review: confirms omissions
- Human reviewer → Findings review: reviews findings
- Validator entry → Level 1 analysis: runs Level 1
- Level 1 analysis → Local semantic models: judges statements
src/validator/level1Engine.ts