Datasets·Medicine·therapy_donovan
Donovan therapy session — sentence-level Health Lens coding
A 70-minute audio therapy session (Donovan, 2026), transcribed and projected through the Health Lens at the sentence level. Pure qualitative artefact — no regression model is fit; the value is in observing how PsyProxy surfaces therapy-level meaning inside raw clinical text. Coded with scripts/donovan_exploratory_coding_v3.py .
Distribution of Speaker (Patient vs Therapist)
1
2
51 at floor50 at ceiling
101
items
20
holdout n
Speaker (Patient vs Therapist)
target
qualitative
kind
0
systems compared
Criterion validity
Reported holdout systems from the verified card
Reported score uses score as the task-primary metric. Secondary columns keep the companion metrics visible so binary, ordinal, regression, and multiclass cards are not compared through one flattened score.
Model-family mix
SystemFamilyVariantscorePrimary scale