2026 · Q1
User-reported outcomes after 8 weeks of conversation with Calyn
Pre-registered cohort study (n=1,204) measuring PHQ-9 and GAD-7 movement against a waitlist control.
OutcomesPre-registered
Back to home
Research
Evidence over vibes.
We publish what we learn. Quarterly clinical reports, open-sourced evaluation suites, and outcome studies.
If you can’t measure it, you can’t improve it. And if you don’t publish it, no one can hold you to it.
Every quarter, our research lead publishes a report on how Calyn is performing across three axes: outcomes (are people actually feeling better?), safety (are responses adhering to clinical practice?), and equity (does the experience work across cultures and contexts?).
Latest publications
2025 · Q4
Safety benchmarking for therapeutic AI: a clinician-authored evaluation suite
We open-sourced the 412-scenario safety eval used to score model updates.
SafetyOpen
2025 · Q3
Cross-cultural calibration of therapeutic conversations
Comparing model responses across eight cultural contexts.
EquityMethodology
2025 · Q2
Crisis detection and warm hand-off: protocol design and live performance
How Calyn detects moments needing escalation, and what happens after.
CrisisProtocols
What we measure
- PHQ-9 and GAD-7 movement over 4- and 8-week windows, against a matched waitlist control.
- Adherence to clinical protocols across CBT, ACT, IFS, and motivational interviewing.
- Safety incident rate, including missed crisis detections (we aim for zero, audited weekly).
- Hand-off success, when Calyn refers someone to a human therapist or crisis line.
Open science
We pre-register our outcome studies and publish methodology before we publish results. Our safety evaluation suite is open-sourced for the field. Your conversations are not anonymized and are not shared with outside researchers — they stay linked to your account and are handled under our privacy policy. Get in touch if you want to collaborate.