Quality measurement

Synthesis accuracy

We measure synthesis quality against a versioned benchmark corpus of realistic interview notes with expert-defined pass criteria. Results are recomputed on every deploy — a failing benchmark blocks the release.

Rule-based engine, single-interview corpus

11/12

criteria passed

92%benchmark pass rate

Try live demo →

Benchmark: Churn (structured)

4/4

Passed: Surfaces approval or workflow friction as a pain point
Passed: Mentions reporting or Excel/export pain
Passed: Captures a churn-related quote
Passed: Includes a follow-up or next step

Benchmark: Field notes (messy)

3/4

Rule-based often scores 3/4 here — Compare in the live demo shows when AI closes the gap. Try live demo

Passed: Identifies setup or configuration friction
Passed: Surfaces mobile or iPad preview issue
Passed: Includes a user quote or quote-style insight
Failed: Notes audit log or export strength (often needs AI on messy input)

Benchmark: Enterprise HR (structured)

4/4

Passed: Surfaces calibration or performance review pain
Passed: Mentions permissions or role-based access
Passed: Captures mobile/Android sign-off issue
Passed: Includes at least one positive signal

Try the live demo to run Compare on messy notes and see rule-based vs AI side by side. Try live demo