Validation / Technical
Validation and limits
Reported testing scope for v48.0. Replay testing, live recording and feature-specific validation are distinguished.
The record develops while surgery continues.
Testing approach
Hundreds of hours of replay testing across numerous cases from the veterinary surgery database, plus live recordings on the appliance itself.
Each tested case passes through the pipeline and is compared with a hand-built account of what was actually said. Case outputs are compared before changes ship. 184 automated regression tests run before each release.
Processing measurements
Single-camera measurements on the appliance. Draft readiness excludes clinician review and approval.
| Case length | Work per 30-second interval | Stop to finished draft |
|---|---|---|
| 31 minutes | 16 seconds | 69 seconds |
| 64 minutes | 13 seconds | 92 seconds |
| 2 hours | 16 seconds | About 2 minutes |
Interruption and repeatability
Recovery was demonstrated after a live-case interruption. Case state is persisted at Stop and can be used to rebuild the record. Repeated processing produces the same record and ledger identifiers; entries identify the software version.
What is not established
Monitor extraction has been demonstrated on captured frames, not on a live theatre monitor. Dual-camera timings and the previous four-hour dual-camera benchmark are not verified for the shipping appliance.
Recall and precision measurements from replayed room audio are not published as clinic-audio accuracy. No recall percentage is claimed here.
Clinical review remains necessary
The record depends on audible speech. Misheard drug names can reach the draft and need clinician review. Missing clinical values cannot yet be typed into the appliance’s record fields and are completed in the practice system after export.
Review the record structure