End-of-utterance classifier balanced accuracy (owner corpus)
JARVIS tells finished sentences from pauses 94.6% of the time (balanced) on the owner's own speech.
Single reading
94.6 % balanced accuracy
Protocol
Test definition
Text end-of-utterance model separating finished vs mid-thought utterances on the owner's own corpus; residual 4/37 finished utterances held, 0/11 mid-thoughts cut.
Limits
Limitations
- Date is when the figure was first recorded in the licence register (commit 89f8e9cf6); owner-specific corpus of 48 utterances; not a general benchmark.
Log
Recorded values
| Date | Value | Evidence | Samples | Source |
|---|---|---|---|---|
| 28 July 2026 | 94.6 % balanced accuracy | MEASURED | 48 | docs/architecture/MODEL_LICENCES.md |
Source paths refer to the private JARVIS repository and runtime records. They are listed for audit and are not published.