Calls
Search by date, by the number that rang, by what was said, or by all three at once. Every result carries its recording, its speaker-labelled transcript, the mark against each KPI, and the decision that follows from them.
Upload recordings
Drop audio here — wav, mp3, opus, flac, m4a and the telephony formats a recorder exports; anything the model cannot read directly is converted first. A CSV manifest alongside it carries the call time, the CLI and the agent; without one the filename is parsed, and failing that the call is still kept and dated by its upload. Every recording accepted here is written straight into the recording drop folder, so what the reviewer sees and what sits on disk are the same set of calls.
Standards
What the calls are judged against. Upload the policies and scripts, define the KPI set and the category catalogue, and set the thresholds — none of it needs a release, and a change here affects the next evaluation rather than rewriting the scores already on the books.
Analytics
Everything scored so far, aggregated — including how often a reviewer disagreed with the machine, which is the only honest measure of whether these scores can decide anything.