| Term | What it is |
|---|---|
| Transcript | The record of one conversation the agent actually had, with every turn and trace. |
| Evaluation | A criterion a model scores past conversations against, turning quality into a number you can track. |
| Test | A scripted conversation replayed against the agent, with checks asserting what it does. |
| Analytics | Aggregate usage and cost: tokens, calls, conversations, unique users. |
| Credits | The unit Voiceflow bills usage in. |
| Test turn | One side of a scripted conversation: either what the user says, or the agent reply the checks assert against. |
| Test check | One assertion about an agent turn. Compares the reply itself, or the behaviour behind it such as which tool was called. |
| Test run | One execution of a test, carrying how many of its checks passed. Runs are asynchronous, so poll until the status settles. |
| Transcript property | A label you define once and then set on individual conversations, so your own outcome data can be filtered alongside Voiceflow’s. |
Measure
Transcripts, evaluations, tests, analytics and credits: the words for reading back what an agent actually did, and what it cost to do.