Metrics

Bucketed readings, so you can chart quality beside your own numbers.

One endpoint, returning one measure over one window, already bucketed. Every bucket is UTC.

bash

curl "https://app.evidova.com/v1/rollups?metric=...&grain=hour&from=...&to=..." \
  -H "authorization: Bearer ak_live_..."

The parameters

metric	Required. Which measure to read
grain	minute, hour or day
agent_id, client_id, channel	Narrow it to one agent, one client, or voice or chat
from, to	The window to read

metric, grain	Echoed back, so a stored response says what it is
from_ts, to_ts	The window actually served
buckets	The readings, one per interval
agent_id, client_id, channel	The filters in force, null where none was given

Buckets are UTC, so a daily chart in another timezone needs re-bucketing on your side, not an offset on the axis.
A grain finer than the data available is coarsened rather than refused, and the response says which grain it actually served.
An empty buckets array means no traffic in that window. It does not mean zero quality.

This is the same data the Metrics screen draws, so a figure you chart here and a figure a colleague reads on screen agree by construction.