Skip to main content

Evaluation dashboard

Every evaluation has a Dashboard tab, alongside Runs and Query Lab, that tracks how your relevance metrics move across the runs of that evaluation. It answers the question the runs list can't at a glance: is search quality trending up or down as you iterate?

This is the per-evaluation dashboard. It's different from the workspace home dashboard (described in Getting started), which summarises recent runs, usage, and run status across all of your evaluations. The Dashboard tab is scoped to one evaluation and is entirely about its metrics over time.

Latest run metrics​

The top of the dashboard shows one card per metric for the most recent completed or locked run, each with the change versus the previous run - coloured green for an improvement, red for a regression - as both an absolute and a relative delta. It's the at-a-glance "did my last change help?" summary.

Below the cards, each metric is plotted over time, one chart per metric. A few things worth knowing about the trend charts:

  • Only completed and locked runs appear. Pending, running, cancelled, and failed runs are not plotted. Locking a run keeps it in the trend.
  • The x-axis is real time, not a run counter - runs are placed by when they were created. A time-range picker narrows the view to a rolling window (say the last 7 days) or an absolute date range.
  • Configuration changes are marked. Where a run's configuration (body, query string, content type, candidates mapping, headers, or tags) differs from the run before it, the chart drops a "config changed" marker - so a step up or down in the line can be tied to the change that caused it. This is the same per-run override you set when creating a run.
  • The metrics plotted are the union of the metrics configured across the evaluation's runs. A run that didn't compute a given metric simply leaves a gap in that metric's chart.
curl "https://${RELEVAL_HOST}/api/v1/evaluations/${EVALUATION_ID}/metrics/trends" \
  -H "Authorization: Bearer ${TOKEN}"

See the Metric Trends endpoint for the response shape.

The dashboard trend is the all-runs view. When you want to dig into a single change, compare two runs head-to-head: that view adds per-query metric deltas, how much the result lists changed (RBO, Jaccard, and candidate churn), and execution-time differences.