Get evaluation overview
Returns one evaluation dashboard page, read at one consistent point in time. Current and prior windows are adjacent and equal; gaps in history produce explicit null coverage fields rather than misleading zeroes.
Authorizations
A Recursion API key, created in the console under API keys.
Query Parameters
Length of the current window. The prior window is the adjacent interval of equal length.
7d, 28d, 90d Immutable evaluation-agent version whose rubric and verdicts define this overview.
Optionally narrow page-level metrics, table rows, and retained history to one target agent.
Repeat to explicitly compare up to five target agents; absent values select the page target or the first five whole-filter agents.
5One canonical UUID identifying an organization-scoped comparison agent or immutable version.
Repeat to add scoped previous target-agent versions; automatic latest versions plus these values may total at most twelve traces.
11One canonical UUID identifying an organization-scoped comparison agent or immutable version.
Verdict population plotted for each selected target-agent version.
overall, criterion Rubric key plotted when series_metric is criterion; required exactly for that metric.
Opt into bounded adaptive chart positions from exact timestamps through days. Omit for the legacy fixed daily grid.
adaptive Maximum target-agent rows to return. Defaults to 25.
1 <= x <= 100Opaque signed continuation bound to the organization, frozen watermark, range, evaluator, singular table target, and effective limit. Comparison agents, versions, and metric may change without resetting the table page. Expires after one hour.
Response
Complete fixed-window evaluation Overview plus an optional signed target-agent continuation.
Complete fixed-window evaluation Overview plus an optional signed target-agent continuation.
Ranked target-agent metric rows for this signed cursor page.
Durable analytics watermark freezing every query in this response.
Immutable agent-version and model markers when one target agent is selected.
Whether markers are complete or require selecting one target agent.
complete, select_target_agent Bounded multi-agent and immutable-version traces for one selected metric.
Stable key order used to align every agent row's criterion_metrics array.
Aligned criterion verdict series for the complete frozen window; legacy responses use a daily grid.
Inclusive UTC start of the current Overview window.
Inclusive UTC start of the adjacent equal-length prior window.
Exclusive UTC end of the prior window, equal to from.
Fixed headline coverage, verdict, criterion, and exact-cost metrics.
Exclusive UTC end of the current window, equal to as_of.
Adaptive position semantics: exact timestamps are criterion-local, while interval granularities use shared aligned positions; omitted for the legacy daily response and empty adaptive collections.
exact, second, minute, hour, day Signed one-hour continuation for the next ranked target-agent page at the same watermark.