Agent performance
Agent performance measures how each agent configuration (agent, model family, persona and review tier) does on your organization's tasks: success rate, first-pass acceptance, defects, rework, completion time, cost per success and security findings, each rate with a Wilson 95% interval. A configuration with fewer than 20 judged outcomes is not ranked. Every endpoint requires the Org Owner or Org Admin role and reads your own organization only.
Compare agent configurations
Takes 2 to 4 `config` keys and returns `shared_strata` (the persona and complexity-bucket pairs every configuration has a judged outcome in) and each configuration's metrics over those strata only. With no shared stratum every `metrics` is null. Requires the Org Owner or Org Admin role (else 403).
Export the outcome ledger as CSV
Returns every outcome ledger row of your organization as CSV, one line per row, with its attribution links as `kind:state:confidence` joined by `;`. Requires the Org Owner or Org Admin role (else 403).
Get the agent leaderboard
Returns the ranked configurations, best success rate first, each with its `rank`; a configuration with fewer than `min_n` (at least 20) judged outcomes is left out. Takes `persona`, `project_id`, `period` and `language`. Requires the Org Owner or Org Admin role (else 403).
Get agent performance profiles
Returns one group per agent, model, configuration or project (`group`; `org` for the whole organization), each with `n_final`, `ranked`, `insufficient` and the seven metrics: `success`, `first_pass`, `defects_per_100` and `security_per_100` with `value`, `lo`, `hi` and `n`; `rework`, `completion` and `cost_per_success`. Filters: `persona`, `project_id`, `period` (`30d`, `90d`, `365d` or `all`), `language` and `min_n`. Requires the Org Owner or Org Admin role (else 403).