Skip to main content
Everything the web UI does goes through this API, so a script or coding agent can do it too: start runs, read results, add annotations or export. It’s served by ezvals serve at http://127.0.0.1:8000. Requests and responses are JSON. For reading saved results without a server, ezvals query and ezvals run --json are usually simpler.

Run JSON

Endpoints that return a run use the same shape as ezvals run --json and ezvals export -f json:
  • created_at is when the run last started (Unix seconds).
  • trial, trial_of and regradable appear only when they apply. Runs with trials also have trials, pass_at_k and pass_all_k.
  • status is one of not_started, pending, running, completed, error or cancelled.
  • correction_history lists edits made in the UI or through this API: field (annotation or scores), before, after and timestamp.

Results

run_id can be latest for the active run. index is the result’s 0-based position in results. /results and /api/runs/{run_id}/data add score_chips (per-score summaries) and eval_path. For the active run they also add is_paused, selected_total, and discovery_error when finding or importing the evals failed. A PATCH body looks like this:
Only annotation and scores can be changed. Each change adds an entry to correction_history. Works on any saved run, not only the active one.

Running

Starting a run or regrade while one is in progress returns 409.

Sessions and runs

Export

Settings

Pages