Eval

Last 0 runs

Run an eval

docker compose exec web npm run eval -- --name "baseline"
docker compose exec web npm run eval -- --arms retrieval   # no VLM spend
docker compose exec web npm run eval -- --name "after-feedback"

To see the learning loop move the number: run a baseline, resolve scans in the review queue, then run again with a different --name. Both land in the trend below.

Loading…