Evals
Is the change better?
Run an agent on a fixed set of cases, score every answer by hand and compare it with the baseline. There is no automatic scorer — every score in Orbit comes from a person.
Loading …
Review queue
3 answers wait for a human score
Experiments
One row per run of a dataset. Open a row to compare it case by case with the baseline or another experiment.
| Agent · change | Dataset | Cases | Avg score | vs baseline | Pass rate | Date |
|---|---|---|---|---|---|---|
| SEO checkPrompt asks for the file and the concrete fix per finding | Key pages | 5/5 | 4.0 | +18% | 80% | 10/02, 03:40 PM |
| Ask the projectFirst version of the prompt | Common questions | 2/4 | 4.5 | Baseline | 100% | 09/28, 09:05 AM |
| SEO checkFirst version of the prompt | Key pages | 5/5 | 3.4 | Baseline | 40% | 09/24, 10:12 AM |
Datasets
The cases each agent is tested on — written by hand or taken from real runs on the Runs page.
Key pagesSEO check · 2 experiments5 cases
The five pages that bring most visitors. A good answer names the file and the fix for every finding.
Common questionsAsk the project · 1 experiment5 cases
Questions the team asks every week. A good answer cites its source and says when it could not check something.
1 from real runs