Second opinion
Your eval score went up. Is it real?
Upload the per-case scores another tool gave your system before and after a change. Oloproof says whether the rise is real, how many more cases would settle it, and whether anyone has checked the judge that scored them. No account, nothing to install.
What a file holds
| Column | What it holds |
|---|---|
| case_id | The case's id. The two files pair only on an id both contain. |
| criterion | The metric's name. |
| score | A number on the metric's scale, or empty if the tool did not score that case. |
| passed | true or false instead of a score, or empty. |
| score_min, score_max | The scale a score lies on (0 and 1 for most tools), required for a score. |
| judge, judge_version | The model that judged, if a model did. Optional. |
A short script turns each tool's results into this file: DeepEval, Braintrust, LangSmith and Langfuse. Recipes for each tool
What happens to them
Checked in your browser. A file with a column beyond these is refused before anything is sent, so no question or answer ever leaves your machine.
Decided by the engine. The same one oloproof scores import runs: each file becomes a run, the two are compared case by case, and each criterion is tested for a rise above zero.
Kept nowhere. The files and the runs live in a temporary store, deleted before the answer comes back. Up to 2 MB and 20,000 scores per file.