# ariadnev eval

This is the reference for the `ariadnev eval` command, projected from the release.

Score kit quality — tier-1 static; tier-2 behavioral suite; optional tier-3 LLM judge

## Options

| Options |  |
| --- | --- |
| `--concurrency <count>` | maximum parallel isolated runs |
| `--deep-repeats <count>` | repeats for every golden task |
| `--model <name>` | pinned model identity |
| `--runner <json-argv>` | strict JSON argv array; prompt is sent on stdin |
| `--runtime-provider <name>` | pinned runtime provider identity |
| `--runtime-version <version>` | pinned runtime version identity |
| `--skill <name>` | only evaluate one skill (bare or av: prefixed) |
| `--skill-repeats <count>` | repeats for every skill routing cell |
| `--suite` | run the tier-2 behavioral scenario suite |
| `--timeout-ms <milliseconds>` | per-run timeout |
| `--variant <name>` | benchmark variant: ariadnev or reference |

[Back to CLI index](/en/stable/reference/cli/)