Testing and evals
| Suite | Result |
|---|---|
| Scripted orders | 30 / 30 |
| Rogue agent | 29 / 29 (+ 2 client approval) |
| Match eval | 52 / 52 · calibration ~0.01 |
| pytest | ~1,000 tests |
make seed
uv run python -m evals
make test
uv run python -m evals.match.run
Replay modes (live / record / replay) keep demos and CI offline. Recordings live in gateway/recordings/ and engine/recordings/.
Next: wire the same suites into Langfuse datasets and CI gates so every policy/MCP change stays at 30/30 and 29/29.