Running & reading results
Run one simulation or a whole suite, choose draft or published behavior, and read the verdict, assertion results, and full conversation trace.
Once a simulation is saved, you can run it as often as you like and inspect the results in detail.
Run one, or run all
- Run one — execute a single simulation to check a specific scenario.
- Run all — execute every simulation as a suite. This is how you catch regressions: one click confirms a change didn't break anything else.
Draft or published behavior
You choose what the simulation runs against:
| Mode | Runs against | Use for |
|---|---|---|
| Pinned | The published version | The authoritative result |
| Draft | Your current working draft | Checking edits before you publish |
A draft run never becomes the official status of a simulation — the pinned result is always the verdict of record. This lets you experiment freely while keeping a trustworthy history.
Reading a run
A completed run shows:
- The verdict — passed, failed, or error (see Assertions).
- Assertion results — one row per assertion, with what was expected versus what actually happened.
- The conversation trace — the full back-and-forth, every tool call, and the procedure steps the agent took.
- AI credits used — what the run cost.
History and comparison
Suite runs are kept over time so you can compare results across changes and spot when something regressed. A pulse indicator on the Simulations tab shows when a suite is currently running.
A green suite before publishing
Make it a habit to get your simulation suite passing before you publish a version. It's the cheapest way to keep a reliable agent reliable.
Next steps
Assertions
Define the checks that decide whether a simulation passes — from how the conversation ends to which tools the agent must and must not call.
Suggested simulations
Let Auxx propose ready-to-run test cases for a procedure — happy paths, branches, and edge cases — so you can build coverage with one click.