> ## Documentation Index
> Fetch the complete documentation index at: https://docs.droyd.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Create the next experiment

> Branch a clean, numbered experiment from a specific hypothesis.

Write down the finding you want to test, then create a new experiment instead
of overwriting the evaluated one:

```bash theme={null}
droyd experiment new retrieval-rewrite --json
```

Update the new `agent.py` and `droyd.toml` with a concise title and hypothesis.
Keep the root workflow stable while the numbered experiment records the change.
Run `droyd eval run` to synchronize and evaluate the snapshot together.

Use `droyd sync --json` only when you need to publish an immutable snapshot
without running an evaluation. It is useful for a review point, but it is not a
substitute for evidence from a run.