> ## Documentation Index
> Fetch the complete documentation index at: https://docs.10et.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# tenet train

> Train the policy head on accumulated experience

Train a lightweight RL model on your project's (state, action, reward) tuples. The trained policy biases agents toward actions that historically improved YOUR metrics.

## Usage

```bash theme={null}
tenet train                    # Train on all accumulated tuples
tenet train --dry-run          # Show what would be trained without running
```

## How It Works

1. Reads `.tenet/training-buffer.jsonl` — tuples from agent rounds
2. Extracts features: action type, files affected, scope, prior score
3. Trains a small neural network to predict reward from state+action
4. Saves weights to `.tenet/policy-weights.json`
5. Future agent runs consult the policy head before choosing actions

## When to Train

* After accumulating 50+ tuples (minimum for useful signal)
* After a batch of overnight agent runs
* The overnight build script runs training automatically

## See Also

* [Policy Head](/learning/policy-head)
* [Training Buffer](/learning/training-buffer)
* [RL Loop](/learning/rl-loop)