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

# Policy engine

> Five core predicates that authorize execution outside the LLM.

The policy engine is a **deterministic authorization layer**. The LLM (or `local_rules` fallback) proposes plans; predicates decide whether execution may proceed.

## Core predicates

| Predicate         | Question it answers                             |
| ----------------- | ----------------------------------------------- |
| `drawdown_ok`     | Is 24h PnL above the max drawdown floor?        |
| `whitelist_ok`    | Are all assets in `ALLOWED_TOKENS`?             |
| `trade_size_ok`   | Is each leg below `MAX_TRADE_SIZE_USD`?         |
| `slippage_ok`     | Is expected slippage within `MAX_SLIPPAGE_BPS`? |
| `daily_volume_ok` | Will this trade exceed the daily notional cap?  |

All five must pass:

```
plan_approved ⇔ drawdown_ok ∧ whitelist_ok ∧ trade_size_ok ∧ slippage_ok ∧ daily_volume_ok
```

## Where code lives

| Layer                                    | Path                                                    |
| ---------------------------------------- | ------------------------------------------------------- |
| Predicate definitions                    | `apps/worker/ameo_worker/policy.py`                     |
| Extended checks (gas, balance, protocol) | `apps/worker/ameo_worker/services/guardrail_service.py` |
| Graph integration                        | `apps/worker/ameo_worker/graph.py` → `guardrail` node   |

Failed checks emit `guardrail_evaluated` with a `violations[]` array. The replay UI surfaces each rule as pass/fail in the Policy validation node.

## Formal spec

See the full predicate table and failure semantics in [Policy specification](/policy-spec).

To extend predicates, follow [Write a policy](/guides/write-a-policy).