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

# Thresholds and decisions

> fraud_score, risk_level, decision, and threshold_used in scoring responses

## fraud\_score

Estimated fraud probability in **\[0.0, 1.0]**. Higher values indicate greater risk. The score is calibrated with chargebacks and confirmed labels.

## Default global thresholds

| Band | Score range | risk\_level | Typical decision |
| - | - | - | - |
| Low | ≤ 0.25 | `LOW` | `APPROVE` |
| Medium | ≤ 0.55 | `MEDIUM` | `REVIEW` |
| High | ≤ 0.80 | `HIGH` | `REVIEW` |
| Critical | > 0.80 | `CRITICAL` | `REJECT` |

Your client may have custom thresholds. When they apply, the response includes `threshold_used`:

```json theme={null}
{
  "threshold_used": {
    "low_max": 0.25,
    "medium_max": 0.55,
    "high_max": 0.80
  }
}
```

| Field | Description |
| - | - |
| `low_max` | Maximum inclusive score for LOW band |
| `medium_max` | Maximum inclusive score for MEDIUM band |
| `high_max` | Maximum inclusive score for HIGH band. Above → CRITICAL |

`threshold_used` may be `null` when the decision comes only from deterministic rules.

## Rules vs model

The rules engine (global + per client) can **override** the ML model decision. In that case `decision_source` is `RULES`.

## uiaf\_reportable

`true` when the transaction meets SARLAFT/UIAF criteria (typically `REJECT` or `CRITICAL`). Your compliance team should review these cases.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.