Fuzzy Match Threshold

# Fuzzy Match Threshold

Auto-accept high-confidence matches, confirm uncertain ones.

## Category
Tool Resilience

## Context
Tools that resolve natural identifiers.

## Problem
Always requiring confirmation is slow; never confirming risks mistakes.

## Solution
Use confidence thresholds:
- **High confidence (>90%)**: Auto-accept
- **Medium (50-90%)**: Return options
- **Low (<50%)**: Ask for different input
- **Configurable**: Let users set threshold

## Example (Python)
```python
@tool
def find_user(
    name: str,
    auto_accept_threshold: float = 0.9
) -> User | ConfirmationRequest:
    """Find a user by name.
    
    Auto-accepts matches above 90% confidence.
    Returns options for lower confidence matches.
    """
    matches = fuzzy_search_users(name)
    
    if not matches:
        return ConfirmationRequest(
            message="No users found",
            suggestions=["Try a different spelling", "Use email instead"]
        )
    
    best_match = matches[0]
    if best_match.confidence >= auto_accept_threshold:
        return best_match.user  # Auto-accept
    
    return ConfirmationRequest(
        message=f"Multiple possible matches for '{name}'",
        options=[{
            "user_id": m.user.id,
            "name": m.user.name,
            "confidence": f"{m.confidence:.0%}"
        } for m in matches[:5]]
    )
```

## Considerations
- Document the threshold behavior
- Allow threshold customization
- Log auto-accepted matches for audit
- When threshold is not met, return a clear error message with potential matches for confirmation

## Related Patterns
- confirmation-request
- natural-identifier
- parameter-coercion

---
Source: https://arcade.dev/patterns/fuzzy-match-threshold