Source: https://bridgeeme.com/docs/antifraude-score-mmm.html

Documentation / Antifraud, score & MMM

# Antifraud, score & MMM

Animated Scanner

## Evidence before decision

✓ 

Anti-fraud

## Signs observed

### CTIT

extreme ranges and distribution by channel.

### Replay

click ID duplicated or reused in multiple sessions.

### Conflict

Incompatible native IDs in the same click.

### Automation

burst, cadence and incoherence of interaction.

### Journey

active time, PDP, search and funnel consented.

### Integrity

separate low-quality collection failures.

Score

## Components, not black boxes

```
trust_score = f(
  attribution_confidence,
  traffic_quality,
  measurement_integrity,
  fraud_risk,
  score_coverage,
  model_version
)
```

The components and reason codes are the operational decision. The aggregate number is a versioned summary.

Responsible use

## Operating rules

| Situation | Action | 
| --- | --- |

| high score + high coverage | maintain monitoring; do not assume causality | 
| low score + high fraud | investigate evidence and partner before blocking | 
| low score + low integrity | correct collection; do not penalize channel | 
| behavioral unavailable | use only technical risk | 
| version change | compare shadow and recalibrate baseline | 

MMM

## From quality to model

The MMM works on a channel × day/week × geography basis. Use trusted installs, fraud rate, score coverage, physical signs, investment and external controls. Perform adstock, saturation, backtesting and uncertainty intervals.

Do not multiply income by trust score. Use quality as a feature, observation weight or sensitivity analysis and document the specification.