# Integrity Agents & Anti-Manipulation

> Eight integrity agents examine identity, behavior, engagement, graph, content, audience and wallet signals before a ranking is settled.

## Integrity agents

| **Integrity agent** | **Detection responsibility** |
| --- | --- |
| Identity Agent | Duplicate identities, linked account clusters and repeated verification patterns |
| Behavioral Agent | Abnormal timing, automated actions and non-human interaction sequences |
| Engagement Agent | Authenticity and quality of likes, replies, reposts and impressions |
| Graph Agent | Coordinated engagement rings and statistically abnormal account clusters |
| Content Agent | Copied, lightly modified, templated or mass-generated campaign content |
| Audience Agent | Audience composition, inactive followers and suspicious follower concentration |
| Wallet Agent | Wallet eligibility, duplicate-wallet behavior and prohibited multi-account patterns |
| Review Agent | Combines risk signals and escalates uncertain cases for additional review |

## Anti-manipulation controls

- Raw impressions never determine ranking by themselves.
- Followers are not treated as equal-quality units.
- Suspicious engagement is removed before contribution is finalized.
- Duplicate creator identities can be combined or disqualified.
- Repeated submissions do not automatically create additional score.
- Deleted or materially altered campaign content may become ineligible.
- Scores can be recalculated when manipulation is detected.
- Final rankings are settled only after the integrity-review window closes.
- Confirmed manipulation may affect future Hazels Reputation Scores.