# Integrity & Anti-Manipulation

> A multi-agent integrity system inspects separate manipulation surfaces before their findings are combined into one campaign integrity decision.

Pulse uses a proprietary multi-agent integrity system operated by Hazels.

Different agents inspect separate manipulation surfaces before their findings are combined into a campaign integrity decision.

## Integrity agents

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

## Bot and manipulation detection

Pulse does not classify an account as authentic based on a single metric.

The system evaluates combined signals including:

- Unnaturally synchronized engagement
- Repeated interaction clusters
- High-volume actions within non-human time intervals
- Sudden audience creation or follower-acquisition patterns
- Engagement from inactive or low-history accounts
- Repeated content structures across unrelated creators
- Semantic duplication and coordinated phrasing
- Wallet reuse across multiple creator identities
- Improbable view-to-engagement relationships
- Circular engagement between fixed account groups
- Geographic and temporal inconsistencies
- Campaign actions disconnected from normal account behavior

Each detected signal contributes to an internal risk vector.

```text title="integrity-factor"
Integrity Factor = 1 - Validated Manipulation Risk
```

The Integrity Factor is constrained between 0 and 1.

| **Integrity Factor** | **Result** |
| --- | --- |
| 1.00 | No validated score reduction |
| 0.75–0.99 | Limited suspicious activity; proportional reduction |
| 0.40–0.74 | Material integrity risk; significant reduction and review |
| 0.01–0.39 | Severe manipulation indicators; near-total score reduction |
| 0.00 | Invalidated contribution and removal from settlement eligibility |

> **INFO:** Exact detection thresholds are not published because exposing them would make the integrity system easier to exploit.

## 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.
- Hazels may request additional wallet or account verification for high-risk cases.
- Confirmed manipulation may affect future Hazels Reputation Scores.

## Design principle

Pulse measures the relationship between who a creator is, what the creator contributes and whether the resulting attention is authentic.

Reputation can improve the context of a contribution, but it cannot replace contribution.

Reach can increase distribution, but manipulated reach cannot create valid impact.

The objective is to make creator attention measurable without allowing follower count, bot activity, FCFS access or undisclosed selection to control the outcome.