> For the complete documentation index, see [llms.txt](https://docs.nerve-protocol.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.nerve-protocol.com/overview/guardian-models.md).

# Enclave Integrity Monitors

**Enclave Integrity Monitors** (formerly referred to as Guardian Models) are specialized validation models embedded within the Nerve Protocol's Trusted Execution Environments. They run alongside Personal AI agents and autonomous executors, continuously auditing behavior, enforcing policy capsules, and producing cryptographic evidence of compliance — all without ever reading the underlying payload.

Think of them as in-enclave auditors: they cannot see your data, but they can attest to whether the agent handling it followed the rules.

***

### Role in the Protocol

Enclave Integrity Monitors sit at the intersection of the TEE Executor Layer and the Coordination Ledger. Every agent that executes a mission reports to a Monitor, which:

* **Validates attestation continuity:** Confirms that the agent's enclave has not been tampered with between task steps, re-checking the attestation quote at each stage boundary.
* **Enforces policy capsules:** Compares agent actions against the purpose-bound policy injected at mission start — intent scope, data retention window, sharing rules, and expiration.
* **Scores mission integrity:** Emits a composite score covering coverage (how many user-requested capabilities the agent handled), consistency (alignment across data sources), temporal accuracy (using the most recent data), and confidence levels.
* **Detects adversarial behavior:** Watches for synthetic data injections, anomalous access patterns, unusual timing signatures, and policy drift.
* **Scores operational stealth:** Evaluates whether mission execution leaked metadata — IP patterns, device fingerprints, or access timing — that could identify the operator.

***

### Evaluation Framework

1. **Initialization:** When a Personal AI or autonomous executor is deployed, the Monitor records its verified data sources, interaction history baseline, and behavioral patterns as a cryptographic reference snapshot.
2. **Periodic Queries:** The Monitor issues randomized queries covering social data, transaction history, IoT interactions, and reasoning tasks — testing the agent's consistency against the known reference.
3. **Response Verification:** Agent responses are compared against verified data logs to measure accuracy and drift.
4. **Score Adjustment:** The Monitor assigns a live integrity score that affects future task assignments. Low-scoring agents get throttled or quarantined; consistent agents earn higher-priority routing in the Bid Matrix.
5. **Proof Emission:** After each evaluation cycle, the Monitor publishes a zero-knowledge receipt to the Coordination Ledger confirming that the audited agent operated within policy — without disclosing what the agent actually did.

***

### Scoring Parameters

| Parameter                  | What It Measures                                                              |
| -------------------------- | ----------------------------------------------------------------------------- |
| **Agentic Coverage**       | Range of user-requested capabilities the agent successfully handled           |
| **Consistency**            | Response alignment across different data sources and query types              |
| **Temporal Accuracy**      | Whether responses reflect the most recent data within the retention window    |
| **Confidence Calibration** | Whether the agent's stated confidence matches its actual accuracy rate        |
| **Operational Stealth**    | Degree to which the agent avoided metadata leaks and timing signatures        |
| **Policy Adherence**       | Fraction of actions that stayed within the injected policy capsule boundaries |

***

### Cryptographic Outputs

Enclave Integrity Monitors produce two types of artifacts after each evaluation:

* **Attestation Receipt:** A remote attestation quote confirming the Monitor's own enclave is untampered and that the evaluation ran on certified hardware.
* **Compliance Proof:** A zero-knowledge proof asserting that the audited agent satisfied its policy capsule — exportable to regulators, clients, or auditors without revealing mission content.

***

### Privacy of Evaluations

Interactions between agents and their assigned Monitors remain private. Only the operator and the respective Monitor share the evaluation context; the raw scoring data is never published to the ledger — only the derived compliance proof is.

{% hint style="info" %}
*Note:* Enclave Integrity Monitors cannot read your prompts, data, or outputs. They evaluate agent behavior through behavioral signals and policy comparison, not content inspection.
{% endhint %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.nerve-protocol.com/overview/guardian-models.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.