> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zen.land/llms.txt
> Use this file to discover all available pages before exploring further.

# Dispute Resolution Security

> How the agent system protects against bad actors.

## Multi-Layer Protection

Dispute resolution in Zenland has multiple security layers:

```mermaid theme={null}
graph TD
    A[Disagreement Occurs] --> B{Try Mutual Split?}
    B -->|Yes, Agreed| C[Split Executed]
    B -->|No Agreement| D[Invite Agent]
    D --> E{Agent Valid?}
    E -->|Yes| F[Agent Reviews]
    E -->|No| G[Return to Disputed]
    F --> H[Agent Decides]
    H --> I[Funds Distributed]
    D --> J{Agent Timeout?}
    J -->|7 days| K[Claim Timeout]
    K --> G
```

***

## Agent Staking

Agents must **stake funds** to participate:

### Dual Staking Model

| Stake Type           | Purpose                                   | Amount                 |
| -------------------- | ----------------------------------------- | ---------------------- |
| **Stablecoin Stake** | Determines MAV (max value they can judge) | Min \~\$100 equivalent |
| **DAO Token Stake**  | Shows protocol alignment                  | Min \~100 tokens       |

### How MAV Works

```
Agent stakes: $1,000 USDC
MAV multiplier: 20x
Max escrow they can handle: $20,000

Your escrow: $15,000
→ Agent is eligible ✓

Your escrow: $25,000
→ Agent cannot be selected ✗
```

***

## Protection Against Agent Attacks

### Attack: Agent Unstakes After Selection

**Scenario:** Agent is selected, then unstakes before dispute, then colludes for a bribe.

**Protection:**

1. Agent eligibility is re-checked at *invite time* (not just creation)
2. `activeCases` counter prevents unstaking during disputes
3. 30-day cooldown after last case before unstaking allowed

### Attack: Agent Colludes with One Party

**Scenario:** Agent takes a bribe to rule unfairly.

**Protection:**

1. Economic: Bribe must exceed stake value (at risk of slashing)
2. Reputation: Bad decisions visible on-chain, affect future selection
3. DAO: Token holders can vote to slash bad agents

### Attack: Agent Goes Inactive

**Scenario:** Agent stops responding to requests.

**Protection:**

1. 7-day timeout after invite
2. Either party can claim timeout
3. Escrow returns to disputed with no agent
4. Parties settle via mutual split

***

## Two-Stage Agent Validation

Agents are validated at two checkpoints:

<Steps>
  <Step title="At Creation (Factory)">
    * Is this agent registered?
    * Is their MAV sufficient?
    * Are they active and available?

    *If fails: Cannot create escrow with this agent*
  </Step>

  <Step title="At Invite (Escrow)">
    * Is agent still registered?
    * Is MAV still sufficient?
    * Are they still active and available?

    *If fails: inviteAgent() reverts*
  </Step>
</Steps>

<Note>
  This two-stage validation prevents "nothing to lose" attacks where agents unstake between creation and dispute.
</Note>

***

## Seller Refund Escape Hatch

Even in disputes, the seller can always refund:

<Card title="sellerRefund()" icon="undo">
  Seller can return 100% to buyer at any time — no approval needed.
</Card>

This provides an ultimate escape:

* Agent is unresponsive? Seller can refund.
* Dispute is taking too long? Seller can refund.
* Seller made a mistake? Seller can refund.

***

## Locked Escrow Considerations

For locked escrows (no agent):

<Warning>
  There is NO third-party intervention. Security comes purely from game theory.
</Warning>

| Protection          | Standard Escrow | Locked Escrow |
| ------------------- | --------------- | ------------- |
| Agent arbitration   | ✓               | ✗             |
| Seller refund       | ✓               | ✓             |
| Mutual split        | ✓               | ✓             |
| Timeout protection  | ✓               | ✗             |
| Permanent lock risk | ✗               | ✓             |

***

## DAO Oversight

The DAO provides backstop governance:

* **Slash bad agents** for proven misconduct
* **Adjust parameters** (fees, timeouts, minimums)
* **Blacklist tokens** that behave maliciously
* **Upgrade factory** for future escrows

<Info>
  The DAO cannot intervene in individual disputes or override agent decisions. This is by design for decentralization.
</Info>

***

<Card title="Learn About Agents" icon="gavel" href="/agents/what-is-an-agent">
  Understand the agent system →
</Card>