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

# Reputation

> How Lucid builds verifiable trust scores for AI assets from real traffic

Lucid's reputation system transforms raw traffic data into verifiable trust scores. Unlike self-reported ratings, every reputation signal in Lucid is anchored to a cryptographic receipt from a real interaction.

## Three-Layer Architecture

### Layer 1: On-Chain (Solana)

The `lucid_reputation` program stores immutable reputation data:

* **Feedback entries** -- Score (1-100), category, receipt hash, asset type
* **Validation entries** -- Third-party verification that a receipt is valid
* **Passport stats** -- Rolling aggregates: feedback count, validation count, average score
* **Revocation** -- Original submitters can revoke feedback; scores adjust atomically

Average scores are stored with 2-decimal precision (value \* 100) and recomputed on every submission or revocation.

### Layer 2: Off-Chain (API)

The `IReputationProvider` interface aggregates data from multiple sources:

* Database-backed provider for fast queries
* On-chain provider for verified data
* Multiple scoring algorithms via `IReputationAlgorithm`

### Layer 3: Gateway (PayReputation)

Credit scoring derived from payment history through TrustGate:

| Tier       | Description                   |
| ---------- | ----------------------------- |
| Unverified | No payment history            |
| Bronze     | Basic payment history         |
| Silver     | Consistent payments           |
| Gold       | High-volume verified payments |
| Platinum   | Extensive verified history    |

Scores are cached for 5 minutes and recomputed from `receipt_events`.

## Reputation Per Asset Type

Reputation applies to all five AI asset types:

| Type ID | Asset Type | What Is Measured                   |
| ------- | ---------- | ---------------------------------- |
| 0       | Model      | Accuracy, latency, consistency     |
| 1       | Compute    | Uptime, throughput, reliability    |
| 2       | Tool       | Correctness, availability          |
| 3       | Agent      | Task completion, user satisfaction |
| 4       | Dataset    | Quality, freshness, coverage       |

## Submitting Feedback

Feedback requires a valid receipt hash, proving the submitter actually interacted with the asset:

```typescript theme={null}
import { RaijinLabsLucidAi } from "raijin-labs-lucid-ai";

const sdk = new RaijinLabsLucidAi({
  serverURL: "https://api.lucid.foundation",
  security: { bearerAuth: process.env.LUCID_API_KEY },
});

await sdk.reputation.lucidComputeReputation({
  passportId: 'agent-abc123',
  assetType: 'agent',
});
```

## Revoking Feedback

Only the original submitter can revoke. The on-chain program atomically adjusts the total and average scores. The `feedback_count` is never decremented (it tracks total entries ever created, used as PDA seed), but `total_score` and `avg_score` reflect only non-revoked entries.

## Cross-Network Reputation

Lucid's reputation system syncs with external identity standards:

| Standard           | Integration                                                                       |
| ------------------ | --------------------------------------------------------------------------------- |
| Metaplex (MIP #52) | Bidirectional sync via `syncReputationPlugin()` (prep code, awaiting MIP release) |
| 8004 / SATI / SAID | Consume scores from external protocols and feed validated scores back             |

This creates a bidirectional reputation mesh: Lucid consumes external signals and feeds its own verified scores back into the broader ecosystem.