3320c24afa
Add CRC32C checksums to WAL record format (v2), implement crash recovery with automatic truncation of corrupt records, add feature-gated group commit buffer for batched fsync under concurrent load, and implement log rotation via segment files with global offset addressing. Key changes: - Record format v2: [len:u32][crc32c:u32][blake3:32][payload:N] - recover_file() scans and truncates corrupt tail records - GroupCommitBuffer batches fsync via MPSC channel (tokio feature gate) - SegmentManager with binary search resolution and cursor-based cleanup - Journal::read() auto-refreshes segments on miss for writer/reader split - Split recovery.rs and key_codec.rs into directory modules for 500-line max Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
97 lines
2.8 KiB
Markdown
97 lines
2.8 KiB
Markdown
# TrustRank (Reputation)
|
|
|
|
**Last Updated:** 2026-02-01
|
|
**Confidence:** High
|
|
**Status:** Implemented in `stemedb-storage` v0.1.0
|
|
|
|
## Summary
|
|
|
|
TrustRank is a reputation system for agent credibility. Agents who make accurate claims gain reputation; agents who contradict settled truth lose reputation.
|
|
|
|
**Key Facts:**
|
|
- Stored in `TR:{AgentId}` key in KV store via `TrustRankStore` trait
|
|
- Scores bounded to [0.0, 1.0], default 0.5 for new agents
|
|
- Used by `TrustAwareAuthorityLens` to weight assertions
|
|
- Confidence decays over time (30-day half-life by default)
|
|
|
|
**File Pointer:** `crates/stemedb-storage/src/trust_rank_store.rs`
|
|
|
|
## TrustRank Record
|
|
|
|
```rust
|
|
pub struct TrustRank {
|
|
pub agent_id: [u8; 32],
|
|
pub score: f32, // 0.0 to 1.0
|
|
pub last_updated: u64, // Unix timestamp
|
|
pub assertions_count: u64,
|
|
pub accuracy_count: u64,
|
|
}
|
|
```
|
|
|
|
## How It Works
|
|
|
|
### Score Updates
|
|
|
|
```rust
|
|
// Record an outcome for an agent
|
|
trust_store.record_outcome(&agent_id, was_accurate, timestamp).await?;
|
|
|
|
// Accurate prediction: +0.05 (capped at 1.0)
|
|
// Inaccurate prediction: -0.1 (floored at 0.0)
|
|
// Higher penalty for inaccuracy discourages spam
|
|
```
|
|
|
|
### Confidence Decay
|
|
|
|
```rust
|
|
// Apply time-based decay to all scores
|
|
trust_store.decay_trust_ranks(current_timestamp, Some(custom_half_life)).await?;
|
|
|
|
// Default half-life: 30 days
|
|
// Score approaches 0.5 (neutral) over time without activity
|
|
```
|
|
|
|
### Usage in TrustAwareAuthorityLens
|
|
|
|
```rust
|
|
use stemedb_lens::TrustAwareAuthorityLens;
|
|
use stemedb_storage::{HybridStore, GenericTrustRankStore};
|
|
use std::sync::Arc;
|
|
|
|
let store = HybridStore::open("./data")?;
|
|
let trust_store = Arc::new(GenericTrustRankStore::new(store));
|
|
let lens = TrustAwareAuthorityLens::new(trust_store);
|
|
|
|
let resolution = lens.resolve_async(&candidates).await;
|
|
// Winner = assertion with highest (confidence * trust_rank)
|
|
```
|
|
|
|
**Resolution Strategy:**
|
|
1. For each candidate, lookup signer's TrustRank (O(1))
|
|
2. Calculate: `weighted_score = assertion.confidence * agent.trust_rank`
|
|
3. Return assertion with highest weighted score
|
|
4. Tiebreaker: most recent timestamp
|
|
5. Unsigned assertions treated as 0.0 trust
|
|
|
|
## Score Interpretation
|
|
|
|
| Score | Meaning |
|
|
|-------|---------|
|
|
| 0.0-0.3 | Unreliable (history of inaccuracy) |
|
|
| 0.3-0.5 | Below neutral (more wrong than right) |
|
|
| 0.5 | Neutral (new agent or unknown) |
|
|
| 0.5-0.7 | Above neutral (more right than wrong) |
|
|
| 0.7-1.0 | Reliable (strong track record) |
|
|
|
|
## API Access
|
|
|
|
**Via LensDto:**
|
|
- `lens=Authority` → Routes to `TrustAwareAuthorityLens`
|
|
- `lens=TrustAwareAuthority` → Routes to `TrustAwareAuthorityLens`
|
|
- `lens=Confidence` → Uses confidence field only (no TrustRank)
|
|
|
|
## Related Topics
|
|
|
|
- [Lens](../services/lens.md) - TrustAwareAuthorityLens integration
|
|
- [Assertion](../services/assertion.md) - Signature and agent_id fields
|