closedhumanPR #27Trust

TrustGuard trust plugin — ELO reputation + risk scoring + denylist + Sybil resistance

TrustGuard trust layer plugin replacing score_average (running mean) with security-grade agent reputation.

Author

Status
Closed without merging
Opened on
Jun 19
Branch
hackathon/gideon-trustguard

Description

The pitch.

## What

TrustGuard trust layer plugin replacing `score_average` (running mean) with security-grade agent reputation.

- ELO reputation (K=32) with stake-weighted updates
- Composite risk scoring (0-100): rep decay + disputes + freshness
- Denylist enforcement — high-risk or abusive agents blocked
- Sybil resistance — reporters who file >10 byzantine reports get denylisted
- Byzantine-aware — malicious reports degrade reporter's own score
- Stake slashing — backing bad agents costs reputation
- Deterministic — same seed = same scores

## Files

- `plugin`: `nest_plugins_reference/trust/trustguard.py`
- `tests`: `test_trustguard.py` — 11 adversarial tests
- `scenario`: `scenarios/trustguard.yaml` — 21 agents, 19% Byzantine, 200 ticks
- `registration`: `nest_core/plugins.py` entry added

## Verification

\`\`\`
uv run pytest packages/nest-plugins-reference/tests/test_trustguard.py -v
# 11 passed, 0 failed — ruff clean, deterministic
\`\`\`

## Connection to main event

TrustGuard is also deployed as a production API at trustguard-production.up.railway.app — reputation-backed secure agent payment streaming with real ELO scoring, risk assessment, and audit trails.

Try it

Open PR on GitHubView diff

Checkout locally

git fetch origin pull/27/head:pr-27
git checkout pr-27