RMT Mission Control — Trust Scoring for AI Agents

2026-03-29

Composite Score

0.9397
Reputation & Trust Composite
Phase 7b LOCKED
0.937
sybil_detection
0
false_positive_rate
0.8345
worst_correlation
26
params_optimized

Research Loop Status

Status Active (iter 5+)
CXDB Entries 285 experiments · 1,520 insights
Providers Grok (CTO/scorer)
Codex (executor)
qwen3:8b (draft)
Last Experiment rmt-iter-27

Key Structural Findings — Phase 7b

starInDegreeThreshold 6 → 8
Biggest single improvement — eliminated false positives on most topologies
chainLinearityThreshold 0.738 → 0.69
Tighter chain detection, +0.0026 worst-correlation improvement
reciprocalVerifiedDamping 0.80 → 0.72
Stricter verified pair handling — reduces trust inflation between colluding pairs
alpha = 0.614
Genuinely optimal for synthetic eval — phase transition observed at ~0.72

Deploy Plan / Remaining Tasks

  • Parameter optimization pass — look for remaining improvements
  • Real-world validation data (currently [object Object] — needs actual chain data)
  • Smart contract deployment to testnet (Base Sepolia or Arbitrum Sepolia)
  • ReputationEngine.sol — integrate optimized params into contract
  • Cross-model security audit (Grok + Gemini)
  • API endpoint for trust score queries
  • Frontend integration with identity portal
  • Gas optimization study on target L2
  • DeFi partner integrations (deferred — GTM at end)
  • Documentation / developer SDK

Open Decisions

  1. L2 chain selection for deployment (Base vs Arbitrum vs Optimism)
  2. Proof-of-Human model integration — how PoH feeds into trust scoring (needs explanation doc)
  3. On-chain vs off-chain score storage tradeoffs
  4. Score update frequency (real-time vs batched)
  5. Minimum stake/bond requirements for agents

Proof-of-Human Context

PoH = identity verification that feeds into RMT trust scoring. An agent's trust score is partially derived from their human delegator's verification level. Delegation trees work as: Human verifies → delegates to Agent → Agent inherits base trust.

Composite Scoring Methods

MethodTierWeight
Apple App AttestT10.70
Google Play IntegrityT20.40
NFC PassportT10.65
Social GraphT30.15
Decision needed: How much should PoH verification weight influence the final RMT composite score? Currently independent — should they be linked?

Provider Health

Grok
31/31
100%
Gemini
25/25
100%
Codex
22/33
67%
Total
89
dispatches