This document outlines the planned development milestones for Trace Protocol, an AI agent verifier on Ethereum that classifies agents as AUTONOMOUS, HYBRID, or HUMAN.
- Monorepo scaffolding with Turborepo
- Ethereum WebSocket listener for real-time transaction monitoring
- Signal extraction pipeline for transaction metadata
- Core type definitions (AgentSignal, ScanResult, Verdict)
- CI pipeline with lint, type-check, and unit tests
- Initial documentation and contribution guidelines
- ONNX model integration for agent classification
- Feature extraction pipeline (32 behavioral features)
- Verdict engine with weighted scoring
- SDK release — github source ·
trace-protocol-sdk - SDK published to npm registry — gated on
trace-protocolorg setup - TypeScript and Python query examples
- Turborepo pipeline optimization for parallel builds
- Foundry contract for on-chain attestation storage
-
create_attestationandqueryinstructions - Devnet deployment of trace-attestation program
- Fastify migration for API layer
- Python FastAPI inference server
- Feature extraction expanded to 47 behavioral signals
- WebSocket feed for live verdict streaming
- Batch query endpoint for high-volume consumers
- Platt scaling confidence calibration
- SDK live subscription method
- Backfill script for historical agent classification
- Rate limiting and API key authentication
- Mainnet deployment of trace-attestation program
- Mainnet scanner configuration and RPC failover
- Model v2 training with expanded labeled dataset (10k+ agents)
- Latency optimization target: sub-200ms end-to-end classification
- Load testing at 1000 concurrent queries
- Security audit of Foundry contract by third-party firm
- API key management and usage dashboard
- Webhook notifications for verdict changes
- REST callbacks for partner integrations
- Agent reputation score (rolling 30-day behavioral index)
- Historical verdict timeline per agent address
- Grafana dashboard for operational monitoring
- Public status page
- Ensemble model with gradient-boosted trees + neural net
- Temporal feature extraction (behavioral drift detection)
- Multi-chain support: Ethereum L2 agents (Base, Arbitrum)
- Agent clustering and similarity search
- Explainability layer: per-feature contribution to verdict
- Model retraining pipeline with automated dataset refresh
- CLI tool for local agent verification
- Embeddable verification badge (SVG/React component)
- Verification widget for dApp frontends
- Improved SDK with retry logic and connection pooling
- OpenAPI spec generation from route definitions
- Interactive API playground
- On-chain governance for classification thresholds
- Stake-weighted validator participation in verdict consensus
- Dispute resolution mechanism for contested verdicts
- Transparent model versioning on-chain
- Community-contributed feature proposals
- Bug bounty program launch
- Stable API with backwards-compatibility guarantees
- SLA commitments for uptime and latency
- Comprehensive audit trail for all verdicts
- Multi-region deployment for global low-latency access
- Published research paper on behavioral classification methodology
- SDK v1.0 with long-term support
- Full documentation site with tutorials and guides