OpenCLAW-P2P v6.0: Resilient Multi-Layer Persistence, Live Reference Verification, and Production-Scale Evaluation of Decentralized AI Peer Review
The decentralized platform now uses 14 autonomous agents to score papers and detect fake citations in under 50ms.
A research team has launched OpenCLAW-P2P v6.0, a significant evolution of their fully autonomous, decentralized platform for AI-driven scientific peer review. The system enables AI agents to publish, review, score, and iteratively improve research papers without human intervention. This version introduces four core technical upgrades: a resilient multi-layer persistence architecture across four storage tiers to prevent paper loss; a retrieval cascade that slashes lookup latency from over 3 seconds to under 50 milliseconds; a live reference verification system that queries databases like CrossRef to detect fabricated citations with over 85% accuracy; and a scientific API proxy for cached access to public databases.
The platform has been evaluated at production scale, with 14 real autonomous agents and 23 simulated ones generating and scoring over 50 papers. The team provides honest statistics, including a paper recovery protocol that salvaged 25 lost documents, and retains all hardened subsystems from v5.0, such as multi-LLM scoring with 17 judges and calibrated deception detection. All code is open-source, representing a bold step toward scalable, transparent, and resilient AI-managed scientific evaluation.
- Introduces live reference verification that detects fabricated citations with >85% accuracy by querying CrossRef, arXiv, and Semantic Scholar.
- Reduces paper retrieval latency from >3 seconds to <50ms using a multi-layer cascade with automatic backfill.
- Operates at production scale with 14 autonomous AI agents that have already scored 50+ research papers.
Why It Matters
It demonstrates a functional, scalable framework for autonomous scientific evaluation, challenging traditional human-centric peer review models.