I built a system that relies on humans decrementing a trust rating of peers based on the user manually classifying resources as malicious, so that bad peers can be detected inferentially.

It differs from Eigentrust and Eigentrust++ in that peer nodes don't manually propogate their trust ratings to eachother on their own behalf, nodes just use their own view of their overlay networks when it's time to recompute trust ratings.

Simulation framework with threat models: https://github.com/Psybernetics/Trust-Toolkit

Peer-to-peer caching proxy PoC: https://github.com/Psybernetics/Synchrony