There is also ForgeTrace: A Failure-Aware, Provenance-Native Trajectory Dataset for Software Agents we are working on: https://mcp.undisk.app/forgetrace

RFC for a stricter way to evaluate software agents: not just whether they finish with a correct patch, but whether they can operate safely over persistent state, recover from failure, obey policy, coordinate with other agents, and explain what changed with verifiable provenance.

Happy to answer questions on the edge architecture or MCP transport layers! Diagrams on https://mcp.undisk.app shows How It Works including:

Every write → immutable version → SHA-256 hash-chained audit entry Any version → restore in <50 ms (metadata flip, no blob copy)