ConcurProof

What it does

ConcurProof solves the problem of multi-agent systems claiming to be concurrent when their agents may only be running at the same time without meaningful collaboration. It executes the same controlled task using sequential, naive-parallel, and reactive-concurrent agents, then records their timing, interactions, reactions, and answer quality. ConcurProof uses interaction graphs and ablation tests to determine whether agent-to-agent reactions genuinely improved the result, producing an auditable concurrency certificate.

How the agents run concurrently

The Evidence, Hypothesis, Critic, and Verifier agents run as separate asynchronous Mozaik v4 loops. In reactive-concurrent mode, they overlap in time and exchange events through a shared interaction stream, allowing one agent’s output to trigger another agent while the others are still active. ConcurProof records timestamps and causal links for every reaction, then compares this behavior against sequential and isolated-parallel baselines.