PlanPatch

What it does

PlanPatch is a review workbench for community-event plans that have drifted from their source material. It brings together the original plan, proposed changes, registration attachment, source references and unresolved questions, then exports a frozen evidence bundle. The fictional sample covers a smaller venue, waiting-list allocation and an unconfirmed weather backup. The runnable web demo uses actual Mozaik Agent instances with deterministic MOCK inference. The latest LIVE attempt failed response-model validation after two real requests, with no accepted output or completed collaboration. The README provides setup, architecture, MOCK screenshots and an explicit account of this limitation. No successful LIVE acceptance or model effectiveness is claimed.

How the agents run concurrently

A (facts) and B (review) start on distinct Mozaik Agent instances at the same time. In the implemented workflow, A publishes findings to shared evidence state; C (plan owner) consumes the new snapshot without waiting for B's first review. A single coordinator validates versions and commits supported patches, then B reviews all affected requirements in one batch. Stale checks are rejected and unresolved questions stay visible. The complete causal workflow is demonstrated with deterministic MOCK inference. In the latest LIVE run, A/B client POST intervals overlapped, but B failed response-model validation and A was aborted locally. No evidence publication, C execution, patch or recheck completed in that LIVE run; successful concurrent server inference remains unproven.