Detection of invalid gossip creators requires Parsec to be aware of the consensus state of other nodes based on the graph. This means we cannot just reset all the consensus metadata after reaching a decision, because other nodes might not have reached one yet, and we wouldn't be able to tell when they do after metadata erasure. On the other hand, not erasing the metadata will block us from starting the calculation of the next stable block.
The task is to find a feasible way of detecting the "invalid gossip creator" malice type.
It used to be a part of and MAID-3033, but we extracted it because we are prioritising dynamic membership handling.