Side project: a second HiveD implementation in go — replays up to 30M blocks before failing. Curious whether a fresh AI pass with full repo context could outperform the original. Plans to maybe run a top-100 witness on it for the experiment.

This sounds very interesting! Code public by chance?