3568 passes all tests and runs successfully in most scenarios but I suspect it has introduced a subtle edge case (still not tracked down). I'm going to revert for now once tests pass on this.
It appears to be possible to leave the output MMR in a bad state after chain compaction, not yet reliably reproducible.
Going to revert on master and continue investigating on a branch.
3568 passes all tests and runs successfully in most scenarios but I suspect it has introduced a subtle edge case (still not tracked down). I'm going to revert for now once tests pass on this.
It appears to be possible to leave the output MMR in a bad state after chain compaction, not yet reliably reproducible. Going to revert on master and continue investigating on a branch.