It turns out that there is an issue with the implementation of the old witness refresh code and we’re going to disable these chain variables until we can fix the issue and re-enable them.


Our chain implementation keeps two different copies of the ledger, one which is delayed 50 blocks behind main one. When we change something that happens to the ledger outside of a transaction we have to be sure that we’re doing the same things to both versions (at that same delay). In this case we failed to do so, and since the lagging ledger is used to validate old transactions, we’re seeing differences in the validity of transactions on different nodes.