As an optimization, we could keep a simple index of the "latest" head per log, instead of [traversing *all* heads](https://github.com/mafintosh/hyperlog/blob/master/lib/replicate.js#L214). Note that this must be backwards compatible with the old state (i.e. generate the index if it doesn't exist already).
As an optimization, we could keep a simple index of the "latest" head per log, instead of traversing all heads.
Note that this must be backwards compatible with the old state (i.e. generate the index if it doesn't exist already).