Follows on from (or maybe is the final part of) #537
We'd like to make it as easy as possible to see, on real-life examples from #536
- whether cellfinder performance has changed after code changes
- whether cellfinder results regress correctly (i.e. find the same cells) after code changes
