Yeah that is a good idea. We could:

• Include all ERROR (these are typically few)
• Increase the number of rows to 1000
• Strip out owlclassexpression hashcodes to reduce the diff
• stratify the cut-off by shrinking large groups where necessary

I think the last will only impact very few ontologies.

Originally posted by @matentzn in #34 (comment)