daniel.gottlieb had a failure today, and I took a look.
This is in master, commit c5bfa9391f:
What's happening is WT_CURSOR.search found a match in an in-memory page insert list, and the WT_INSERT had a single update on it that consisted of a valid key and a WT_UPDATE_MODIFY record.
We didn't find a record to back the modify record in the update list, so we fell back to the page itself, but that page had no entries, and we eventually dropped core because the cursor's slot didn't reference anything useful.
I suspect the most likely problem is we should have some lookaside records and don't, which could have quite a few explanations: we didn't store the records, we stored them but didn't read them, we tried to read them but they weren't there.
As far as I can tell, we should not have skipped reading the lookaside records:
it's certainly possible we read the lookaside table but didn't find the records we expected, but I can't think of any way to prove that one way or the other at this point.
I'm putting this into "Needs Triage", but absent any way to proceed, I guess we close it as not-reproducible until we see it again.