Ooh I wonder if this is related to T87810

I'm not sure, @atgo! It's strange that the numbers from 2/2 - 2/6 look
normal, and then jump on 2/7.

@Ejegg is this related to the email you sent about HU earlier?

This was my fault. On 2/5, I deployed a change to the payments cluster but not to the machine that runs our orphan script to check on transactions which didn't clear instantly. Because we changed the format of messages sent between machines, the orphan script repeatedly marked some valid transactions as rejected and delayed recording them in CiviCRM. On 2/17 I updated the orphan runner to the same version. It cleared up 100 donations that had been stuck in limbo, but I'm still determining whether a number of older orphan donations are still wrongly marked as rejected.

For now, you can filter out the bad rows from the orphan script by adding a clause involving the 'server' column (machine name sent via email).

Update: Katie points out that many of the spurious 'reject' lines are actually abandoned transactions.

Between the updated orphan script and the nightly audit processor, all valid transactions should have made it into CiviCRM.

Thanks, @Ejegg! I got the server info and will work on getting the data for
Pats. I think this is solved now...