Random reordering of some assessment columns in exported gradebooks

[moveisi]

We recently noticed an unexpected PL behaviour that causes exported gradebook to reorder assessment columns apparently randomly, when assessments share the same set and number. See the screenshot below.

"set": "Homework",
"number": "3",

In our case, we had two assessments both with same set and number above (the older one was incorrect and omitted). They show up under Gradebook tab as expected: always the older assessment column comes before the newer one.
When you export the gradebook, they show up as HW3 and HW3.1. However, the problem is that when you export multiple times, which assessment will be called HW3.1 seems to be randomly selected, as shown in the screenshot.
The links to these assessments HW3 (old) and HW3 (new) are available in the related discussion in Slack pl-help channel.

[nwalters512]

A relevant piece of information: the .1 suffix actually comes from Pandas, not PrairieLearn. AFAICT, Pandas does that deterministically based on the ordering of the original columns, so it's still on us to produce output in a consistent order. But the raw CSV just has two columns named HW3.