One commenter in that Twitter thread mentions that the Unix epoch embedded in the URL represents the time of the download, not the time of the test.
On HackerNews someone gave a plausible explanation for the timestamps: They are regenerated when you download the PDF with the result.
This explains the inconsistencies in the timestamps – but not in the confirmation codes – because they remain the same.
And the inconsistencies also exists in them. The confirmation codes are ascending, so the result from the 16th should have a lower number than the one from 22nd. However, it is the other way around.