Every few years I seem to forget that slightly different base64 strings can decode to the same bytes, even after excluding whitespace and the = padding.
For example, 0xd5 is the decoded result for 1a=, 1b=, 1c=, 1d=, 1e=, and 1f= -- it makes total sense given the encoding algorithm, but sometimes throws a curveball into testing, especially if you assume different inputs are always going to lead to different outputs.
I chased a "broken" test for an hour tonight before it clicked again. Happy Friday!