This is an insidious issue.

We take string labels for our endpoints (as JSON). These strings get encoded as b64. Then, when the response is parsed, they are rendered back to UTF-8 for display as JSON.

This means that 1 out of every 8 labels is invalid. For example, '516d593559505355376d454b61374751577146467a47754658396d516a685674716b7663744b376b4b666a35336d' is not valid, because its b64 representation can't be serialized as UTF-8.

Here's a discussion of this matter: https://ptb.discordapp.com/channels/411401661714792449/411401661714792451/564353305887637517

@KPrasch suggests in #508 that we adopt UTF-8; I can now fully concur.