Essentially, if there are multiple different options for the pronunciation of a word, occasionally the wrong pronunciation will be chosen. From just looking at the behavior, it seems like Yomitan just takes the first reading in the list and goes with it, but I haven't confirmed that. Two examples of this happening today was it choosing いも instead of いもうと for 妹, and it choosing いります over はいります for 入ります. This seems like it isn't so much a bug as it is a design decision that sometimes has undesirable results.

I'm not even sure if fixing this issue is possible (at least in all cases, or without a ton of work), but I figured I'd make an issue to track it. Two solutions that would take comparatively little work that I can think of are:
1 - Using the word that is being added to anki to back-reference the sentence and set the readings for any words that match it or will conjugate to it
2 - When multiple readings for a given word are present in a sentence, pop up a dialog box to choose which is the correct reading. (and maybe add a setting to turn this on or off)

I don't really think either of these are good solutions, but those are the only partially viable solutions I could think of while typing this up. If this really starts to bug me I might work on the problem more, but I doubt I'll find the motivation to.