All of these violate this proposed rule.

After reading some of these comments, there are more exceptions than that, and I wrote it confusingly. So how about this: you cannot have more than one consonant at the same side of a syllable without extenuating circumstances. Having two of them have the same place of articulation (like the s and t in backstop), is a common one. s and z seem to only be possible to place after consonants that are voiced and unvoiced respectively. Neither can be placed after a ʒ (the second half of a j sound).

That should be "consistently pronounced." However the native speakers consistently pronounce something is right.

The pronounce it in a way that violates the theory of what can and cannot be pronounced in Japanese. As far as I can understand, the Japanese alphabet has one character for each syllable. Each syllable has one consonant, then one or more vowels, then possibly an n. There is no syllable "sem".

I don't know if they always pronounce it "sempai". I know it is at least sometimes written "senpai". I just meant that it's very common to pronounce that way, even though it shouldn't be possible at all. If they have gratuitous English that has a syllable ending in a consonant, they stick a vowel after it. For example, "red" becomes "redo" (and the d is particularly t-like, and the r is something that has no English equivalent).