Thx. I reviewed the code again, the described sequence is impossible for all intents and purposes.
Reproduction tests with the 5.0a versions I archived was not successful in producing that behaviour , all worked as designed (pc and mac).
I'm at a loss, but still curious how this can occur.
The only possible way for an erroneous code flow as described to be even theoretically possible is if the "from _call... " commands renpy adds when the distro is compiled are somehow jumbled/ wrongy compiled / inserted
In that case it should be reproducible, however, which it isn't.
Anyway, I went into the code once more, stripped all the 300+ from commands throughout the codebase per script and recompiled. The mac file is available in the OP as v0.5b mac. Can you D/L and check the behaviour?