I think what Tach is saying is that since A and B do not line up in either frame any measurements made in the two frames could not be compared directly.
I'm not sure that is really necessisary though. It seems if you can establish that the flashes occur simultaneously in one frame and sequentially in the other, you have proof of RoS. Einstein's hypothetical goes further than that but it was a teaching tool. You don't really need to know anything about what parts of the train were aligned with the embankment points A and B, or even that the flashes were simutaneous in the embankment frame, or even simutaneous at all. All you need to know is that the event(s) were measured to occurr with a different timing when measured from the two frames.
All Pete needed was the embankment frame, a rest frame for the simultaneous measurement of two events, and an observer in motion relative to the embankment frame. One observer measures the events as simutaneous (whether they are in fact or not) and the other measures them as sequential = RoS.