This turned out to be significantly harder to solve than I was expecting, but my final solution is more elegant than the one I had to abandon.
What I discovered is that textbox.text is not the same during the routine as when it ends. That’s the point at which the newline characters get stripped out. My first attempt was fine for every word apart from the last one, where it lost most or all of the letters.
My tip for today is how to define a split function so you can use an Auto code component in the trial routine, and how to call the split function as the routine is ending, before the newline characters have been stripped out.