Ah, my mistake. I didn't mention (forgot, in fact) that the collates.txt I used was from a different source. Sorry for sending you on the wrong track! Can't track down its origin anymore (would have liked to give credit), but I'm attaching it here. As you see, it has the "syllable characters" added at the end.
I must have phrased something wrong, because the apostrophe is not an issue - my bad for making it sound like that, English is not my first language.
BUT the key seems to lie in your last paragraph, about the symbol width. I think you found the source of the problem there. Upon closer observation, PB seems to
always try to select three characters as a "word" to look up. No more, no less. With practice, I can tap in the right place (second or third character) to determine where the string begins. If my word consists of only the first character (as it often happens in Japanese), either the dictionary automatically drops the end characters, or I can edit the search and delete them. If it consists of more than three characters, the "edit search" button allows me to find it. Bottom line: the search is not perfect, but it works well enough, if you get used to it, within the limits of the three-character selection, which I guess is determined by PB's software itself.
Taking another look at the three "accessory" files (collates, keyboard, and morphems), it suddenly occurred to me: is keyboard.txt supposed to somehow influence the on-screen keyboard? Because I've never been able to type Japanese characters in the dictionary, all I get is the Latin ones. (Just idle curiosity, it's not that important.)
Despite the minor inconveniences, the dictionary works! Once again, I am very grateful. Have a happy New Year, you and everyone else here (assuming you celebrate it tonight), and I hope life gives you all the good things you deserve and more!
(I tried attaching the .dic itself, but it's seen as an "invalid file" for some reason.)