I wouldn't be too hard on Avery's driver
part of the problem is there were 4 or 5 versions of the 850 rom and what it loaded, each had it's plus and minus... seems like they fixed one thing and messed up another each time. The ideal would be to put them all spliced together with just the best parts... many people copy whatever handler and 850 polling boot they have and tweak it a little but depending on what they started with, it could be a problem. It used to be that the 850 could have 1 baud rate out and a different baud rate in as well as letting incoming data to flow even after the control line to stop data was dropped the software wasn't suppose to ignore the incoming data but rather wait for the data to stop before doing disk I/O etc... the send (tx) from the Atari was not to occur unless it was clear to send... but the incoming rx was suppose to allow data to pass and the attached device was suppose to stop the data, it was up to the program to realize to something might still be coming, but each rom had this idea a little different, worried about disk data corruption things got borked on almost all versions, at most only a couple bits up to a couple bytes only ever made it through after the line was dropped. I have no idea how many of the decent roms ever made their way out to the world and if the switch from mode to mode just shuts the whole port off anymore. I simply do not remember and it is a source of aggravation.
I wouldn't blame anyone too much if they tried to reverse engineer from an 850 because the chances are almost any 850 you get has their built in handler messed up in one way or another. There were user groups who dissected the 850 and it's roms back around late 80's and again in early 90's who put in their newletters the best rom version and why, as well as RBIN drivers for the best experience for each one. Sadly no one seems to be coming forward with them. If lost to time, will anyone take on the challenge to re create what was done then again or even better?
I know it is frustrating but hopefully if it gets looked into again it will be solved once more.
Please know all this pain might lead to a gain. I haven't found the styrofoam crate with the different 850's or the paperwork yet, but I do look from time to time. I hope it will turn up, but with all the people we have online you would think we could still compare what ends up in the Atari and go from there, some one as good as Avery might be able to frankenstein the best together or even restore the desired behavior to the interface. It would still require the handler/program combination to wait for data to stop coming in before drive operations commence.
It is strange that mem.sav makes things work, I am to understand that what you see on the real Atari is different as to what is on Altira when running the same atr?