I've just been trying to work out that if the screen uses 25 rows of 40 chars = 1000 Bytes, what happens to the remaining 24 Bytes in the screen RAM. And with some searching, I found a copy of this book which might be useful to PET enthusiasts on here:
http://www.primrosebank.net/computer...Raeto_text.pdf
And on page 4, it confirms that the last 24 Bytes are not used. However, the daver2 PETTESTE2KV04 program writes, then reads all 1024 Bytes.
So you could have the situation that there is some faulty bytes in only these last 24 Bytes, that do not appear to actually be required to work OK, but the test would fail and you wouldn't be able to see why.
And it might have been better if it hadn't checked these / showed a screen RAM read-back failure in a different way (maybe after a delay, clear screen and display a faulty locations list (but would need most to work OK, to see this)