A bit of a commentary on the bugs:
#1: Early machines did not save X into L when \GS+ and \GS- were done.
This was just a bug, and a pain!
#2: Early machines allowed RCL IND nn and STO IND nn when nn was outside the memory range.
These allowed the user to reach into program memory and change things (it allowed the birth of synthetic programming on the HP41C)
#3: Early machines allowed SF,CF,FS?C, and FC?C IND nn with nn any number (but not a stack register) 0 to 55.
The first bug to be fixed. So to check for all bugs do: 49, STO 01, SF IND 01. On an allbugs calculator the BAT flag is displayed, on a non bug 3 calculator you get NONEXISTANT.
#4: Early models compute the SIN of small angles incorrectly.
Just a pain.
#5: CLP would only clear 1089 program lines.
...and even less with the printer connected (224 steps from memory)
#6: -67A/-97A prorams that contain combinations of number entry, EEX, and CHS do not translate correctly. Example:
Also a problem that pointed the way to synthetic programming.
#7: The second nybble of the seventh byte of Alpha is copies along with the first six bytes during an ASTO.
Horrible if it bites you!
#8: Early machines do not decompile programs if they are turned off during PRGM mode.
Amusing bug - and not fixed until quite late (2035)
#9: Executing CAT 1 while in PRGM mode, then interrupting it with R/S causes step labelling oddities.
Provides the "modern" way to bootstrap byte jumper or other synthetic key assignments. Also allows you to un-private programs :-)
#10: If flag 25 is set and MEAN or SDEV causes an overflow, some flags can be altered.
Odd!
#11 (CW1): If a program at the end of CAT 1 is replaced or deleted, a return to that program will return into the new program or off-program memory.
Fun! Almost not a bug IMHO, rather a caution for people who want to remove programs that are still effectively running.
#12: On early models, the function LN(1+X) gives wrong answers for values of X between -.9990234374 and -1.00.
Ooooh! But not the usual range in which you would use this function.
#13: On early models, the operation (.1)^-43 gives an answer of 9.999999999E42 instead of 1E43.
Ooops :-)