I have seen some of the interpreter source codes, but I don't know if they have been "officially" published. These also include some other things such as test files, and a picture file that I have never seen a decoder for (other than the decoder (and encoder) that I wrote myself).

Many modern implementations do not support permanent shifts in Z versions 3 and above (although all of my own implementations do, and I think all of the official implementations also do, even though Infocom never used that feature (this isn't too surprising since the algorithm they described for deciding when to use permanent shifts is worse than not using them at all; I worked with someone else to make a better algorithm for making this decision)).

Some of the official implementations check the Z version number and some don't; even some that do, do not check if it is a small-endian story file (and the ones that do will only display an error message if it is, and refuse to run it). My own implementations do check for small-endian story files (as well as the Z version number), although some will display an error message and refuse to run it in that case, some actually are able to run both big-endian and small-endian story files (as far as I know, there are no small-endian story files; Infocom never used this feature and no modern compilers support this).

Something else I might mention is that some people say that Infocom used many tricks in the programming, although I have looked at disassembled code in the debugger and found that they could be optimized a lot more (e.g. by using SET->BCOM optimization, and many other things), and the source code for the interpreters also shows some things that could be optimized much better. (Another thing revealed from the source code of the interpreters is a undocumented command-line switch for the DOS version that allows you to specify the name of the story file.)