I have done loads to fine tune the clearance between the lid and the box, with lots of testing (each test takes hours, during which I have other work to do). It is a bit of a trade off - you need the lid to not scrape on the maze, or be too tight to move, but also not be too loose, and you also need to adjust the "nub" a bit and the "ridge" where it parks when complete. All these are now options on https://www.me.uk/puzzlebox anyway, and tested on my TAZ6. I will have to test on a Makerbot Replicator 2 to confirm if they work on the current defaults, or need more tweaks.
I have also gone for three nubs on the maze by default (again, this is a setting), to be more stable, especially as it is now a tad looser. This works well, but means less maze area and so simpler mazes. Even so, I think the new maze logic does create a good puzzle. I find it hard to tell as I don't have any problem solving these, sorry.
I changed the park ridge and nubs to directly calculated polyhedrons to make them the exact correct shape and size and skew to fit the maze, that was a challenge. It is quite fun getting your head around three dimensional geometry though.
Some of the tinkering has been a bit more complex. One of the changes I made was to allow any number of sides on the outer box. It used to insist on a multiple of the nubs - this ensured you could make the maze any way around (e.g. if 3 nubs, any of the 3 ways to start the maze would work). However, I thought it would be more fun to allow for other combinations, so as per the image above, 7 sides and 3 nubs. This means only one of the 3 ways to start will line up with the base at the end. To help, I added a subtle mark on the lid and the maze for alignment as you may be able to see. That was a slight challenge to code as the starting point of the maze could be wrapped around several times compared to the end point, and the combinations of maze inside, outside, both or neither, made for some fun testing.
Also, I found the lettering was not ideal on the ends. I have an option to do a proper chiselled effect (45 degree angle) in to the letters, but you ideally need a thin/fine font. I have made an ExtraThin variant of my 5x9 font for this now - and overall the 5x9 JTD font looks pretty good in 3D print I find.
I also used parallel command to make the sample boxes without trashing my machine in device waits!
All in all I am running out of things to improve. I have one idea which is to do with fonts: I made a bodge to use Noto Emoji if the string starts with top bit set, which is to allow some hearts and rings, but ideally each block of text should use fontconfig to find if the chosen font has the glyph and have fallbacks and make OpenSCAD for each separate block (as OpenSCAD does not seem to allow a list of fonts with fallback). Doing that could also ensure text does not go off the end of the space. Not sure if I'll go that far yet as the text is usually fixed, and I have already added a scale setting.
Apologies to the many people downloading from thingiverse. I have tried to ensure the various versions I have uploaded are all "good" even if they have slight changes, but just a few times I have messed up and only realised when the test print finished. Thingiverse is very slow at deleting and uploading new models, so I am sure it looked wrong at times. I included a "version" box (as shown above) in the uploads for reference. There are a good selection of test boxes on thingiverse but always best to make a new maze yourself as they will all be different. Still loads of accesses, likes, comments and makes on this puzzle, and even some people making the code direct from github (thought some of the comments were verging on trolling).
I was actually surprised to find at least two people selling these on ETSY! I hope they like the new version.
I did have fun making one the right size for an Annoy-A-Tron though - not sure where to put it :-)
Anyway, I have a bit of printing to get done for this small project :-)