Haha, that's hilarious!
I had to re-carve a track loop because it had originally been engraved exactly the wrong way around. Although i guess with the size of the current partial project, that single mis-carve hardly counts, esp. considering the fully-additive-and-subtractive 8 bits of memory (32 hatches, ~140 links) ran without a hitch right out of the box and that part is always on, shutting it down for repairs would be nigh impossible. I had to debug both operation loops, both the subtraction one - aforementioned wrong track loop - and the multiplication one - the moment i started the test drive, i knew it was going to go wrong: two straight accelerators in a row are always too much for logic purposes.
A high-friction track stop kept it in check and the second test succeeded with flying colours.
I've upgraded the Marble Bookkeeper, now it can also subtract and multiply. I've done both the lazy way and with very small capacity - the subtractor can handle eight bits, the multiplier a whopping four (the product can be eight bits large, though). Reading the tale of the Mighty Dwarven Calculator was very inspiring and moved my thoughts in the right directions. I suspect multiplying with bigger numbers would become infeasible even with less space-consuming and latency-plagued logic systems. Unless i'm missing some great opportunity, multiplication of two n-bit numbers requires up to n^2 read/write operations, where for each active bit of the 'first' factor, each bit of the second must be evaluated and written to the operator array in the appropriate location, then added up. Multiplying two 10-bit numbers could take half a month game-time this way.
I added an extra switchable 'set' loop to my counter/memory cell, so now it can handle both addition and subtraction, being able to execute the commands 'set to one', 'set to zero', 'set to zero and send positive carry' and 'set to one and send negative carry'. It's a rather unwieldy 8x9 tiles this way, though.
The calculations handed to the machine were 233-94 and 11x7, both of which generated the correct output. I'll need to add some features to ease operation and emulate negative numbers, for an expected 100 additional mechanisms or so. The current tally for all machinery on site is 440 installed hatch covers and 1540 mechanisms.