If there's something strange
In your neighborhood
Who you gonna call?
Ghostbusters!

If there's something weird
And it don't look good
Who you gonna call?
Ghostbusters!

--------------------

I had a decent start on TI (was able to get 7/9 atoms in 29 instructions) but the last bit just kept not working so I abandoned TI in favor of RI. However as a final reprisal of week 1, I was narrowly unable to get my rate submission done in time.

I started my rate work with an extensive theoretical analysis. The main theoretical limitation on instructions is due to the purifications; each purification requires a minimum of 2 grabs, 2 moves, and 2 drops. Together with projection, the absolute minimum is 1999 operations per 272 cycles, which is about 8 instructions per cycle assuming only single arms. Since period 4 is about the minimum reasonable period, this makes about 32 instructions as a minimum, or less with proper hexarm usage.

This laid the groundwork for my plan. I believe the best approach is to use wanding to disassemble the input and distribute the metal atoms, then use hexarms on track to do 4 purifications in parallel with each arm. After making the needed atoms, an arm on track can manipulate a big brick once every 272 cycles to make 15 outputs.

An alternative plan involved some cleverness with routing the gold atoms to conditionally grab mercury, which in turn conditionally grab silver, so that the machine doesn't need to calculate the correct ratios and the atoms are produced at the correct ratio on demand.

Of course once I tried implementing my plan, none of these ideas worked at all. Wanding seemed hopeless, and I couldn't get any parallelism on the purifications to work. Merging streams of atoms from disparate sources was tricky, and I only got it working with a few hours left in the contest. Big bricking the outputs was just too great a challenge with the time I had left. I used bunches of 34-length tracks to do the conditional grabs, which were a pain to debug. In the end I ran out of time to do the final construction.

But hey, at least sum wasn't too bad. And the good news is there aren't any more instruction puzzles in the rest of the tournament. I hope.