My first solve had a tape loop of length 18, and I ended up with a pretty smooth wasteless way to handle the metals with Rate 54, but it was very instruction heavy.

I thought to retry with a tape loop of 12, to really trim down the instruction count, but I had to completely redo the metals path. In the end the two were surprisingly similar in score, but the 12 track layout had much more space to optimise so I went with that. There's only a handful of arms with 10+ instructions (ignoring repeats), so probably I could have reduced the tape loop down to 8 or 9, but it's a pain to re-time all the initial setup. Taking a small hit to the rate for the Salt elements would be worthwhile if it makes the metals loop faster, since this setup needs 6 tape loops to produce the metals and there's more cases. 

I tried to work out ways to make some of the metals that could loop in 2/3 cycles but I couldn't find anything which didn't cost a ton of extra machinery and cost is already the largest part of my sum metric. Maybe longer tape loops with more quicksilver draws per loop is the way. Probably it's worthwhile to extract the lead path at least, since that can be handled with a chain of inputs, though knowing how to enable/disable it is an interesting problem.

Anyway, this was a really fun computation problem and the template atom style really makes it easy to retry the different setup of the machine. Much easier than look-and-say from last year!