Designed for WGC.

So here's my thought-process on how I arrived at this.
My first question was what even is minwidth for this puzzle? I'm not super experienced and had never done width before, so I wasn't sure. I proved to myself that W3 (ie, can fit a disposal) is possible really easily, with a piston arm on a really long track with work-stations laid out to its right. I also proved to myself that W2 is impossible, because there is no motion you can do to a W2 input that does not add at least a half-width, and without some kind of movement you can't start reducing the input down to a smaller size. Same deal for the output. 

In setting up a width 2.5 solution, I made a couple observations: first, the input, output, and quintessence-generating glyph (all being W2 and 3 tall) naturally push the design to be arranged as two or three semi-separate areas separated by these components. In my first pass, I had three arms on three tracks: the two arms you see now, doing the same approximate job, plus a third arm past the output that took lead from arm 2 and generated a waste chain. I think this is about your only option for making this machine looping, but it cost a lot so I decided it wasn't worth it. I figured I could just add a bunch of track to arm 2 to give it room to stick a bunch of lead out of the way of the assembly.

This precipitated design 3: It looked a lot like the current design, but had two bonders, two unbonders, approximately 4 track for arm 1, and something like 10 track for arm 2. I managed to find places to permanently store all six lead such that I could entirely ignore them after moving them off the glyph of rejection with just some careful arm maneuvering, but this meant it cost almost as much as just having a waste chain with its own dedicated piston arm. It was at this point I realized though: while it might be slow and take a while, I didn't *have* to put the lead in places that were out of the way. I could just make way in a specific area when I needed to.

After this, the prevailing project was reducing the amount of track I needed. I couldn't think of a way to avoid having two piston arms, so the only real place to save cost was glyphs and track. Due to a degree of uncertainty over how much was possible, I took this very slowly: I would sit for a while, come up with some movement that would allow me to remove a track or a bonder/debonder, implement it, decide *this* was mincost, and then walk away, and repeat after a few hours. This wasted a lot of time, and at this point I have the requisite motions for arm 2 to most efficiently assemble an output with one bonder or two thoroughly memorized. 

But I was able to keep cutting until I arrived at this layout. You need both the end tracks on each side of the machine for sure (to enable picking up the input/dropping the output/passing atoms to the other arm, otherwise there's not enough access I believe) and the middle track-hex is necessary to enable some weirder movements and simply allow movement between ends. This design, which I couldn't save cost on anymore, just needed to be made fast. The submitted version is ~250c faster than the first design using this layout I made, but there weren't any major logic changes other than what I think is a very slick process for disassembling the input and a bit of movement-minimization in my waste-positioning to prioritize more easily-accessed areas over more inconvenient ones.

Some side-notes and predictions to close out my yapping: it was fun playing with a tournament puzzle for the first time, and I figured out a lot of interesting methodology from improving this. If there's any more cost to cut, it's going to be via something deeply weird I've never heard of, or just a specific layout that somehow enables assembly/disassembly with only two track. You can definitely go faster than this though, and I will be surprised if this solution places all that well for that reason alone. I hope y'all found my notes interesting, I had fun making this solution and wanted to talk about it.