Sheesh.

So to start, 43 is optimal.  You can increase the number of atoms on the board by 1 each cycle (counting already-dropped products as 7 atoms on the board).  42 atoms go into the products, adding the final drop on cycle 43 gives you the theory min.  But when trying to achieve it, the following things are challenging:

- you need an atom on the embedded multibonder on every single cycle from 2 to 42 (condition 1)
- you need no extra atoms on the board when dropping the 6th product (condition 2)

Each miss here adds one cycle.  The first design I built, I just wanted to see what a 7 cycle steady state loop would look like.  I allowed 2 extra grabs on the center to get wands, immediately costing 2 cycles to condition 1.  With 5 atoms left at the time of dropping the 6th product, each a lost cycle to condition 2, it clocked in at 50 cycles.  By performing a final "squeeze", making the 6th output out of wands in an ad-hoc manner, I got it to 48 but it wasn't ever going to be 43.

I then used some choice pivots on the initial wand to get new wands in a way that didn't lose any cycles to condition 1.  I initially made a 49 with 6 atoms left on the 6th product, and then squeezed it all the way down to 44.  However, I had made a crucial error: latency.  Whatever atom is the last one to leave the center, has to become part of an output in some other way.  On every step between, there needs to be some other output getting built and dropped.  Otherwise it is a guaranteed miss for condition 1.  This also means that the path it takes can never block the output, so the simple slide pivot is out, though I had already noted that.  The 44 failed because the latency was 4, and there wasn't room to output on the last 4 cycles.  I needed to redesign and either tighten up the latency to its minimum of 3, or plan more space to store more outputs.

Genius had built a solution early on that takes the "store more outputs" approach.  He just extruded bricks of 12 atoms, reshuffled them, and then pushed them onto the output using over 1200g.  This took 54 cycles although could be tightened up.  This is simple and attractive, albeit very expensive.

However, I wanted something slightly cheaper, so I set off on a hybrid solution with a 5 cycle inner loop, allowing 2 reinserted wands per output.  And here it is.  Initially 910 cost, trimmed gradually down to 835.  Its main expense actually comes from gathering and processing all of the atoms generated before the steady state begins.  It may be more economical to make a 6 cycle inner loop to alleviate that burden, but that comes with its own drawbacks.  Hard to say.

At any rate, we finally had the 43, and I expect it's not something many others get.