While the ratio of 5 input sets to 3 output sets is clean, the distribution is not.  Of the 3x5 water, 9 go to one output and 6 to the other.  Of the 2x5 iron, 8 of them purify to copper, in two different places by necessity, while the other 2 project using 2 of the 5 quicksilver.  Splitting everything efficiently is key.

Then there's the matter of what the machines that handle the assembly look like.  3 per 10 is notably faster than 1 per 4.  I was torn between having hex arms and track loops that act every 3rd cycle, with one cycle downtime after 3 repetitions, or extra pipelines for everything so that I could use period 4 mechanisms.  In the end, each choice was appropriate for a different section of the solution.  The purification-only output, with the 3-cycle purification glyph, wanted the tri arms (track loops used sparingly).  The projection side wanted separate and slower constructions.

I am most proud of the single input and assembled trio of water that pass between the two stations.  These help balance everything, and were where I saved the most cost when trimming from the starting point of 825g.