Building two different products from a single set of inputs is always tricky when the goal is a minimal programming tape.  Typically, it involves complicated timing, routing partial products in different ways on subsequent tape loops.  The problem is, most of those techniques consume a fair amount of space, and there is VERY little to spare in this cabinet.

But as I was starting my exploration of this puzzle, I remembered a trick kaliuresis used to win the PIG metric for Probe Module in 2023: build a machine that produces one of the products six times, and then introduce a delayed-start action that converts the entire machine to make the other one forever after.  

And when I reviewed the leaderboards, it turns out that this technique has been successful in freespace TI and production I a modest number of times, notably Mist of Incapacitation, Van Berlo's Pivots, Quintessential Stabilizer, and Fragrant Powders--the last of which, I think, is the nearest thematic cousin of this puzzle.

You can see the implementation of that idea here with the long-delayed start on the arm that pulls copper inputs.

*********

I am still a little dissatisfied with the fact that creating triplex bonds costs 10i, but every more-economical method I found required at least one more hex in the cabinet.  I also think there must be a way to improve the piston that rearranges salt-silver dyads, which burns 6i--the most costly single element of this machine.  Even so, sub-30 feels like a great achievement.  

Final breakdown:
-Building the triplex product costs 10i
-Building the silver product costs 10i (including 3i on the late-start and the funky pivot on the hexarm, which is non-functional in the triplex product)
-8i is shared by both processes