Unlike any other cost puzzle I've encountered.  Zorflax, you've outdone yourself.

Obviously you need to spend all 6 access points on inputs and outputs.  Slightly less obvious but still true: in order to make all 3 types of triplex bonds, the arm must be length 1.  So, crowd all the inputs and outputs around the arm, and place 4 no-access glyphs: duplicator, calcifier, triplex bonder, and debonder.

Now, among no-access glyphs, there are still more and less convenient spaces for them.  I defined three separate classes of accessibility.  Any tile that is not within a radius 2 hexagon of the arm, cannot be reached by either end of a lone input, and is thus called "distant".  The corners of that hexagon can be reached in only one way, and I called that "line-indirect".  The centers of the edges of the hexagon can be reached in two different ways, and I called that "elbow-indirect".

The far sides of the two inputs need to be elbow-indirect to support suppression from both directions.  If either is line-indirect, you end up with parts needing to swap places around the arm in an impossible way, before you get very far at all.  The inputs also need to be adjacent, in order to have one object suppress both to give adequate working space.  There are two ways to place the inputs with these restriction, and the one shown here where they are in a line works much better.  It means some of your useful glyphs are behind the inputs, but that doesn't even feel cursed in this context.

The triplex bonder needs one of its edges to be fully in the hexagon, so that you can build any structures at all.  That means it takes up an elbow-indirect slot as well.  There actually is no freedom on the placement of the triplex bonder that isn't accounted for by a symmetry argument.

Then comes the reality check: try to satisfy all of the above demands, and it turns out either the debonder or the duplicator *must* be partially distant.  Initially I dreamed of some clean intermediary from which you could pull off the outputs in pairs and leave the board empty upon loop, but as I reasoned, the access is not convenient enough for this.  So, we brick.

I settled on the receiving end of duplicator being the distant tile.  So within our nice hexagon of structure-free interaction, we have all parts of the solution except that piece of the duplicator, and the furthest pad of the triplex bonder.  Immediately that means that the earth/water input will never be able to bond to the brick.  We sacrificed its capacity to become fire.  So, the single instance of that input that spawns before simulation begins, must be the only instance we ever see.  Juggling it is not the worst thing, as we already promised that an object would need to spend much of the time suppressing both inputs.  But it does just add to the list of fail conditions for the algorithm itself.

Armed with this, I spent a couple nights tinkering, one night pretty sure I was almost there but too tired to think of the remaining work, and then one morning programming it over from scratch, determined.  The result was just under 8000 cycles, with 55 thousand area.  It worked.

In order to create the yellow triplex bond, either for the explosive output or the partial yellow triplex one, a fire from the spine must be calcified.  There is no simple recovery if a triplex bond ever forms in the other slanted direction.  The other two triplex bonds can be routed without needing to modify the spine.

Originally, every time I used the catalyst, it would be a huge ordeal to be able to recover a stable brick.  I realized a better way and cut 1500 cycles and 22 thousand area.

The previous solutions were 1P, but at this point I wanted to call things done, so I did what I knew must be done.  I 6Pd it.  Among my submissions are my best 1P, my best looping 6P, and this, my best nonlooping 6P.  As part of the final output delivery, it allows a new catalyst input to spawn.  This prevents looping, but saves 8 cycles.