People probably have similar mental model on the constraints but here's mine. On the last subcycle only two elements can be duplicated from the berlo, thus the other two elements must have already been present on the molecule.

The baseline way to achieve this for one element is to "store" it on a select atom: have an atom to duplicate from a source and never calcify. This way the first non-salt element value the source ever has is always on that atom.

The first of two important ideas in this solve is a cyclic approach of storing multiple element values on multiple atoms. If you have written an implementation of swap function there's this trick of involving a temp variable like the solve name, which works great here. To store N element values on N atoms, first make temp copy from (calcify and duplicate from) atom N, make atom N copy from atom N-1, etc, make atom 2 copy from atom 1, then make atom 1 copy from temp, and finally make atom 1 duplicate from source.

This storage mechanism faces the same challenge as the baseline approach: the N values it stores are the first N non-salt element values the source ever has. When previous product is outputted, berlo has the two not-to-be-stored elements next to the salts, which makes it hard for berlo to act as the source right away. It can store wrong value to baseline approach, or add an unwanted value to the cyclic approach.

Enters the second important idea: output on the first half-cycle. This way berlo can rotate right away before glyphs act on the next molecule, leaving a salt and then next cycle one of to-be-stored element to the source duplication glyph. The easiest way to output on the first half-cycle is to have an arm hold the molecule and release on the cycle of output.

The cyclic storage layout in this solve brings these two ideas together to reach min area. It stores fire-fire-earth, and happens to rotate to match the corner of output at the exact first half-cycle output should happen, which means this storage mechanism does the work of the arm too.

The rest of glyphs does the rather mundane work of moving values into specified places. The other two fires are essentially baseline approach storage because it's convenient. It also got away from reusing the duplication path that at output time transports earth, since the first time this path activates it's fire on it.

In case it's useful, here's a bunch of layers of this solve: https://cdn.discordapp.com/attachments/830269136583131157/1121695752838713435/Opus_Magnum_-_DYE_HARD_1830G_36_59_2023-06-22-22-51-18.gif
Top row is storage mechanism and duplication path of earth, bottom row is duplication path of air, water, final touchup of earth, air and water. Layers are overlapped in this order.