First some simplifying assumptions: We know that the berlo has to spin, because only 2 atoms of berlo can talk to the grid, so assume it has a single rotate instruction (period 6). I want the board to look the same every time an input spawns. The first input spawns on a pre-movement half cycle with a particular berlo orientation. Thus the solution expects to output post-movement on cycle 6, 12, etc. Now let's talk about memory. You can add a half cycle delay when propagating some element by having the order of duplication reversed: "b dupes onto c" before "a dupes onto b" means that "a" gets to "c" with a half cycle delay. That's fine, but the berlo only moves every 2nd half cycle. So, any chain like this has two adjacent signals of the same color, whereas the output doesn't have any adjacent atoms of the same color. We solve this by using atoms in the center of the grid as "wires" and having our "memory" cells around the perimeter. Wires allow information to propagate to nonadjacent atoms, by doing chains of correctly ordered duplicators (no half cycle delay), and then re calcifying the atoms that were part of the wire. This is the general strategy, now let's look at the specifics. Label the atoms in page order as 0123456789abcdef Berlo talks to e and f. When the thing outputs, we need e to be earth, so our assumptions dictate that berlo's initial position has earth on e and fire on f. Alphabetically convenient. For wiring, the least convenient atoms are 0 and f, but they can just be set to fire and ignored. Since we have fire in the initial orientation of berlo, we don't even need calcifiers under these atoms, they just become fire and stay that way. e is how the information makes it into the memory cells. So now let's look at how many half cycles of delay we need to keep air and water around by post-movement cycle 6. This number is lower if we have ccw berlo rotation, which is why I picked ccw. Air is 5 half cyles old on the subcycle we intend to output, so that is the maximum delay we need. In order to have 5 half cycles of delay, we need to designate atoms to be 4, 3, 2, and 1 half cycle delay as well. My layout uses the rectangle of water-to-be atoms here, aka 2, 7, 8 & d. Some water is forced because the age 3 and 4 half-cycle memory cells want to be water when things are output. The 1 and 2 half-cycle memory cells are unforced at the time of output because the berlo gave them salt. We can make them hold water very easily by having the 1 half-cycle memory cell act as a berlo override. One conditional duplicator (target atom may be already an element) exists to do this. The deepest glyphs in the stack are responsible for, in order: fill the non-wire 5-half-cycle delay cells (b and c) based on the content of the 4-half-cycle memory cell (d) fill d from 8 fill 8 from 2 fill 2 from 7 fill 7 from 3 berlo to 3 all of the above is just passing information one cell deeper every half cycle then we do the conditional dupe from 3 back to 7, which places water when the berlo has salt and makes sure we have water in 2 and 7 at the time of output then we fill in everything else - on cycle 6 every element is available somewhere usable so this step is just routine. We first need to calcify everything apart from 0, 4, and f which get filled correctly the first time, and any memory cells. In total this solution has 36 duplicators and 32 calcifiers, down 50g from my first attempt. Not sure how much lower it goes and what I maybe missed.