Excellent answer.

Well I did the math and you should do it. This actually is very efficient.

The problem with this reasoning, especially when we see all the articulations of coherent multipart deductive thinking working in the latest language model and transformers AI, if that it is already off with reality.

And if we understimate the rate at which AI takinging over language related functions brings it much closer to AGI, we risk being set for a dangerous awakening !

[edit : sorry, that's the repost under my account. My bad.]

On the first point I would recommend you stick with a simple ecosystem of same vendor / same model, or better then, with a spare one of another model of another vendor, just in case ;) (as far as monoculture is a concern, that would solve it). But the point is that routers of the same kind tend to have more reasons to interoperate better.

Another interesting choice : Make a map of the estimated coverage interlap of the routers in the field. Then colorize it with a color for each choosen channel, so that never two neighbors (closest overlapping person in that direction) share the same color.
With channel 1, 6 and 11, you would get 3 colors, but nstead of using 3 colors, use 4 colors by letting frequencies overlap better if 3 don't suffice. That would make channels 1, 5, 8, 11 (I prefer to give the extra space to the lowest frequency, on the logic that it spreads a bit better, so would be the least at ease).

Then, of course, if you could use some power control. Listen to neighbors, estimate their activity relatively to yours, and scale your power according to that difference. The less you're active, the less you merit to dispensate your imprint on the local spectrum.

My 2 cents, of course. But I admit I did a Ph.D in a related field.