Stand close to a pointillist painting and you see dots. Step back and you see a face. Nothing about the painting changed — your description of it changed scale, and at the new scale some information (each dot's exact color) is gone while other information (the face) is not just preserved but finally visible. Physics has a precise version of this step-back, and it is one of the great ideas of the twentieth century. This section puts the knob for it in your hand.
Take a lattice of the site's usual kind — a grid of little units, each in one of two states. You can read the two tones as a phase snapped to its two extremes: gold for "up", dark for "down". (Everything here works for full continuous phases too; two tones just make the bookkeeping visible.) Now do the following, exactly:
Cut the grid into 2×2 blocks. Replace each block by a single cell carrying the block's majority vote. You now have a lattice half the size. Repeat.
That is the whole operation — physicists call each application a blocking step, and the repeated procedure coarse-graining. Each press of the button below throws away three quarters of the information. The question that makes the idea deep is: which three quarters?
Play with it. Two things reward attention. First, the speckle — single flipped cells and tiny clusters — vanishes in one or two presses: a lone dissenter is always outvoted by its block. Second, the domains — the big connected regions — survive press after press, keeping their shapes while losing their texture. Squinting is not a uniform forgetting. It is a filter: it forgets the small and keeps the large. Whatever in the picture was organized at scales bigger than the block survives to the next level; whatever lived at one-cell scale is gone.
To see how sharp this filter is, feed it two extreme diets. On the left below: a field with genuine large-scale structure (smooth blobs, thresholded to two tones). On the right: pure coin-flips, no structure at any scale. Same squint button for both.
This pair of behaviors is the working definition this module runs on: structure is exactly the part of a description that survives a change of scale. Noise is the part that does not — and no amount of squinting turns one into the other. When the essays say a thought, a molecule, or a level of the scale ladder "emerges", the honest cash value is always this: there is a description that survives blur, sitting inside a sea of detail that does not.
One more thing before the next section, because it is where everything is headed. As you squint the thermal lattice above, its agreement number a changes — the coarse lattice is not just smaller, it is statistically a slightly different lattice, usually a cleaner or messier one. So squinting doesn't merely compress a system. It moves it. Where systems move to, when you squint them over and over, is the subject of the next section — and the answer is the mechanism behind the site's whole picture of criticality.