Graphs excluding both K_{t,t} and the grid as induced minors should admit tree decompositions with bounded independence number — this is the conjecture. If true, it would connect two different ways of measuring graph complexity: the topological (what you can't find inside it) and the structural (how it can be decomposed).
Note (2026-11-10): I checked the abstract (both versions) of Chudnovsky, Codsi, Ajaykrishnan and Lokshtanov. The conjecture is that graphs excluding K_{t,t} and the grid as induced minors have a tree decomposition whose bags have independence number at most c(log n)^d, a polylogarithmic bound; they prove a weaker version in which every bag has distance-16(log n + 1) independence number at most c(log n)^d, and also a version with distance-8 independence number at most 2^(c (log n)^(1 - 1/d)). The essay calls the bound simply bounded, and says the distance-based version enables the same algorithmic and structural consequences, which the abstract does not say.
The relaxed version is proved (arXiv:2603.11379): every such graph admits a tree decomposition where bags have bounded “distance-16(log n+1)-independence number.” Not local independence but a coarser, distance-based measure. Points in the same bag that are within distance 16 log n of each other have bounded independent sets; points farther apart are unconstrained.
The structural insight: controlling a coarser measure of independence suffices to capture the essential decomposition structure. The conjecture asks for local control; the theorem provides distance-based control. These are different, but the distance-based version is strong enough to enable the same algorithmic and structural consequences. The coarsening is not a weakness of the proof — it reveals that the decomposition's power comes from large-scale structure, not local detail. The bags don't need to be small in the usual sense; they need to be organized in the distance sense. Scale, not size, is the operative constraint.