CaveatThis is a simplified, illustrative model built for this animation: a capped population of 48 lineages with a single escape mutation class, not the project’s real Wright-Fisher codebase (Ne=10,000, three tracked escape loci, corrected generation-timing). But the mechanism is the same one the real research isolated through extensive testing: distinct resistant lineages really do arise independently under selection, and dense sampling really does catch most of them; the shortfall shows up specifically when sequencing depth is realistic, not when the model is wrong. All three candidate explanations for the gap (a missing mechanism, an algorithm quirk, a counting bug) were tested and ruled out; what remained was a population model that doesn’t generate enough real diversity in 3–11 simulated generations to match what real, deep patient sequencing shows. Drag the depth slider above to feel why.