The reframed question cuts right to the physics. According to Budkewitsch and Robin (1994), the surface of a cooling lava flow initiates cracking in a chaotic rush — producing T-type and X-type quadruple junctions, curved lines, and wildly irregular polygons. This is the 'immature' stage. Their VOPONUCE algorithm models what happens next: each propagating crack must follow the steepest thermal gradient, meaning it tracks the hottest nearby point — invariably the centre of an adjacent, larger column. This creates a feedback loop in which crack boundaries continuously renegotiate territory between unequal columns, functioning as a natural shape-corrector. Mathematically, when this process drives all junctions from T-type to Y-type triple junctions, the average polygon count is forced to exactly six — not by design, but as a topological inevitability. Yet Hetényi et al. (2012), across 50 field sites, measured a mean of only 5.71 sides per column. The shortfall is time. Cooling rates outpace the maturation process before the network can converge. Columns near the surface, where cooling is fastest, have the fewest sides on average. Deeper columns, cooled more slowly by groundwater infiltration, approach the hexagonal ideal more closely — a gradient that itself encodes the thermal history of the entire flow. The 'six comes free,' as one recent analysis put it, from Y-junction topology alone; the imperfection is simply the thermodynamic race being lost.