Every cell holds its own thermal diffusivity α, so the grid is a genuine heterogeneous medium, not one global constant. Each step uses an explicit finite-difference update of the 2D heat equation ∂T/∂t = α∇²T, but with the flux to each of the four neighbours weighted by the average of the two cells' own α — the standard way to discretise conduction across a material interface. Cells you paint as a heat anchor skip that update entirely and are held pinned at the anchor temperature every step, acting like an idealised heater or cold reservoir embedded in the material, not a one-off splash of colour that just fades away.
The outer edge of the grid is an insulated (zero-flux) boundary: energy can only enter or leave through the anchors you place, never leak out through the walls. That means a copper anchor next to a foam anchor will drive the whole board toward a real steady-state distribution instead of decaying to zero, the same way a real insulated enclosure with a heater and a cold plate settles into a fixed temperature profile.
The three material presets use each material's actual thermal diffusivity α = k / (ρcp) in m²/s, scaled down by one constant factor so the explicit scheme stays numerically stable — the ratios between materials are exactly the real ones. Copper conducts roughly 1,300× faster than pine wood in this model because it does in reality too; raise Steps / frame to fast-forward through the slow materials.