From the guys that make smokeless powder. kernel gets hot from primer blast. Surface MELTS and creates a surrounding cloud of non-burning gas. gas on the outside (plasma) burns, heating the gas and more surface melting. IF the pressure is NOT high enough, the plasma moves away from the gas cloud (gas cloud has pressure), cooling it and slowing MELTING. Plasma needs to be HOT enough and CLOSE enough to start the process on adjacent kernels. Air is a horrible heat conductor so airspace due to large case slows the process. Close kernels increases the total reaction. Great, how does that effect our loads? Compressed loads cause a very fast total reaction - bad. Seating depth changes effective case volume. Heavy bullet, hard crimp or jamming the lands slows bullet release and keeps pressure up. Once bullet moves, pressure decreases. Flake powders have more surface area, rods have lesser so MELTING is slower. Retardants used to slow the MELTING. Filler moves the kernels closer as does high fill. Position sensitivity - one side of the kernels is cold (case) and doesn't melt easy. Kernels are not adjacent, same result. Really low loads have very slow and incomplete reaction. Saw on another site, OP thought bang was the 'explosion', really is the shock wave at the muzzle. Gas fps can be 2-3x the bullet fps. Factories chose powder to get high fill rate for desired performance, we don't often/sometimes. Kinda funny to hear a pressure trace company claim a SEE occurs (sometimes) past the end of the muzzle. They recorded the pressure spike after bullet exit. Reflected down the barrel. Hot supersonic gas tends to illuminate air. Some may claim excess powder burns after the muzzle - it may glow, doubt if it burns. Doesn't help picking powder but may help explain why some don't do what we want.