Discharge schemes in particle models of thunderclouds remove charge from a stochastically grown channel, which does not conserve charge and imposes the channel geometry rather than deriving it. This paper replaces the scheme with a plasma process: when the field in the gap between two oppositely signed charge layers reaches the ionization level the gap is filled with neutral air at the simulation’s own initial density, molecules above threshold separate into a positive ion and an electron, and everything after that follows from the force law that moves every other particle. Charge is conserved at every reaction. Two quantities such schemes leave free are derived rather than chosen: the separation at which a new pair is unbound, from the ambient field, and the timestep, from the electron drift velocity at the reduced field the model itself produces, E/N = 27 Td. Layers assembled by hand show what the mechanism requires. The same charge spread smoothly through each layer gives 19 kV m⁻¹ and never fires; gathered into local concentrations it gives up to 196 kV m⁻¹ and does. The requirement is a layered and locally concentrated charge structure, not a rotating one. Against matched controls the field in the firing gap falls 55.8 ± 6.5 per cent over eleven seeds, with relief of neighbouring gaps attenuating to zero by 2.16 km. Without an electron attachment loss the discharge never terminates; with attachment on oxygen all ten discharges self-terminate, at a median 190 ms and 1.44 × 10⁴ m s⁻¹ against 192 ms and 1.69 × 10⁴ m s⁻¹ for flashes measured in a mapped storm using the model’s own definitions. The reach is a factor of four short, which bounds the result to initiation and local field relief rather than to flash extent. Two predictions follow that observation can test: initiation should track the concentration of charge within a layer rather than the layer-averaged field, and flash duration should scale with the oxygen number density that sets the attachment length.