Behavioral WFC leakage must be driven by double-layer overpotential, not total cell voltage — and needs a mass-transport ceiling

Arising from review of Scotchn's WFC_Behave.sub (LTspice), #neuralstan20-testing 2026-09-14.

The original leakage source was Bgas P N I=(exp(limit((V(P,N)-Vgas)/Sgas,-50,50))-1)/Rgas with Vgas=2.4, Sgas=0.08. Two independent problems, only one of which is the obvious one.

1. FUNCTIONAL FORM WAS CORRECT. Butler-Volmer in the Tafel limit is i = i0exp(eta/b), with b = RT/(alphan*F) ~ 51 mV at 298 K for alpha=0.5, n=1. Sgas=0.08 (80 mV) is a defensible Tafel slope. Exponential onset is real electrode kinetics, not hand-waving. My replacement (tanh with Sgas=40 V) was numerically safe but chemically meaningless, and its soft knee conducted ~85 mA at 168 V, clamping the cell and flattening an entire frequency sweep (168.0 V at 300 Hz falling monotonically to 161.8 V at 1 kHz, 3.7% total swing, no resonance visible). Unclamping (Vgas=2000) at 620 Hz raised cell peaks from 165-166 V to 485-498 V and revealed a real resonance peak at ~3.16 kHz (cell 1 at 2.51 kHz).

2. THE ACTUAL CONCEPTUAL ERROR: the exponential was fed V(P,N), the total cell voltage. Tafel eta is the overpotential ACROSS THE INTERFACE only. In a WFC most of the applied potential drops across bulk water and the dielectric barrier; only a fraction lands on the double layer. Feeding hundreds of volts into a term calibrated in tens of millivolts is what made it deliver ~1e17 A. Leakage must be driven from the double-layer node (Rion/Cdl junction), i.e. V(Ps,N), not V(P,N).

3. MISSING PHYSICS: Butler-Volmer alone is unbounded because it describes charge transfer only. Real electrodes saturate at a mass-transport limiting current. Correct mixed kinetic-diffusion form: i = (i_lim * i_BV)/(i_lim + i_BV) - exponential at low eta, saturating at i_lim.

GENERAL RULE for any behavioral WFC model: check which node the Faradaic term senses. If it senses total cell voltage it will either explode or clamp, and either way it destroys the measurement. This is the Figure 8 split (US 4,798,661) expressed as a netlist: displacement current rides C_dl*dV/dt on the whole cell; Faradaic leakage rides overpotential at the interface. Two different voltages.