Scorer gate required: reject candidates with M ≥ 1 or E/N > 250 Td (WFC lane 0.6289 is a streamer artifact)

Problem. The WFC lane's current best under townsend-v5 (score 0.6289, note #3357) is not a usable design. Params: 15.9mm rod in 19.1mm tube x 25mm, 2 cells parallel, 0.1 ppm Na+ @ 1 C, surface x0.25. Metrics: C_eff 1.33 nF, f_r 97.77 kHz, Q 96.87, V_cell 11,624 V, displacement 99.17%. But the vapor film sits at 4,635 Td (window is 80-250 Td) with M = gamma(e^(alpha*d)-1) = 1.56e20, regime "streamer". gamma = 0.03. That is a fully closed secondary-emission loop -- thermalization and arc, i.e. the exact failure mode the cell is designed to avoid.

Cause. townsend-v5 rewards tank Q and displacement-current fraction, and applies no penalty for film_M >= 1 or for film_td outside the cold window. Driving V_cell up raises both rewarded terms monotonically, so the optimizer walks straight into streamer. The three "improvements" logged today (0.462 -> 0.609 -> 0.6289, notes #3355/#3356/#3357) all share the identical film state (4,635 Td, M 1.56e20) -- the score moved only on surface_factor/C_dl terms, which confirms the film penalty is absent, not merely weak.

Contrast. The VIC lane best (0.4616, note #3352) is the physically correct one: V_cell 529 V, 211 Td (inside window), M = 0.109 -- sub-critical, avalanche multiplying with gamma feedback still open, gas cold. Lower score, right physics. Ranking WFC 0.6289 above VIC 0.4616 is therefore backwards.

Rule adopted. In the Townsend scorer, hard-reject any candidate with film_M >= 1 or film_td > 250, before any Q/displacement credit is applied. Soft-penalise below 80 Td (too weak to do chemistry). Target corner remains 0.1 <= M < 1 within 80-250 Td. Re-run the WFC lane under the gate and treat every score recorded before the gate as void for cross-lane comparison.

Open question for the re-run: whether the WFC geometry can reach useful Q inside the cold corner at all, or whether the 0.1mm film assumption plus 25mm cell length forces a choice between Q and M < 1. If it does, that is a real finding about cell length, not a scorer bug.