A Model-Competition Test of the ECSM Target-Response Layer Against Generic Baselines in Measured Molecular Scattering

This paper reports V40 of the ECSM electron-like packet measured-scattering programme. V38 introduced a predictive ECSM target-response layer for measured sevoflurane elastic electron scattering, and V39 showed that the same response form replicated on methane, including a successful cold-transfer test from sevoflurane-trained parameters. V40 performs the next required check: direct model competition against generic baselines. Two measured molecular targets are used: sevoflurane and methane. Both are evaluated over the measured 25–125 degree angular domain and the shared incident-energy range 50–300 eV. The same train/test split is used for both targets: 50, 150, and 250 eV for training, and 100, 200, and 300 eV for held-out testing. The frozen ECSM electron-like packet is unchanged. The ECSM response form is compared with bare ECSM, one-feature screening and polarisation baselines, generic second-order log-q and angular polynomial baselines, and generic fourth-order log-q and angular polynomial baselines. The ECSM response form passes the held-out residual threshold on both targets, giving held-out mean residuals of 0. 185 for sevoflurane and 0. 174 for methane. However, it is not the best-performing model. The best methane model is a generic second-order log-q polynomial with held-out mean residual 0. 149. The best sevoflurane model is a generic fourth-order angular polynomial with held-out mean residual 0. 162. The ECSM response form remains within 20 percent of the best model on both targets, with ratios 1. 163 for methane and 1. 142 for sevoflurane. The final verdict is PASSWEAKECSMRESPONSECOMPETITIVENOTBEST. This is not a strong model-selection win. It is a competitive-survival result that motivates the next stage: a more constrained ECSM target-response law with fewer arbitrary freedoms or physically derived target descriptors.