We report an environment-dependent weak-lensing measurement using KiDS shapes and a bright photometric lens sample split into low- and high-environment subsamples via a κ-density proxy (δk/Σk). For each subsample we compute excess surface density profiles ΔΣ(R) and form the differential signal ΔΔΣ(R)=ΔΣlow(R)−ΔΣhigh(R), using the cross component as the primary shear-systematics null test. Uncertainties are estimated with spatial jackknife, and the full covariance is propagated through all inference steps. The tangential differential signal is highly significant and persists under strict (M⋆, z) matching. An additional star-formation-rate control reduces the overall amplitude by ≈20% but leaves a strong residual. After removing the best-fit constant offset per redshift slice, a coherent radial residual shape remains in the stacked tangential component (15 radial bins), while the cross component is consistent with null in the matched construction. The residual exhibits an outer suppression/tilt toward zero at large radii, also captured by a low-order polynomial in ln R. We compare the stacked residual vectors to external-field effect (EFE) lensing templates (Bílek et al.) using a joint full-covariance framework with a shared log-radius rescaling s and per-variant amplitudes. Model selection via AIC strongly favours the strong-field template ge/a0=0.07, with only mild radial rescaling (s≈−0.065, i.e. a ≈0.86× radius compression). The template fits original, matched, and SFR-controlled variants with ordered amplitudes |Aorig|>|Amatched|>|Amatched,sfrw|, consistent with a stable shape whose normalization decreases under tighter confounder control. To address standard ΛCDM-oriented critiques, we test halo-physics competitors (including a two-population central–satellite truncation model) and an extended three-component HOD-proxy basis. We also perform a shared intrinsic-alignment hardness test by adding a single IA-like contaminant amplitude common to all variants. None of these flexible alternatives reproduces the residual shape at comparable information criterion; the preferred EFE template remains strongly favoured (ΔAIC ≳ 60 relative to HOD-proxy and shared-IA competitors). The repository includes the analysis products required to reproduce the figures and tables.
Jan-Frederik Flügge (Fri,) studied this question.