Can Beam Reports Update a Digital Twin? Identifiability of Reflector Geometry and Material from Wide-Beam RSRP

Cellular networks already collect phaseless, 1 dB-quantized power reports on a few wide beams. We ask what they reveal about a reflecting wall in a radio digital twin. Without a second coherent path the reflection phase is invisible; one user never locates the wall, two generically do, and two at distinct incidence angles identify a Fresnel slab's permittivity and conductivity. In ray-traced Munich and Paris scenes, with user positions known to 1 m, ten users bound the wall offset to 0.8-3.0 m. A maximum a posteriori fit meets the bound locally, but a multimodal cost sends 2-9% of fits at 1 dB noise (up to 24% at 0.25 dB) to distant basins. The material is locally but not globally identifiable and is better classified, with 78-86% accuracy among seven classes. A map prior helps only until the wall has moved by about the radio-only error. If the twin models the wall's two-bounce paths, a likelihood-ratio test keeps its 1% false-alarm rate on fully ray-traced reports. With ten users it detects a 4 m move in 92-100% of model-generated cases and in 86-91% when the moved facade is re-traced, and a 2 m move in 39-59% and 71-82%. The same factorization reduces a 64-element malicious surface to one complex number per user; searching it directly lowers the top-1 accuracy of a learned beam predictor from 26.8% to 11.3%. Author-prepared preprint for the IEEE International Symposium on Joint Communications & Sensing (JC&S) 2027. Not peer reviewed.