Source Geometry and Receiver Ascertainment Limit Spatial Perturbation Inference

Spatial perturbation assays locate targeted cells, but neighboring-cell responses can also reflect tissue structure, imperfect guide detection, and dependent sampling. We evaluated these alternatives with a spatial perturbation field diagnostic workflow combining independently specified receiver scores, spatial adjustment, and explicit source references. In 3 Spatial Perturb-Seq chips spanning 4 mice, a Clu-associated score was not exceptional under the stricter reference; training-only mouse holdouts also failed to establish stable program transfer. Known-effect injections showed that refitting spatial background retained 73.7% of the contrast gain at a 150-μm decay scale and 63.3% at 600 μm. In a separate count-level simulation, 3.82% receiver contamination by missed sources produced 33/100 zero-field rejections with source-autonomous expression changes, versus 3/100 without them. Perturb-map stromal contrasts were positive in 2 animals, but dependent sections, unresolved clone identities, and insufficient matched-placement support prevented population-level confirmation. External assays, injury, and clinical cohorts did not establish broad transportability or incremental prediction. These results distinguish limits in receiver ascertainment, source-reference validity, and biological replication, providing reproducible diagnostics rather than a generally calibrated test of causal propagation.