Update README.md

README.md

@@ -3,6 +3,15 @@ license: mit
 ---
 # GeoSpatial Prior: Synthetic 3D → Geometric Substrate Training
 
+# Incomplete Documentation
+
+Claude provided an early document and it's full of problems that will need smoothing and refining.
+
+This will not behave exactly as Claude says it will and there will be multiple refactors and compromises along the way.
+
+# Claude below
+
+
 ## Abstract
 
 A system for teaching geometric spatial reasoning to neural networks by rendering deterministic 3D scenes where every spatial relationship — position, occlusion, depth, lighting direction, scale — maps directly to known simplex coordinates. Rather than inferring geometric structure from 2D pixel statistics, we construct ground-truth spatial labels from a sectorized 5×5×5 perspective volume and use those labels to pretrain both a geometric classifier and a geometric CLIP variant. The result is a transferable spatial reasoning backbone that can be finetuned into any vision model, providing compositional understanding that current models lack.