diff --git a/notes/obj-to-data-structures.md b/notes/obj-to-data-structures.md
new file mode 100644
index 0000000..9c4c334
--- /dev/null
+++ b/notes/obj-to-data-structures.md
@@ -0,0 +1,31 @@
+# How to go from a parsed .obj file to the starting data structures for subdivision
+
+Here's the data structures we need to derive:
+
+```
+vertices_size         # how many vertices in level-0 + level-1
+                      # this is the entire vertex buffer
+base_vertex_positions # given in .obj
+base_faces            # given in .obj
+subdiv_1_faces        # must be generated
+triangle_indices      # f(base_faces, base_face_valence)
+                      #   and subdiv_1_{faces, valence}
+subdiv_1_triangles_count
+base_face_valence     # f(base_faces)
+base_face_offset      # exclusive-scan(+, base_faces)
+base_edges            # f(?)
+base_vertex_valence   # f(?)
+base_vertex_offset    # 2 * exclusive-scan(+, base_vertex_vlnc)
+base_vertex_index     # f(?)
+base_vertices         # f(?)
+```
+
+## Proposed data structure changes in subdivider code
+
+`offset` and `valence` can probably be put into one data structure that covers all vertices (and is the size of the vertex buffer). `valence` could be done as a segmented scan if necessary.
+
+Be better at explicitly exposing hierarchy (many data structures will be indexed internally by hierarchy level)
+
+## Q: When to convert from 1-based indexing (in .obj) vs. 0-based indexing (in subdivision code)?
+
+Probably after parsing (don't change the parser), then we can update the parser / get a different parser and use that unchanged.