Clarify converter output is not style guidance

docs/research/geometry-script-fork-notes.md

@@ -112,13 +112,15 @@ Current acceptance:
 - Converter runs in Blender 5.1.1.
 - Generated drafts compile as Python for the tested groups.
 - The full 176-node `Field Value` source group emits a 179-line draft script.
-- Generated text is a bootstrap draft, not doctrine-quality final code.
+- Generated text is a bootstrap draft, not doctrine-quality final code or style
+  guidance.
 
 Known limits:
 
 - Nested source group calls are emitted by source group name, so dependency
   groups must be converted or replaced by hand-authored equivalents.
 - The generated script preserves node-level wiring and therefore can be noisy.
+- Generated names and ordering reflect extraction mechanics, not spatial intent.
 - Behavioral equivalence still needs the same evaluated-geometry or scalar
   harness verification used by the hand translations.
 

docs/research/procedural-doctrine.md

@@ -16,6 +16,11 @@ Code snippets are excerpts from the verified translation scripts unless noted.
 They may omit surrounding setup or sibling branches, but they should keep the
 same explicit Geometry Script call style that passed Blender verification.
 
+`nodes_to_script` output is evidence, not style guidance. Use generated drafts
+to recover wiring, defaults, socket order, and hidden graph structure. Do not
+copy their naming, ordering, or mechanical transcript shape into accepted
+Geometry Script. A useful draft can still be ugly enough to need supervision.
+
 ## Start With A Pipeline
 
 Reach for this when the form feels like it has stages: a seed shape, a sampling