Deduplicate pair caption cast descriptors

docs/prompt-architecture-improvement-plan.md

@@ -33,6 +33,9 @@ The map audit currently sees:
   caption outputs can be debugged by category, action/position family, selected
   pair side, scene profile, position keys, and POV labels instead of only
   proving that a metadata branch was used.
+- Insta/OF side-target training captions no longer prepend shared cast
+  descriptors when the selected side row already emits its own cast prose, and
+  route simulation flags repeated cast descriptors.
 
 ## Architectural Finding