Semantic extraction - The code still needs real per-policy parsing that reads values from Apigee XML and JavaScript instead of selecting static Flex YAML templates.
Confidence scoring - The current mapping score is still mostly stubbed, so the code can validate output files while still reporting many low-confidence source XML mappings.
Generated target ownership - The repo needs a decision on whether Flex YAMLs under Adhoc/flex-gateway-policies are permanent golden files, temporary references, or seed templates for generated output.