Where cix falls short.

Lua is the most-discussed gap (Kong scores 1–4/15 for this reason). Erlang, Elixir, Crystal, Zig, Lisp/Scheme dialects, Haskell, and Scala are also unsupported today. On these projects, orientation explicitly tells you parsing is unsupported and you should expect to fall back. cix never returns a confident-looking but wrong answer when the parser doesn't exist.

Loki registers most of its primary HTTP API through a runtime module manager. rclone uses a runtime registry function. Static analysis can't fully see this. cix captures the static portion and flags the dynamic gap honestly. On these projects, treat the route view as partial.

When two unrelated types share a name across packages, impact analysis sometimes resolves to one when you meant the other. The output flags the ambiguity, but in time-pressed workflows it's real friction. Package-qualified lookups by default are on the roadmap.

Custom migration runners, ALTER-without-CREATE migrations, Postgres-specific types (tsvector, custom domains), and ORM-only fields are partial or best-effort. The schema confidence tier surfaces all of this — when the schema view says 'partial,' it explains what's missing.

Some projects register routes for testing purposes. Directory heuristics handle the common cases (test/, spec/, __tests__/), but cases that don't match leak through. Better classification is in progress.