Act as a senior TypeScript engineer. Parts of my {{tech_stack}} project are loosely typed: plain JavaScript files, any everywhere, or types that don't match the real data. Start with {{page_or_feature}}. Add real types without changing what the code does. 1. Survey first: count the any types, untyped files and type assertions in this area, and list where data enters the app: API responses, database queries, form inputs, URL parameters, local storage and third-party libraries. 2. Type from the edges in: - Database types generated from the schema with the project's tooling, never written by hand - API and form data validated at runtime with the schema library already in use (or Zod if there's none, after asking me), with types inferred from those schemas - Shared domain types, such as a booking, an order or a profile, defined once and reused - Then component props, hook return values and function parameters 3. Make the types honest: fields that can be missing are optional and handled; data you haven't checked is unknown until narrowed; real states such as loading, error and success are a union, not several booleans. 4. Rules: no behavior changes. When a type exposes a real bug, stop and list it for me rather than fixing it in the same step. No any, no non-null assertions on data that can be missing, no ts-ignore. Small commits, one area each, with typecheck, tests and build passing after every one. 5. Turn on stricter compiler settings gradually, one flag at a time, only once the code passes. Finish with before and after counts, the bugs the types uncovered, and the next area to type.