Compilation model

JIT treats a schema as compile-time input. Schema traversal, wrapper resolution, strategy selection, optimization and source generation happen once; hot calls run plain JavaScript specialized for one known shape.

JIT builders -> schema AST -> operation plan -> optimizer -> code generator
                                                       -> runtime Function
                                                       -> pure AOT module

Runtime JIT

Runtime operations compile through globalThis.Function and are cached by schema identity and operation plan. External values such as predicates and query constants are passed as bindings, never interpolated into generated source.

const isUser = JIT.validate(User).is().compile();

isUser.source;
isUser.hash;
isUser.explain();

The default cache, cold-compilation configuration and the separate hash/index caches are covered in cache, hash and indexes.

Build-time AOT

jit generate evaluates explicit declarations and writes self-contained ESM, CJS and declaration files. Generated operations have zero imports from the JIT engine, so the schema builders and compiler do not ship in the production bundle.

Why the generated path is fast

• known properties use direct access;
• arrays use indexed loops;
• validators order cheap checks before regex and transforms;
• queries fuse filtering and projection;
• binary rowsets replace string discriminators and booleans with integer codes;
• unused operations are never generated.

Compile cost and execution cost are reported separately in benchmarks. A complicated optimization is accepted only when the hot-path result justifies it.