What data types cross the bridge, and their limits?

WHAT THIS TESTS: Knowing the serialization boundary of the bridge and the practical patterns for working around it when data is not JSON-friendly.

A GOOD ANSWER COVERS: The legacy bridge transmits only values that survive JSON serialization: strings, numbers, booleans, null, arrays, and maps with string keys. Anything stateful or non-serializable cannot cross directly, including JavaScript functions, native class instances, file handles, and raw binary buffers. Callbacks are a special case handled by the bridge assigning an id, not by passing the function itself. To move complex native data, you typically keep the object on the native side and pass an opaque identifier or token that JS uses in later calls, encode binary content as a base64 string, or use a URI the native layer can resolve. Under the new architecture, JSI host objects let JavaScript hold a direct reference to a native object without serialization.

COMMON WRONG ANSWERS: Saying you can pass function references or live class instances. Assuming large binary data is fine to send inline; serializing big blobs is slow and floods the bridge. Forgetting that map keys must be strings.

LIKELY FOLLOW-UPS: Why callbacks work despite functions not crossing, the performance cost of base64 over the bridge, how JSI changes this story, and the risk of leaking native handles referenced by stale ids.

ONE CONCRETE EXAMPLE: A native image-processing module produces a bitmap. Rather than serialize pixels over the bridge, it stores the bitmap natively, returns a string handle, and exposes operations keyed by that handle; JS passes the handle back for each step. Final output is delivered as a file URI the Image component can load.