Runtime without upfront layer

I'd like to define atoms with service dependencies in a shared package and provide different layer implementations per platform (e.g. web vs native).

Currently Atom.runtime(layer) requires the layer at definition time, which couples the atom definitions to a specific implementation.

Would something like this make sense?

// core/runtime.ts - create runtime without layer
export const runtime = Atom.lazyRuntime<Api | Storage>()

// core/atoms.ts - real atoms with real dependencies, defined once
export const users = runtime.atom(Effect.flatMap(Api, a => a.getUsers()))
export const posts = runtime.atom(Effect.flatMap(Api, a => a.getPosts()))
export const userCount = Atom.make((get) => Result.map(get(users), u => u.length))

// web/main.tsx - provide layer
<RegistryProvider initialValues={[Atom.initialValue(runtime.layer, WebApiLayer)]}>

// native/main.tsx - different layer, same atoms
<RegistryProvider initialValues={[Atom.initialValue(runtime.layer, NativeApiLayer)]}>

Start by reading the existing Atom.runtime implementation and the runtime-related code in core/runtime.ts and core/atoms.ts. Trace how RegistryProvider consumes Atom.initialValue, then assess whether one shared atom definition can use different web and native layers. Done means the proposed lazy runtime behavior is clearly validated for both platform entry points.