Limitations & non-goals

Read this before you build something that assumes a capability BridgeKit doesn’t have yet. The line between built, deferred, and out of scope matters.

Built today

Production-validated, bidirectional, and cross-platform — exercised on a real device on both Android (Kotlin) and iOS (Swift):

• The Kotlin runtime — registry, router, epoch manager, park buffer, state store, ServiceLoader discovery.
• The Swift runtime — Router, StateStore, StreamHub, OutboundCaller, the Nitro C++/Swift seam, and BridgeKitRuntime.default. At parity with Android.
• The TypeScript runtime — registry, dispatcher, typed proxies, state mirrors, React hooks, marker contract hooks.
• The code generator for both targets — bridgekit generate --platform kotlin and --platform swift, both schema (t.*) and marker authoring styles, with hash parity, bridgekit.lock, drift --check, and --into.
• All four markers in both directions: Async / Void / Stream / State, native→JS and JS→native, including JS→native state push.
• Local-first resolution for pure-JS providers.
• The generic Nitro transport (the single Nitrogen-built component) on both platforms.
• Shared and feature-owned host contracts — a globally-provided host contract plus contracts owned and provided by a single feature at Scope.Feature(...). See migrating a feature.

Still deferred

Part of the design and the API is shaped to accommodate them, but not built today:

Area	Status
Web transport	Deferred. LoopbackTransport runs pure-JS providers in-process (web / test / standalone), but there is no formal web bridge yet. Local-first resolution already enables a pure-JS web target without a native side; a real transport is the remaining work.
Native binary payloads	t.binary() exists in the TS DSL with a base64 codec, but a native binary round-trip is not validated end-to-end (no native binary-specific codec; the wire protocol carries no blobs). Treat it as a JS-level, experimental capability.
Gradle-plugin discovery	Discovery is ServiceLoader-based on Android and explicit registration at app init on iOS; a Gradle plugin remains a non-goal.

Incremental adoption

BridgeKit is built to land in an existing app incrementally. You can move a feature onto a BridgeKit contract a few actions at a time and leave the rest on whatever bridge mechanism the app already uses — the two coexist at runtime, so a migration never has to be a big-bang rewrite. The migration guide walks through moving a feature one contract at a time.

Explicit non-goals

Stated outright in the design and still true:

• A web transport today (see above) and native binary payloads (t.binary() is JS-level / experimental, not a validated native round-trip).
• Full version negotiation — replaced by hash diffing + the additive-only rule.
• Multi-React-instance beyond instance scoping.

Hard transport limits

These are properties of the wire format, shared by both platforms — not roadmap items:

• No binary payloads in AnyMap (no blobs / ArrayBuffer; a top-level ArrayBuffer param is a possible future extension). AnyMap is map-only — scalars and arrays cross wrapped as { v: <value> }.
• No nested functions in payloads.
• No sync path native→JS — onInvoke is Promise-returning. querySync is native-provided only; JS cannot be called synchronously from native, and the provider side must be an in-memory lookup (dev asserts < 2ms).
• Stream backpressure is DROP_OLDEST, capacity 64 — the oldest element is silently dropped and counted in streamDrops.
• Readiness timeout is 5000 ms; the re-provide grace window is 1500 ms — both hardcoded.

Correctness notes worth knowing

A few state/epoch edge cases to keep in mind:

• A state observer may not receive the current value on subscription in every path — prefer the snapshot-on-connect hydration and useBridgeState’s status field.

How to read this page

None of the above blocks the core promise — typed bidirectional contracts on Android and iOS, with streams and state, validated on both. It’s here so you size your work against what’s real today versus what the design still reserves room for.