fix: park early-arrival chunks for un-adopted channels (open/first-data race)

The connect side adopts a channel (installs local routing state) only
after the open-op response arrives, but the accept side's OpenHandler
can start pumping data the moment the channel opens — the two race and
the demux's lenient unknown-channel drop (REQ-CH-04) silently lost the
producer's first chunks (a TTY backend's banner, a sub protocol's
greeting).

route_payload now parks up to 64 payloads per unknown channel_id in a
bounded early-arrival buffer; adopt_channel drains them into the new
receiver in order. Beyond the cap the chunk drops with the existing
debug log + dropped_unknown_chunks counter (which now also counts
overflow). clear_all drops parked buffers with the connection.

Surfaced by alktty's consumer end-to-end test (review #001 L3): the
session never resolved because the producer's first chunks (stdout
sentinel + exit chunk for an immediately-resolving backend) arrived
before the adopt and were dropped. REQ-CH-04 wording updated by this
behavior; ADR-039 §demux loop describes the lenient drop for genuinely
unknown channels, which remains the case past the cap.

Verification: cargo test 597 (2 rewritten for the new semantics +
route_payload_to_unknown_channel_parks_until_adopt gate);
--all-features 614; clippy -D warnings clean; fmt clean; doc 0
warnings; publish --dry-run ok; standalone probe (handler-writes-first
e2e over one connection) shows 0 dropped chunks with the fix vs 1
without.

AGENTS.md

@@ -215,13 +215,14 @@ cargo semver-checks check-release             # before a release
 ## Architecture Context
 
 - `docs/architecture/` — the authoritative spec. Read it before
-  non-trivial changes. ADRs are numbered 001..047; OQs (open questions)
+  non-trivial changes. ADRs are numbered 001..048; OQs (open questions)
   track resolved/deferred decisions.
 - This crate unifies `alknet-call` and `alknet-channels` from the
   alknet mono-repo (`/workspace/@alkdev/alknet`). The source
   architecture docs were ported from
   `/workspace/@alkdev/alknet/docs/architecture/` and renumbered as
-  alkcall ADRs (001..047). The ALPN strings (`alk/call`,
+  alkcall ADRs (001..048; ADR-048 was authored in this crate). The ALPN
+  strings (`alk/call`,
   `alk/channels`) are wire-stable going forward (renamed from
   `alknet/` to `alk/` in v0.1.1, before the first published consumer).
 - Key ADRs that inform this crate's design:

CHANGELOG.md

@@ -4,6 +4,275 @@ All notable changes to this crate are documented here. The format is
 based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and
 this crate adheres to [Semantic Versioning](https://semver.org/).
 
+## [0.4.1] - 2026-09-05
+
+Bug-fix release: chunks arriving for a not-yet-adopted channel are
+parked instead of dropped (the open-op response / first-data race).
+No API changes.
+
+### Fixed
+
+- **Early-arrival chunks for un-adopted channels are parked, not
+  dropped.** The connect side adopts a channel (installs local routing
+  state) only after the open-op response arrives, but the accept side's
+  `OpenHandler` can start pumping data the moment the channel opens —
+  the two race. Previously the connect side's demux dropped those
+  chunks (REQ-CH-04's lenient unknown-channel drop), silently losing
+  the first chunks of any push-first producer (a TTY backend's banner
+  or greeting, a sub protocol's initial frame). `route_payload` now
+  parks up to 64 payloads per unknown `channel_id` in a bounded
+  early-arrival buffer and `adopt_channel` drains them into the new
+  receiver in order; chunks beyond the cap drop with the pre-existing
+  debug log and `dropped_unknown_chunks` counter (the counter now means
+  "early-arrival overflow or genuinely unknown channel", not just the
+  latter). `clear_all` drops the parked buffers with the connection.
+  Surfaced by alktty's consumer end-to-end test (review #001 L3): the
+  session never resolved because the producer's first chunks (the
+  stdout sentinel + exit chunk for an immediately-resolving backend)
+  arrived before the adopt and were dropped.
+
+## [0.4.0] - 2026-09-05
+
+`input_schema` is now enforced at call time (the advertise-vs-enforce
+leg ADR-016 promised: `INVALID_INPUT` is the registry's
+schema-mismatch error code). Minor bump — behavioral break for any
+caller that was passing schema-violating inputs to ops declaring a
+non-trivial `input_schema` and relying on the validation being
+documentation-only.
+
+### Changed
+
+- **`OperationSpec.input_schema` is enforced by the registry on every
+  dispatch.** Until now the schema was advertise-only: `services/schema`
+  disclosed it, the wire and gateway paths disclosed it, but no dispatch
+  entry point consulted it (the only schema alkcall actually enforced
+  was `publish_schema` per-chunk on Pub ops, P-03). The schema now
+  compiles once at registration (same fail-closed rule as
+  `publish_schema`/CF-003: an un-compilable schema is a registration
+  error, never a silently-skipped contract) and is checked after the
+  ACL gate in all three dispatch entry points — `invoke`,
+  `invoke_streaming`, and `invoke_sink` (via `resolve_sink_handler`,
+  preserving the P-08 single-source-of-truth property). Violations
+  return `CallError::invalid_input("input failed input_schema
+  validation")` with the input echoed in `details`. Raw-JSON-Schema
+  semantics (permissive on unknown keys) — adapters that want
+  closed-by-default enforcement keep doing their own hardening, as
+  alkhttp's `CompiledInputSchema` already does; the two checks compose.
+  Built-in ops are unaffected: alkcall's own specs declare the empty
+  object schema (accepts anything) and `services/schema` already
+  returned `INVALID_INPUT` for a missing `name` before its handler ran.
+  Motivated by alktty's code review #001 (L1): the channels open-op
+  wrapper hands the registry-checked `input` to the `OpenHandler` as
+  the authoritative params, which requires the registry to actually
+  validate it.
+- `OperationRegistry` gained an `input_validator` cache (compiled at
+  registration, mirrored on `fork` and in `OperationRegistryBuilder`
+  like the publish validators); `input_validator()` is `pub(crate)` —
+  the public surface is unchanged.
+
+## [0.3.1] - 2026-09-04
+
+Bug-fix release: `services/list-peers` can now list peer-announced
+ops. No API breaks — `OperationEnv` gains one defaulted trait method
+(non-breaking for all implementors; `cargo semver-checks` 196 checks
+pass against the 0.3.0 baseline).
+
+### Fixed
+
+- **`services/list-peers` shows each peer's operations**
+  (UP-03, surfaced by alkhttp's review 006). `PeerCompositeEnv`
+  overrode `peer_ids` only, so `peer_operations` fell to the trait
+  default (`Vec::new()`) and every peer listed with an empty
+  operations array — the ADR-022 amendment's "announced op is
+  discoverable via `services/list-peers`" promise never resolved on
+  the wire. ADR-030 prescribed the fix but it had never been ported
+  into alkcall; the existing `list-peers` unit tests passed because
+  they mock `peer_operations` with hand-rolled envs. Implemented per
+  ADR-030: `OperationEnv::list_operation_names` (default `Vec::new()`)
+  + overrides on `OverlayOperationEnv` (overlay map keys),
+  `PeerCompositeEnv` (`peer_operations` delegates to the peer's
+  overlay; its own aggregate mirrors `contains()`: session +
+  connections + base), `LocalOperationEnv` (registry names), and
+  `ChannelsSessionEnv` (delegates to base). Gate:
+  `announced_op_is_discoverable_via_services_list_peers` exercises the
+  exact `compose_root_env` shape (announce via `op/register`, then
+  assert both the `peer_operations` probe and the `services/list-peers`
+  wire shape attribute the op to the peer) — verified load-bearing
+  (reverting the override fails the gate). ADR-030 status is now
+  Accepted with the UP-03 provenance note.
+
+## [0.3.0] - 2026-09-04
+
+The connect side can serve (two-way ops over one channels connection),
+per-session fork registries, and the `op/register` bootstrap op — the
+remediation of reviews 004 and 005
+(`docs/reviews/004-*.md`, `docs/reviews/005-*.md`). Two
+`OperationRegistry` methods changed their return types from borrowed to
+owned (source-breaking for annotated call sites; minor bump per 0.x
+semver rules) — verified with `cargo semver-checks` (196 checks pass
+against the published 0.2.0 baseline; the return-type changes were
+caught by manual diff review, as the tool has no lint for that
+pattern).
+
+### Added
+
+- **`OperationRegistry::fork`** — deep-copies a registry (handlers,
+  provenance, composition authority, capabilities, cached
+  publish-schema validators) into an independently-mutable copy. The
+  per-session shape: fork the deployment's base registry, register the
+  session's ops on the fork, dispatch the session over the fork
+  (ADR-047 §4 amendment, 2026-09-03). Internally mutable, so a fork
+  shared as an `Arc` can receive registrations after the dispatcher
+  was built — `install_bootstrap_discovery` relies on this.
+- **`OperationRegistryBuilder::from_registry`** — seeds a builder from
+  an existing registry's registrations (the fork surface expressed
+  through the builder; registration order is not preserved — HashMap
+  iteration).
+- **Connect-side serving (two-way ops over one connection).** The
+  channels connect side's channel-0 read pump previously resolved
+  outbound responses only and silently dropped inbound
+  `call.requested` frames. New `ChannelClient::from_connection_with_serving`
+  takes `Option<ServingConfig>` (`registry` + `identity_provider`);
+  with `Some(..)` the read pump becomes the full-duplex serving loop
+  (`Dispatcher::serve_single_stream`), so a connected peer can call the
+  consumer's ops. `from_connection` keeps the resolution-only pump
+  (pure-consumer default). Serving is opt-in: the protocol is
+  symmetric, the API is explicit (ADR-022 amendment, 2026-09-03).
+- **`Dispatcher::serve_single_stream` and `Dispatcher::dispatch_start`
+  + `StartedDispatch`** — the shared serving-loop machinery. Both
+  single-stream loops (accept side and connect side) now run the same
+  concurrency model: Once invocations, Sub pumps, and sink response
+  writers are spawned, so same-connection nested composition resolves
+  concurrently instead of deadlocking until the 30 s sweeper (review
+  005 G-01). Handles are tracked and aborted at loop exit; no lock is
+  held across an await.
+- **`registry::op_register` module — the `op/register` bootstrap op**
+  (review 004 F-05). A connected peer announces the ops it serves over
+  the wire: `OP_REGISTER_NAME` (`op/register`), `OpRegisterRequest`
+  (wire round-trip via `to_json`/`from_json`), `op_register_spec` /
+  `op_register_handler`. Announcements land in the connection overlay
+  and are served through forwarding handlers, so hub→consumer import
+  (`from_call`) and consumer→hub announce are symmetric. **Collision
+  policy** (review 005 G-03): a peer-announced op may replace other
+  peer-announced ops (when `replace` is set) but never the serving
+  side's own registrations — a name present on the serving registry
+  rejects with the new `CallError::already_exists` (`ALREADY_EXISTS`,
+  non-retryable), regardless of `replace`. The collision gate checks
+  the serving registry, the session fork, and the connection overlay
+  itself.
+- **`install_bootstrap_discovery`** (review 004 F-06) — registers
+  `services/list`, `services/list-peers`, and `services/schema`
+  against a registry with handlers closed over that same `Arc`, so
+  per-session forks are the discovery source for their own openables
+  (handlers see every op the registry serves at call time). ACL
+  filtering stays per-caller. The bootstrap-op set on channel 0 is
+  closed and now includes `services/list-peers` (review 005 G-05 —
+  doc alignment; the code had installed it since the amendment landed)
+  plus `op/register` (ADR-022 amendment).
+- **`spec_to_json_pub`** — public serialization of an `OperationSpec`
+  into the `services/schema` wire shape (the shape `op/register`
+  announces with and `services/schema` serves). **`resource_id_path`
+  now survives the spec wire round-trip** (review 005 G-04): both the
+  serializer and the `op/register` parser carry the field.
+- **`CallConnection::overlay_contains` /
+  `CallConnection::overlay_registration`** — read access to the
+  connection overlay, for `op/register` replace semantics and
+  composition reachability checks.
+
+### Changed
+
+- **`OperationRegistry::registration` returns
+  `Option<HandlerRegistration>`** (was `Option<&HandlerRegistration>`)
+  and **`list_operations` returns `Vec<OperationSpec>`** (was
+  `Vec<&OperationSpec>`). Source-breaking for call sites that annotate
+  the borrowed types — clone-at-the-boundary instead. Minor bump per
+  0.x semver rules.
+- **Registration is `&self` throughout** — `OperationRegistry::register`,
+  `ChannelOperations::register_on`, and `ChannelCore::register_openable`
+  take `&OperationRegistry` (was `&mut`). Source-compatible for
+  callers (reborrow); this is what makes post-construction
+  registration on a forked, `Arc`-shared registry possible.
+- **`from_call` composition reachability** — the forwarding-handler
+  path declares reachable namespaces on the composing handler's
+  `scoped_env` (empty is deny-by-default) and the connection overlay
+  is attached only for identity-carrying connections (ADR-030 §5).
+
+## [0.2.0] - 2026-08-31
+
+Consumer-findings remediation (CF-001..004 from
+`docs/reviews/consumer-findings-ledger.md` — alkhttp as the first real
+consumer), the promoted `gateway` dispatch spine (ADR-048), and a
+docs-hygiene pass. One behavior change noted below; otherwise additive.
+
+### Added
+
+- **`gateway` feature: the transport-neutral dispatch spine**
+  (ADR-048). New `alkcall::gateway` module behind the opt-in `gateway`
+  cargo feature (default off; adds no dependencies). `GatewayDispatch`
+  is the deadline-bounded, re-rooted-context invoke spine over
+  `OperationRegistry` (`invoke` / `invoke_streaming` / `invoke_sink`)
+  promoted from alkhttp's gateway after it proved transport-agnostic —
+  hubs and spokes relaying calls (ADR-042 translate path) need the
+  identical root-context discipline (`internal: false`,
+  `forwarded_for: None`) without any HTTP. `schema_disclosure_denial`
+  is the shared is-internal + ACL check for `services/schema`
+  inner-op-name disclosure (one implementation so transports cannot
+  drift; ACL denial returns `FORBIDDEN`, Internal visibility returns
+  spec-404 — see ADR-048 for the split from the wire handler's
+  conservative spec-404). The handler deadline is a constructor knob
+  (`with_deadline`); the default remains 30 s. alkhttp migrates to
+  this module in a follow-up and drops its local copy.
+
+### Changed
+
+- **Un-compilable `publish_schema` values are rejected at registration
+  time** (CF-003). `OperationRegistry::register` and every
+  `OperationRegistryBuilder` method now return `Err` when a Pub op's
+  `publish_schema` fails to compile. Previously the bad schema was
+  accepted and the dispatch path failed **open** (warn log, chunks
+  unvalidated). The compiled validator is now cached per-op
+  (`OperationRegistry::publish_validator`) and consumed by the dispatch
+  path — the per-request compile is gone. Code that registered
+  un-compilable schemas will now get a registration error instead of a
+  silently-unvalidated op.
+
+### Fixed
+
+- **`services/schema` no longer discloses Internal or ACL-restricted op
+  specs** (CF-004). The schema handler now applies the same gates as
+  `invoke()` — Internal-visibility rejection and `AccessControl::check`
+  with matching identity resolution. Restricted ops return spec-404
+  (`NOT_FOUND`), consistent with "restricted ops don't exist" elsewhere;
+  no information leaks about the restricted surface. The gate is inside
+  the handler, so every transport is covered.
+- **Demux `TooLarge` skip no longer allocates from the peer's header**
+  (CF-002). The skip now streams through a fixed 64 KiB buffer, and a
+  cumulative 256 MiB skipped-bytes budget tears down connections that
+  loop oversized headers. Previously the buffer was sized from the
+  untrusted `u32` length (up to ~4 GiB pinned per dribbling peer).
+- **Write failures before request delivery are retryable** (CF-001).
+  New `CallError::connection_closed` (`CONNECTION_CLOSED`, `retryable:
+  true`) is returned when the `call.requested` frame write fails on any
+  consumer path (`call`/`subscribe`/`publish`, both stream modes) — the
+  call provably never reached the producer, so reconnect/retry is safe.
+  Mid-publish write failures and producer-side connection-teardown
+  failures remain non-retryable `INTERNAL` (delivery ambiguous). The
+  new code string is additive; the `retryable` flag is the
+  machine-readable signal for consumers.
+
+### Fixed (continued)
+
+- **Doc hygiene.** Resolved the broken intra-doc link on `lib.rs`'s
+  feature-gated `gateway` mention (rustdoc warned on default-feature
+  builds), and re-pointed 40 `src/` doc references from the old alknet
+  mono-repo ADR numbering (ADR-049/050/052/065/070/074/092) to this
+  crate's numbering (ADR-021/011/034/007/008/009/005) so published docs
+  reference ADRs this crate actually carries. Failure-path coverage was
+  added alongside: the CF-002 skipped-bytes teardown and skip-EOF arms,
+  CF-001's subscribe write-failure sites (both stream modes), the
+  publish request-frame failure in single-stream mode, and
+  non-retryability pinning for mid-publish failures.
+
 ## [0.1.1] - 2026-08-17
 
 A minor release that renames the ALPN prefix from `alknet/` to `alk/`
@@ -61,5 +330,8 @@ Vendored core types (`Connection`, `ProtocolHandler`, `BiStream`,
 (ADR-046), the channels protocol with openable-ALPNs-as-operations
 (ADR-047), and the `ChannelClient` transport-agnostic client.
 
+[0.3.1]: https://git.alk.dev/alkdev/alkcall/releases/tag/v0.3.1
+[0.3.0]: https://git.alk.dev/alkdev/alkcall/releases/tag/v0.3.0
+[0.2.0]: https://git.alk.dev/alkdev/alkcall/releases/tag/v0.2.0
 [0.1.1]: https://git.alk.dev/alkdev/alkcall/releases/tag/v0.1.1
 [0.1.0]: https://git.alk.dev/alkdev/alkcall/releases/tag/v0.1.0

Cargo.lock

@@ -27,7 +27,7 @@ dependencies = [
 
 [[package]]
 name = "alkcall"
-version = "0.1.1"
+version = "0.4.1"
 dependencies = [
  "async-trait",
  "bytes",

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "alkcall"
-version = "0.1.1"
+version = "0.4.1"
 edition = "2021"
 rust-version = "1.85"
 license = "MIT OR Apache-2.0"
@@ -9,13 +9,14 @@ readme = "README.md"
 repository = "https://git.alk.dev/alkdev/alkcall"
 keywords = ["rpc", "json-rpc", "multiplexing", "wire-format", "alpn"]
 categories = ["network-programming", "asynchronous", "encoding"]
-exclude = [".opencode/", "AGENTS.md", "docs/reviews/", "docs/sdd_process.md", "Cargo.lock"]
+exclude = [".opencode/", "AGENTS.md", "docs/reviews/", "docs/sdd_process.md"]
 
 [lib]
 name = "alkcall"
 
 [features]
 default = []
+gateway = []
 
 [dependencies]
 jsonschema = { version = "0.46", default-features = false }

README.md

@@ -21,7 +21,7 @@ use alkcall::registry::{
     spec::{OperationSpec, OperationType, Visibility, AccessControl},
 };
 
-let mut registry = OperationRegistry::new();
+let registry = OperationRegistry::new();
 registry.register(HandlerRegistration::new(
     OperationSpec::new(
         "echo/run",
@@ -98,6 +98,34 @@ for reg in registrations {
 let response = conn.call("remote/status", serde_json::json!({})).await;
 ```
 
+### Serving your own ops as a connected consumer
+
+The call protocol is symmetric — both sides of a connection can serve
+ops. A `ChannelClient` built with `from_connection` is a pure consumer
+(inbound `call.requested` frames are dropped); pass a `ServingConfig`
+to also serve your registry to the peer, and use `op/register` to
+announce which ops you serve:
+
+```rust
+use std::sync::Arc;
+use alkcall::channels::client::{ChannelClient, ServingConfig};
+use alkcall::registry::discovery::install_bootstrap_discovery;
+
+let registry = Arc::new(OperationRegistry::new());
+// ... register your ops on the registry, then:
+install_bootstrap_discovery(&registry)?;
+
+let client = ChannelClient::from_connection_with_serving(
+    connection,
+    Some(ServingConfig {
+        registry: Arc::clone(&registry),
+        identity_provider: provider,
+    }),
+).await?;
+// peer-callable ops resolve against `registry` on channel 0;
+// `client.call_open_op` still works — both directions share the pump
+```
+
 ## Architecture
 
 alkcall is a pure protocol crate — no networking, no transport
@@ -107,7 +135,7 @@ Downstream crates compose on top of it in a layered dependency chain.
 | Role | Call protocol | Channels protocol |
 |------|---------------|-------------------|
 | **Producer** | Registers ops on an `OperationRegistry`, runs a `Dispatcher` | Runs a `ChannelsAdapter`, registers openable ALPNs via `ChannelCore::register_openable` |
-| **Consumer** | Uses `CallConnection` to call ops, uses `from_call` to discover/import remote ops | Uses `ChannelClient` to open channels via `call_open_op` + `open_channel` |
+| **Consumer** | Uses `CallConnection` to call ops, uses `from_call` to discover/import remote ops; may also serve its own ops (`from_connection_with_serving`) | Uses `ChannelClient` to open channels via `call_open_op` + `open_channel` |
 | **Hub** | Both: runs a `Dispatcher` for ops it produces, holds `CallConnection`s to spokes for ops it consumes | Both: runs a `ChannelsAdapter` for inbound connections, holds `ChannelClient`s to spokes |
 | **Spoke / Worker** | Both: produces ops (its own services), consumes hub ops | Both: produces channels (TTY, tunnel), may consume hub channels |
 

docs/architecture/README.md

@@ -100,6 +100,7 @@ are wire-stable and unchanged — see ADR-004.
 | [045](decisions/045-alknetclient-native-dial-seam.md) | AlknetClient Dial Seam | spawn_dispatch / from_connection take-over; dial in consumer |
 | [046](decisions/046-publish-operation-type-and-handler-kind-sink.md) | Publish Operation Type and HandlerKind::Sink | `OperationType::Pub` (producer→consumer streaming); `SinkHandler` + `HandlerKind::Sink`; `call.published` wire event; `invoke_sink()` dispatch; `Subscription` renamed to `Sub` |
 | [047](decisions/047-openable-alpns-are-operations.md) | Openable ALPNs Are Operations | `channel/open` dissolves into per-ALPN ops `channels/<alpn>/sub`/`pub`; `channel_open` marker on `OperationSpec`; `ChannelCore` wrapper; extension-trait `ChannelOperationEnv`; connection-owner allocates `channel_id`; opener ledger (Gap 2 fix); ALPNs are call apps |
+| [048](decisions/048-dispatch-spine-gateway-module.md) | Dispatch Spine (feature-gated `gateway` module) | `alkcall::gateway` behind the `gateway` feature; `GatewayDispatch` invoke spine (deadline knob, re-rooted context) + `schema_disclosure_denial` (FORBIDDEN for ACL deny, spec-404 for Internal); promoted from alkhttp for hub/spoke reuse |
 
 ## Relevant Open Questions
 

docs/architecture/decisions/022-call-protocol-client-and-adapter-contract.md

@@ -2,7 +2,7 @@
 
 ## Status
 
-Accepted (amended 2026-06-26, 2026-07-13, and 2026-07-16 — see "Amendments" below; the 2026-07-16 amendment per ADR-045 §5 removes `CallClient::connect`)
+Accepted (amended 2026-06-26, 2026-07-13, and 2026-07-16 — see "Amendments" below; the 2026-07-16 amendment per ADR-045 §5 removes `CallClient::connect`; amendment 2026-09-03 — the bootstrap-op set and the connect-side serving loop, see "Amendment (2026-09-03)" below; amendment 2026-09-04 — the `op/register` collision policy, in that amendment's "Collision policy" paragraph)
 
 ## Context
 
@@ -359,6 +359,123 @@ same as `from_openapi` receives HTTP credentials.
   prior art
 - POC at `/workspace/@alkdev/dispatch` — head/worker dispatch over SSH+axum
 
+## Amendment (2026-09-03): bootstrap-op set and the connect-side serving loop
+
+Review 004 (F-04/F-05) verified two gaps between this ADR's
+bidirectionality promise (§2 "connection direction is independent of
+call direction") and the single-stream channel-0 implementation:
+
+1. **The connect side's read pump resolved responses only** — inbound
+   `call.requested` frames were silently dropped. Serving existed only
+   on the accept side (`run_loop_single_stream`).
+2. **No wire mechanism announced client-side ops** — `from_call`
+   imports hub ops into the consumer, but a connected peer had no way
+   to announce "here are the ops I serve" over the wire.
+
+The amendment (implemented in alkcall; the e2e gates are
+`serving_loop_hub_to_consumer_call_resolves` and
+`op_register_announce_then_hub_call_routes_back_to_consumer` in
+`src/channels/client.rs`):
+
+### The bootstrap-op set (one-way)
+
+Each side of a channels connection **may serve** the bootstrap ops on
+channel 0. The set is closed:
+
+- `services/list` — discovery (the op `from_call` dials on every
+  import; each side is expected to serve it)
+- `services/schema` — per-op schema disclosure
+- `services/list-peers` — peer-keyed discovery (the op that makes
+  peer-announced ops discoverable; `from_call`-import discovery
+  relies on it). Added to this list 2026-09-04 (review 005 G-05) —
+  `install_bootstrap_discovery` has installed it since the amendment
+  landed; the set was declared closed and the doc lagged the code.
+- `op/register` — peer op announcement (below)
+
+These are the ops a peer may assume are reachable (subject to each
+op's `AccessControl`); a peer that does not serve one answers
+`NOT_FOUND` and the caller treats it accordingly (`from_call` already
+surfaces discovery failure as `AdapterError::DiscoveryFailed`).
+
+### Connect-side serving is opt-in (two-way door at the API level)
+
+`ChannelClient::from_connection_with_serving(connection,
+Option<ServingConfig>)` — with `None` (the pure-consumer default) the
+read pump resolves outbound pendings only (previous behavior). With
+`Some(ServingConfig { registry, identity_provider })` the read pump
+becomes the full-duplex serving loop (`Dispatcher::serve_single_stream`):
+inbound `call.requested` frames dispatch against the configured
+registry and resolve back to the peer; outbound pendings still resolve
+in the same loop. Serving is opt-in because a pure consumer has no
+registry to serve; the *protocol* is symmetric, the *API* is explicit.
+
+Direction disambiguation in the loop is by table membership, not
+framing: an id that is one of *our* outbound pendings resolves there;
+an id that is an inbound request dispatches; `call.aborted` tries both
+tables (in-flight sink aborts and the pending map's cascade). IDs are
+UUID-generated per side, so cross-correlation is not a hazard.
+
+### `op/register` (one-way in wire shape)
+
+The peer→hub direction of `from_call`'s bundle flow: a peer sends
+`call.requested` for `op/register` carrying the `OperationSpec` in the
+`services/schema` wire shape (`spec_to_json`) plus a `replace` flag.
+The serving-side handler (`registry::op_register::op_register_handler`):
+
+1. Rebuilds the spec (the same parser `from_call` uses — one spec
+   serialization on the wire).
+2. Wraps a **call-forwarding handler** that issues a nested
+   `call.requested` back over channel 0 to the announcing peer (the
+   same shape `from_call`'s imported bundles use — the in-process twin
+   at `protocol/adapter.rs`).
+3. Writes the bundle into **that connection's overlay** via
+   `register_imported` with `FromCall` provenance and
+   `Visibility::Internal` (composition material, ADR-017 — never
+   directly callable from the serving side's own wire).
+
+The announced op is discoverable via `services/list-peers` (the
+overlay is peer-keyed, `compose_root_env` attaches it) and invocable
+via nested composition (`env.invoke`). Announced Sub/Pub ops register
+as stubs that answer `INVALID_OPERATION_TYPE` — nested composition is
+request/response-only (`OverlayOperationEnv`'s contract); the
+streaming/sink forwarding shapes ride on the `from_call` import path.
+
+Access control: `op/register` itself carries an `AccessControl` (an
+unprivileged peer cannot reach the handler — the registry's normal
+invoke path enforces it). Replace semantics: a collision with an
+existing overlay registration is rejected with `ALREADY_EXISTS`
+unless `replace: true` (the reconnect path re-announces). The
+overlay dies with the connection (Layer 2), so reconnect re-announce
+is naturally scoped.
+
+Collision policy (amended 2026-09-04, review 005 G-03): a
+peer-announced op may collide with other *peer-announced* ops on the
+same connection (`replace` governs) but **never** with the serving
+side's own registrations — a name present on the serving registry
+rejects with `ALREADY_EXISTS` regardless of `replace`. The connection
+overlay shadows the base registry in `PeerCompositeEnv` (connections
+resolve before base, ADR-024 §1), so an unscreened same-name announce
+would silently rewrite what a wire-dispatched handler's
+`ctx.env.invoke` resolves for any name the deployment registered:
+composition authority (ADR-018) belongs to the composing handler's
+deployer, not the connected peer. Visibility is irrelevant to this
+gate (`Internal` ops are as shadowable as `External` —
+`OverlayOperationEnv` gates on `AccessControl`, not visibility; the
+composed child is `internal: true` by design).
+
+The envelope kind set stays closed at six — bootstrap ops over channel
+0 are the door (AGENTS.md §7 allows adding kinds; none is needed).
+
+### Cross-references
+
+- Review 004 (`docs/reviews/004-per-connection-dispatch-and-client-serving-review.md`)
+  F-04/F-05 — the verification and the design decision
+- ADR-047 §4 amendment #2 (2026-09-03) — the per-session fork the
+  bootstrap ops compose on
+- alkhttp OQ-05 / ADR-048 — the browser data-channel wiring this
+  unblocks (the alkhttp-side gap was wiring; these two mechanisms are
+  what it wires to)
+
 ## Amendments (2026-06-26)
 
 This ADR left four decisions as two-way doors (§1 Consequences flagged DC-1's

docs/architecture/decisions/030-peer-composite-env-peer-operations.md

@@ -2,7 +2,15 @@
 
 ## Status
 
-Proposed
+Accepted (implemented 2026-09-04 — surfaced as UP-03 in alkhttp's
+review 006 `docs/…/006-alkcall-0.3.0-consequence-review.md`: the
+`op/register` amendment's "announced op is discoverable via
+`services/list-peers`" promise did not resolve on the wire because
+this override had never been ported into alkcall; the gate is
+`announced_op_is_discoverable_via_services_list_peers` in
+`src/registry/op_register.rs`. The `services/list-peers` unit tests
+did not catch it because they mock `peer_operations` with hand-rolled
+envs.)
 
 ## Context
 

docs/architecture/decisions/047-openable-alpns-are-operations.md

@@ -5,7 +5,99 @@
 Accepted (amends ADR-037; refines ADR-044, ADR-046; §4 amended
 2026-08-13 — open ops are registered per-connection, not resolved via
 `context.env` downcast — see "Amendment (§4 per-connection
-registration, 2026-08-13)" below)
+registration, 2026-08-13)"; amendment #2 (2026-09-03) — the
+per-connection registration mechanism is the **per-session fork of the
+base registry installed as the session's dispatch registry**, not the
+connection overlay — see "Amendment (§4 mechanism, 2026-09-03)" below)
+
+## Amendment (§4 mechanism, 2026-09-03)
+
+The 2026-08-13 amendment named the registration target as "the
+connection overlay registry (Layer 2 per ADR-019)". Review 004 (F-01)
+verified that this shape cannot dispatch: the top-level dispatch path
+(`Dispatcher::dispatch` / `run_loop_single_stream`) resolves and
+invokes against the dispatcher's **base registry only**; the
+connection overlay is reachable solely as a layer of `context.env`
+(for nested invocations — a handler calling `env.invoke(...)), never
+for resolving the incoming `call.requested` itself. An open op
+registered on the overlay resolves `NOT_FOUND` on the wire.
+
+The one shape proven end-to-end (alkcall's own e2e gate) is different:
+the `install_channel_zero` hook builds a **fresh per-connection
+registry containing the open op and passes it as the dispatcher's base
+registry**. This amendment makes that the operative mechanism.
+
+**The decision: per-connection registration happens on a fork of the
+deployment's base registry, installed as the session's dispatch
+registry.** The `install_channel_zero` hook (and any future
+session-establishment seam):
+
+1. **Forks** the deployment's base registry
+   (`OperationRegistry::fork` — a deep copy carrying handlers,
+   provenance, composition authority, capabilities, and the cached
+   publish-schema validators; review 004 F-02/F-03).
+2. **Registers the per-session ops on the fork** — the generic channel
+   ops (`ChannelOperations::register_on`), the openables
+   (`ChannelCore::register_openable`), and the bootstrap discovery ops
+   (`install_bootstrap_discovery`, closed over the fork itself so
+   `services/list` sees the fork's per-session ops — review 004 F-06).
+3. **Dispatches channel 0 over the fork** (`Dispatcher::new(fork,
+   ...)`).
+
+The fork is possible because `OperationRegistry` is internally
+mutable (`parking_lot::RwLock` around both maps) — a fork shared as an
+`Arc<OperationRegistry>` can receive bootstrap ops after the
+dispatcher was built, and the self-referential discovery closure sees
+every post-install registration.
+
+The **connection overlay (Layer 2) remains what ADR-019/ADR-024
+describe**: the landing zone for peer-announced ops (`op/register`,
+review 004 F-05 — ADR-022 amendment) and the nested-invocation target
+for imported ops. It is not the dispatch-resolution path for the
+session's own ops.
+
+Rationale for the fork shape over an overlay-aware dispatch fallback
+(F-02 option (b)): the fork is the only shape with an end-to-end
+proof, it needs no change to the shared dispatch loop, and it keeps
+the overlay's `invoke_with_policy` shape (namespace-scoped,
+parent-context-driven — built for nested composition) out of the
+top-level call path, where it does not match the frame-handling
+contract.
+
+This preserves every invariant the 2026-08-13 amendment protected:
+
+- **Layering (ADR-044):** unchanged — the open-op wrapper is in
+  `channels-call`; the call crate's `OperationRegistry` gains only
+  `fork` (and interior mutability), no channels types.
+- **Per-connection resolution:** the open op gets the *right*
+  `ChannelManager` because the fork is built per-connection and its
+  openable closes over that connection's `ChannelCore`.
+- **"Marked ops invoked outside a channels session" (ADR-047 §2):**
+  unchanged in effect — a `channels/<alpn>/sub` op registered only on
+  a session fork is not reachable on a bare `alk/call` connection (the
+  fork isn't that session's dispatch registry) — the dispatch path
+  returns `NOT_FOUND`.
+
+### Door type
+
+**Two-way (implementation detail), as before.** The registration
+*target* mechanism (fork as base registry) sits within the same
+wrapper-shape detail the 2026-08-13 amendment already marked two-way.
+The one-way decisions (per-ALPN op names, the `channel_open` marker,
+removal of `channel/open`/`direction`) are unchanged.
+
+### References
+
+- Review 004 F-01/F-02/F-03/F-06
+  (`docs/reviews/004-per-connection-dispatch-and-client-serving-review.md`)
+  — the verification and the mechanism decision
+- ADR-022 amendment (2026-09-03) — the bootstrap-op set (`services/list`,
+  `services/schema`, `op/register`) and the connect-side serving loop
+- ADR-019: operation registry layering (the overlay stays the nested
+  invocation / peer-announced-ops landing zone)
+- The e2e gate: `fork_registry_open_op_resolves_and_is_discoverable`
+  (`src/channels/client.rs`) — open op resolves through the fork,
+  per-session openable in `services/list`, `services/schema` validates
 
 ## Amendment (§4 per-connection registration, 2026-08-13)
 

docs/architecture/decisions/048-dispatch-spine-gateway-module.md

@@ -0,0 +1,151 @@
+# ADR-048: Dispatch Spine (feature-gated `gateway` module)
+
+## Status
+
+Accepted
+
+## Context
+
+The first real downstream consumer of this crate — alkhttp — built its
+HTTP gateway (`POST /call`, `/subscribe`, `/publish`, the MCP `call`
+tool, the to_openapi projections) on a small internal component it
+calls the **dispatch spine**: a thin struct over
+`Arc<OperationRegistry>` that owns the *non-HTTP* half of gateway
+dispatch. The HTTP layer resolves bearer tokens to an `Identity`,
+frames NDJSON/SSE, maps `CallError` to HTTP statuses, and wraps axum
+handlers; the spine does everything that happens after that:
+
+- constructs the root `OperationContext` identically for every
+  transport (`internal: false` — ACL runs against the caller's
+  identity, not a handler's composition authority; `forwarded_for:
+  None` — wire-ingress only; identity supplied per-call),
+- resolves the registration's `composition_authority` /
+  `capabilities` / `scoped_env` into that context,
+- bounds Once-ops and sink dispatch with a deadline while leaving
+  streaming subscriptions unbounded (ADR-021: subscriptions are
+  long-lived),
+- and applies the `services/schema` disclosure guard when the
+  dispatched operation's *input* names another operation.
+
+This is transport-neutral work. It contains no HTTP concepts: no
+statuses, no headers, no body framing. And it is not HTTP-shaped by
+accident — the same shape is exactly what a **hub** needs when it
+terminates channel 0 on both legs and relays calls to spokes
+(ADR-042's "translate, not forward" rule): the hub must re-root the
+context at itself (its own identity, `internal: false`, no
+`forwarded_for`), re-resolve its capabilities for the outbound leg, and
+bound the relay so a hung spoke does not wedge the browser-facing
+connection. alkhttp needed it; the hub relay needs the same thing;
+any protocol crate that exposes a call surface to a less-trusted
+in-transport caller (a WS-native relay, a CLI bridge, a test harness)
+will need it again.
+
+Duplicating it per consumer is the failure mode this crate already
+paid for once: the spine's `services/schema` guard existed because the
+registry handler and the HTTP route were written against different
+disclosure rules (CF-004, filed from alkhttp's consumer review). One
+shared implementation is the fix; a second copy in a second crate
+would re-open the drift.
+
+ alkhttp is not yet published. This is the cheapest moment to move the
+component into alkcall (behind a feature, so the base crate stays lean
+and the surface is opt-in) and have alkhttp consume it rather than
+carry its own copy.
+
+## Decision
+
+Promote the dispatch spine into alkcall as a new `gateway` module,
+**feature-gated**:
+
+- `alkcall/gateway` behind the `gateway` cargo feature (default off —
+  the base crate stays lean; the module adds no dependencies, the gate
+  exists to keep the audit surface explicit and opt-in).
+- `GatewayDispatch` — the spine struct: `invoke()`,
+  `invoke_streaming()`, `invoke_sink()`, registry access, and root
+  context construction. The 30 s deadline alkhttp hardcodes becomes a
+  constructor knob (`with_deadline`) so consumers keep their own
+  policy; the default remains 30 s for drop-in equivalence.
+- `schema_disclosure_denial()` — the shared check that a spec the
+  caller could not invoke is not disclosed: `NOT_FOUND` for
+  Internal-visibility ops, **`FORBIDDEN` for ACL-denied ops**.
+- `CallRequest` stays in alkhttp (payload framing is transport
+  business); `MAX_BATCH_OPERATIONS` stays in alkhttp (batch is a
+  projection concern).
+
+### ACL denial returns FORBIDDEN, not spec-404
+
+The wire-path `services/schema` handler (CF-004, discovery.rs) returns
+spec-404 for both Internal visibility and ACL denial — the right
+answer for an unauthenticated wire caller, where even acknowledging
+the op's existence is a leak vector. The spine's guard is invoked by a
+transport that has *already* resolved the caller's identity and often
+already admitted the op elsewhere (the alkhttp GET `/schema` route
+answers `403` for ACL-denied ops, deliberately). The spine therefore
+keeps alkhttp's split:
+
+- Internal visibility → `NOT_FOUND` (never acknowledged, at any
+  authority).
+- ACL denial → `FORBIDDEN` (informative for a caller whose identity
+  the transport resolved; alkcall's own registry `invoke()` produces
+  the same code for the same caller state, so the guard's answer is
+  never *more* restrictive or *less* restrictive than the invoke that
+  would follow it).
+
+The two layers are consistent by construction: the wire handler's
+spec-404 is the conservative outer bound, the spine's FORBIDDEN is the
+identity-aware refinement. When an unauthenticated caller hits both,
+`FORBIDDEN` and `NOT_FOUND` differ only in which exists — and
+`AccessControl::check` with no identity returns
+`"authentication required"`, which HTTP-side mappers translate to 401.
+ alkhttp consumes this module in a later session and drops its local
+copy; until then the two implementations coexist (byte-identical in
+behavior, one in each crate).
+
+### What the spine deliberately does NOT include
+
+- **No HTTP mapping.** `CallError` → status/body is the consumer's
+  (alkhttp `gateway::error`). The neutral wire error is the module's
+  lowest-level vocabulary.
+- **No body framing, limits, or timeouts beyond the deadline knob.**
+  NDJSON/SSE framing, body caps, keep-alive intervals are transport
+  concerns (GW-15/GW-16 in alkhttp's ledger).
+- **No batch semantics.** `MAX_BATCH_OPERATIONS` and the batch
+  envelope are projections of the registry onto HTTP/MCP payloads.
+- **No `CallRequest` type.** The spine takes `(op, input)` — parsing
+  `{ operation, input }` is transport framing.
+
+## Consequences
+
+- The deadline policy moves from "alkhttp's 30 s HTTP convention" to
+  "spine policy, configured per-consumer" (default 30 s). Sink
+  dispatch is bounded by the same deadline as Once-ops — the wire
+  path's Pub dispatch is unbounded (`deadline: None`); consumers who
+  want the wire behavior pass `Duration::ZERO`-style no-deadline
+  configuration via `with_deadline(None)`. Documented divergence, a
+  choice the *consumer* now owns.
+- alkhttp migrates to `alkcall::gateway` in a follow-up session; its
+  local spine is deleted then, not deprecated in place (unpublished
+  crate — no compat window needed).
+- The feature adds no dependencies; `cargo test --no-default-features`
+  and `--all-features` both pass (the wasm-clean baseline is
+  untouched).
+- Future transports (WS-native relays, protocol crates) reuse the
+  spine instead of re-deriving the context/guard discipline.
+
+## Alternatives considered
+
+- **Fold into `registry` as a method set on `OperationRegistry`.**
+  Rejected: the spine is a *policy* wrapper (deadline, root-context
+  shape, disclosure guard) a consumer opts into, not the registry's
+  dispatch core. `GatewayDispatch::invoke` ≠ `OperationRegistry::invoke`
+  — conflating them invites wire paths to pick up transport-shaped
+  policy accidentally.
+- **Wait for alkhttp to publish first.** Rejected: promotes a
+  duplicate into the wild that immediately has to be deprecated; the
+  cheapest moment to move the type is before either crate's first
+  release.
+- **Promote the whole gateway module.** Rejected: routes (SSE/NDJSON
+  framing, body limits) and error mapping (`IntoResponse`, status
+  mapping) are HTTP by definition; moving them would drag axum/hyper
+  as optional dependencies into a protocol crate for no reuse —
+  alkhttp is their only consumer.

docs/reviews/004-per-connection-dispatch-and-client-serving-review.md

@@ -0,0 +1,527 @@
+# Review 004 — Per-Connection Dispatch Resolution and Client-Side Op Serving (found via alkhttp's WS data-channel drill-down)
+
+## Status
+
+Verified, remediated (2026-09-03 — Units 1–3; Unit 4 remains
+downstream in alkhttp). See "Remediation log (2026-09-03)" at the
+bottom for the landing summary and the gates.
+
+## Scope
+
+Focused review of two design/spec mismatches in the call + channels
+integration, uncovered while drilling into the deferred half of
+alkhttp's WebSocket path (alkhttp review 003, findings WS-24/WS-25 —
+the OQ-05 "browser data channels" deferral). The drill-down
+established that the alkhttp-side gap is mostly wiring, but two
+assumptions underneath it resolve to *this* crate:
+
+1. **Dispatch resolution for per-connection operations.** ADR-047's
+   §4 amendment (2026-08-13) decided that openable-ALPN open ops are
+   registered per-connection on the connection overlay registry
+   (Layer 2). The top-level dispatch path never consults that layer,
+   so an op registered per the amendment's mechanism cannot be
+   invoked over the wire. The only shape proven to dispatch is
+   different from the one the ADR describes.
+2. **Client-side operation serving.** The connection-local overlay
+   (`CallConnection::register_imported` → `overlay_env()`) and the
+   session overlay (`compose_root_env` attaching peer-keyed overlays)
+   are fully built and tested — but only the *accept side* has a loop
+   that serves inbound `call.requested` frames. The connect side's
+   read pump resolves responses and silently drops requests. There
+   is also no wire mechanism by which a client could announce an op
+   in the first place. Together these leave the "both sides can be
+   both" symmetry (AGENTS.md §8; ADR-022 §direction semantics;
+   alkhttp ADR-048's browser bidirectionality) unreachable on the
+   single-stream channel-0 shape.
+
+The pass re-verified every claim directly in source at tree
+`c16b069`. Cross-repo context: these findings block alkhttp review
+003's Unit 2 (data-channel wiring for WS sessions, both browser and
+native-fallback consumers); the decision work and the fixes live
+here. Findings continue alkcall review numbering (F-01..; reviews
+001–003 used P/C/R/A/B/C/D prefixes — F is the new prefix for this
+focused review, findings numbered independently).
+
+## Baseline verification (this pass)
+
+```
+cargo test                                    → 565 passed, 0 failed
+cargo clippy --all-targets -- -D warnings     → clean
+cargo fmt --check                             → clean
+```
+
+The suite is green; these findings are design/spec mismatches, not
+regressions. Notably, one of them (F-04's serving gap) is invisible
+to the suite *because* the connect side has never been asked to
+serve — the existing e2e tests all put the dispatcher on the accept
+side.
+
+## Verdict
+
+- **The channel machinery is done and proven.** `ChannelCore` /
+  `register_openable` / `ChannelOperations`, the opener ledger, the
+  odd/even id split, the demux hardening, and the full e2e wiring
+  test (`channels/client.rs:767-870`) are all in place and tested.
+  Nothing in Part A below re-opens that machinery.
+- **The gap is the dispatch-resolution *mechanism* (F-01/F-02/F-03)
+  and the connect-side serving half (F-04)** — both upstream of any
+  transport (TCP+TLS, WS, whatever). Fixing them here unblocks
+  alkhttp's data-channel wiring, and — since the crate is
+  wasm-targetable and not yet published to consumers — the fixes are
+  cheap now and required regardless of alkhttp.
+- **The recommended resolution is assembly, not invention:** per-
+  session registry forks (F-02/F-03) + a bootstrap `op/register` op
+  (F-01's mechanism choice) + a connect-side serving loop (F-04).
+  Three of the four pieces have working reference shapes in-tree;
+  the genuinely new protocol surface is one op handler and one read
+  loop.
+
+## Severity legend
+
+- **[critical]** — a decided spec invariant is violated in a way that
+  makes a promised capability unreachable end-to-end; or corrupts
+  data.
+- **[major]** — a core protocol path cannot serve a decided behavior;
+  works only via shapes the spec does not describe.
+- **[minor]** — drift, doc/spec inconsistency, or a missing
+  convenience with no correctness impact.
+
+---
+
+# Part A — Dispatch resolution vs the ADR-047 §4 amendment
+
+## F-01 [major] — Top-level dispatch never consults the connection overlay; ops registered per the ADR-047 §4 amendment mechanism resolve `NOT_FOUND` on the wire
+
+**ADR drift:** ADR-047 "Amendment (§4 per-connection registration,
+2026-08-13)": *"open ops are registered per-connection …
+`register_openable` is called on the connection overlay registry
+(Layer 2 per ADR-019), closing over the per-connection `ChannelCore`."*
+**Verified:** YES, three ways:
+
+1. `Dispatcher::dispatch` reads the operation's `op_type` from
+   `self.registry.registration(...)` only
+   (`src/protocol/dispatch.rs:316-320`) and invokes via
+   `self.registry.invoke` / `invoke_streaming` / `resolve_sink_handler`
+   (`:330-357`) — the dispatcher's **base registry** only. There is
+   no fallback to `connection.overlay_env()`.
+2. The connection overlay is reachable solely as a layer of
+   `context.env`: `compose_root_env` attaches it via
+   `env.attach_peer(peer_id, connection.overlay_env())`
+   (`dispatch.rs:195-215`). `context.env` is consulted for *nested*
+   invocations (a handler calling `context.env.invoke(...)`) — not
+   for resolving the incoming `call.requested` itself. An open op
+   registered on the overlay resolves `NOT_FOUND` when the wire
+   request arrives.
+3. The one place this wiring is proven end-to-end uses a different
+   shape than the amendment describes: alkcall's own e2e test
+   (`src/channels/client.rs:767-870`) has the `install_channel_zero`
+   hook build a **fresh per-connection registry containing the open
+   op and pass it as the dispatcher's base registry**
+   (`Dispatcher::new(registry, ...)` at `:837`) — not as an overlay.
+   That shape works; the amendment's shape cannot.
+
+Consequence: ADR-047's data-plane promise (per-connection open ops
+invocable on channel 0) is unreachable through the mechanism the
+amendment names. Every consumer of `register_openable` must either
+adopt the undocumented per-session-base-registry shape or fail.
+
+**Fix (Unit 1):** decide the mechanism deliberately (see F-02) and
+amend ADR-047 §4's wording to describe the mechanism that actually
+dispatches. The amendment's *rationale* (per-connection state, no
+`context.env` downcast) stands under either option.
+
+## F-02 [major] — `OperationRegistry` has no fork/clone surface; the proven per-session-registry shape requires one
+
+**Verified:** YES. `OperationRegistry`
+(`src/registry/registration.rs:97-100`) is
+`HashMap<String, HandlerRegistration>` + a validators map, with
+`new`/`register`/`registration`/`publish_validator`/`list_operations`
+(`:102-180`) — no `Clone` impl and no fork API. Every type inside a
+`HandlerRegistration` *is* clone-able (verified):
+
+- `HandlerKind` — `#[derive(Clone)]` (`registration.rs:51`)
+- `OperationSpec` — `#[derive(Debug, Clone, PartialEq)]`
+  (`spec.rs:174-175`)
+- `OperationProvenance` — `#[derive(Debug, Clone, Copy, PartialEq,
+  Eq)]` (`registration.rs:58-59`)
+- `CompositionAuthority` — `#[derive(Debug, Clone)]`
+  (`context.rs:82-83`)
+- `Capabilities` — manual `impl Clone` (`core/types.rs:75-80`)
+- `jsonschema::Validator` — `#[derive(Clone, Debug)]`
+  (jsonschema 0.46.10 `validator.rs:294`)
+
+So the fork is a small addition (derive `Clone` on
+`HandlerRegistration` + `OperationRegistry`, or an explicit
+`fork() -> OperationRegistry` that copies both maps). Two shape
+options:
+
+- **(a) per-session base registry (recommended).** The
+  `install_channel_zero` hook (and any future session-establishment
+  seam) forks the deployment's base registry, registers the
+  per-connection operations on the fork (generic channel ops, the
+  openable ALPNs, and — with F-05 — the bootstrap op), and dispatches
+  channel 0 over the fork. This is the proven shape; the ADR wording
+  moves from "overlay registry" to "per-connection registry installed
+  as the session's dispatch registry." Cost: `services/list` /
+  `services/schema` registered on the base registry must be
+  re-registered (or inherited via fork) per session to stay
+  discoverable — the fork makes that free if the bootstrap ops are
+  part of the fork source.
+- **(b) overlay-aware top-level dispatch.** Add an overlay fallback
+  in `dispatch` / `run_loop_single_stream`: when the base registry
+  misses, consult `connection.overlay_env()`. Keeps one static
+  registry and makes the amendment's wording true as written, but
+  touches the shared dispatch loop (higher blast radius, and the
+  overlay's `invoke_with_policy` shape — namespace-scoped, parent-
+  context-driven — does not match the top-level call shape, so the
+  fallback needs care: op-type lookup, ACL, visibility).
+
+Recommendation: (a). It is the only shape with an end-to-end proof,
+it needs no dispatch-loop change, and it matches ADR-019's layering
+(the overlay stays what it is today: the nested-invocation landing
+zone for imported ops — which F-05 needs).
+
+**Fix (Unit 2):** `Clone`/fork surface + the ADR wording amendment;
+acceptance = the review-001-style e2e gate (open op resolves over a
+live channels connection through the *fork*).
+
+## F-03 [minor] — Per-session fork + static base: no `Default`/builder seam exists to compose "base ops + per-session ops" without re-registration
+
+**Verified:** YES. Even with a `Clone` surface (F-02), the hook must
+compose three sources into the per-session registry: the deployment's
+base ops (already registered, not retained anywhere as a "source"
+list), the generic channel ops (`ChannelOperations::register_on`),
+and the openables. A fork of the base registry makes the first source
+free — this finding exists only if (a) is chosen *without* the fork
+(i.e., rebuilding per-session registries from scratch). Record it as
+a constraint on the fork API: it must clone *registered handlers*
+(`HandlerKind` clones carry their closures), not just specs, so
+`services/list` keeps its closure over the base registry.
+`publish_validator`'s `validators` map must be carried too (schema
+enforcement would silently vanish otherwise).
+
+**Fix:** fold into F-02's fork surface; the acceptance test covers it
+(`services/schema` on the fork still validates input).
+
+---
+
+# Part B — The connect side cannot serve; no wire path announces ops
+
+## F-04 [major] — The connect side's channel-0 read pump resolves responses only; inbound `call.requested` frames are silently dropped — there is no serving half on channel 0
+
+**Verified:** YES. `ChannelClient::from_connection`
+(`src/channels/client.rs:85-134`) spawns the read pump as
+`read_single_stream_until_closed(single_stream_reader, &pending_map)`
+(`connection.rs:681-686`), whose only job is resolving the
+`PendingRequestMap` — its frame handler (`dispatch_envelope`,
+`connection.rs:705-...`) branches on `call.responded` /
+`call.completed` / `call.aborted` / `call.error` and **has no arm for
+`call.requested`**. The accept side's mirror
+(`Dispatcher::run_loop_single_stream`, `dispatch.rs:727-850`) serves
+requests and has no pending-resolution arm. Each side silently drops
+the other half's frames.
+
+Consequences:
+
+- A consumer that dials a producer via `ChannelClient` can call ops
+  and open channels, but if the producer later calls an op *the
+  consumer serves*, the request is dropped (the caller's pending
+  hangs until the sweeper or connection close). This is the exact
+  shape ADR-022 §connection-direction-independence promises ("both
+  sides can be both simultaneously" — AGENTS.md §8), and ADR-036's
+  channel-0 pre-negotiation makes channel 0 the single control
+  plane where it must work.
+- It blocks the bootstrap registration story (F-05): the hub cannot
+  call anything the WS/native client serves, and the client cannot
+  even receive the call that would let it register.
+
+The fix is well-understood and symmetric: a full-duplex
+single-stream read loop that branches both ways — `call.requested`
+→ dispatch and write a response frame (the `run_loop_single_stream`
+arms), `call.responded`/`completed`/`aborted`/`published` → resolve
+pendings and route sink chunks (the read-pump arms). IDs are
+UUID-generated on each side (`generate_request_id()`), so
+cross-correlation is not a hazard; the abort arm needs to try both
+tables (in-flight sink aborts *and* the pending map's cascade), which
+`run_loop_single_stream`'s `EVENT_ABORTED` arm already does within
+its own scope. The wasm-targetable browser story compounds the
+value: the same serving loop is what a wasm alkcall browser
+compiles.
+
+**Fix (Unit 3):** a serving-capable single-stream loop usable from
+the connect side (e.g., a `Dispatcher` variant or a
+`CallConnection::serve_single_stream` that composes both frame
+directions), wired into `ChannelClient` behind an opt-in (a serving
+registry is not always present — a pure consumer may have none).
+Acceptance gates: hub→consumer call over an existing `ChannelClient`
+session resolves; consumer-served op participates in ACL (its
+`AccessControl` gates the hub's call); disconnect mid-call still
+fails pendings on both sides.
+
+## F-05 [major] — No wire mechanism announces client-side ops; ADR-022's import flow is hub→consumer only and the browser-native path has none
+
+**Verified:** YES. The envelope kind set is closed at six —
+`call.requested` / `call.responded` / `call.completed` /
+`call.aborted` / `call.error` / `call.published`
+(`src/protocol/wire.rs:12-17`). Op registration onto an overlay
+(`register_imported` / `register_imported_all`,
+`connection.rs:162-172`) happens only in-process: `from_call`
+(`client/from_call.rs:82-91`) builds bundles the *local* process
+registers after *it* dialed out — the consumer importing a hub's
+ops. There is no op, frame, or handshake by which a connected peer
+announces "here are the ops I serve." Every consumer-facing
+registration surface today requires the registrant to hold the
+local `CallConnection` handle in-process.
+
+Consequence: the decided bidirectionality model — either side
+registers ops the other can call (AGENTS.md §8; ADR-022 §direction
+semantics; alkhttp ADR-048's connection-local overlay for
+browser-registered ops) — has no on-the-wire expression for the
+non-in-process side. It is exactly the assumed-bootstrap-op set the
+protocol already leans on (`services/list` is dialed and called by
+`from_call` on every import; each side is *expected* to serve it)
+extended by one op: e.g. `op/register` carrying the
+`HandlerRegistration`'s serializable parts (spec + provenance +
+access control), whose hub-side per-session handler writes the
+registration into that connection's overlay via
+`register_imported`. The overlay is already the landing zone
+(`compose_root_env` attaches it keyed by identity,
+`dispatch.rs:211-213`), and discovery of what a peer registered is
+already built: `services/list-peers` reads
+`ctx.env.peer_ids()` / `peer_operations()`
+(`registry/discovery.rs:260-307`) — the same overlay-populated env.
+
+Scope decision this forces (record it in the ADR, either way):
+
+- **Design it** (recommended): `op/register` (+ a deregister or
+  replace semantics for the reconnect path) as an assumed op in the
+  bootstrap set, served per-session; the wire stays six kinds. The
+  `HandlerRegistration`'s `Handler` closures cannot cross the wire —
+  the client sends spec + a call-forwarding contract, and the
+  hub-side handler wraps it as a forwarding handler that issues a
+  nested `call.requested` back over channel 0 (the same shape
+  `from_call`'s imported bundles use — `protocol/adapter.rs:541,610`
+  show the in-process twin).
+- **Or scope-cut**: amend ADR-022/alkhttp ADR-048 to say peer-side
+  op registration is hub-initiated-import only until a contract
+  exists. Cheaper, but it re-opens the same hedge the deferral was
+  criticized for — the promise would remain written-but-unmechanized.
+
+Recommendation: design it. The pieces are all present; the cost is
+one op handler, an `AccessControl`-gated registration surface (the
+registry's existing ACL path covers it — an op with restrictive
+`AccessControl` cannot be overwritten by an unprivileged peer), and
+the F-04 serving loop to make it reachable in both directions.
+
+**Fix (Unit 3):** with F-04 — bootstrap op + client serving loop +
+ADR-022 amendment naming the bootstrap op set (assumed ops each side
+may serve: `services/list`, `services/schema`, and `op/register`).
+
+## F-06 [minor] — `services/list` on a per-session fork needs a closed-over fork reference, or per-session ops are undiscoverable
+
+**Verified:** YES. `services_list_handler` closes over a *specific*
+`Arc<OperationRegistry>` (`registry/discovery.rs:233-258`) — it
+lists that registry's ops. Under the F-02(a) shape, per-session
+openables live on the fork, so a `services/list` handler closed over
+the base registry cannot see them. Two clean outs, both cheap: (i)
+fork *before* registering the bootstrap discovery ops (then the
+handler closes over the fork), or (ii) the fork API registers
+base-registry bootstrap ops fresh on the fork. Either way the
+constraint is: discovery ops must be registered against the
+per-session registry, not the static one. Also note `services/list`
+on the fork still ACL-filters (the handler re-checks per-caller) —
+no privilege regression.
+
+**Fix:** fold into Unit 2 (the fork seam registers bootstrap ops on
+the fork); acceptance: a per-session openable appears in
+`services/list` for an authorized caller and not for an unauthorized
+one.
+
+---
+
+# Non-findings (verified correct, recorded to bound the re-review)
+
+- **The e2e reference shape is real and passes** —
+  `channels/client.rs:767-870` wires `ChannelClient` ↔
+  `ChannelsAdapter` over duplex with an open op registered on the
+  per-connection registry, invoked end-to-end, quota reserved. The
+  F-02(a) shape is not speculative; it is in-tree.
+- **`register_openable`'s wrapper machinery is complete** — ACL via
+  the registry's normal invoke path, `check_open` → `open_channel` →
+  ledger → spawn → teardown decrement, with `Once`/`Stream`/`Sink`
+  arms (`channels/operations.rs:399-453, 483-555`). Nothing in F-01
+  re-opens it; the ops it registers just need a dispatch-reachable
+  home.
+- **Channel-id allocation is collision-safe** for the bidirectional
+  open story (accept side even from 2, connect side odd from 1,
+  `manager.rs:105-125`; `adopt_channel` for the non-allocating side,
+  `manager.rs:273-309`).
+- **The `call.*` envelope kind set staying closed is correct** —
+  AGENTS.md §7 allows *adding* kinds but F-05's resolution does not
+  need one; bootstrap ops over channel 0 are the lighter door.
+- **Wasm-cleanliness is unaffected** — the recommended fixes (fork +
+  handler + read-loop branch) touch no transport, no fs/net tokio
+  features; `cargo check --target wasm32-unknown-unknown` remains
+  the release gate.
+
+---
+
+# Remediation plan
+
+Sequenced by dependency. Units 1–3 are alkcall work; the alkhttp
+consequences are tracked as alkhttp review 003 Unit 1 (updated to
+point here).
+
+## Unit 1 — Decision recording (F-01, F-02 direction)
+
+ADR work only:
+
+- ADR-047 §4 amendment #2: the open-op registration mechanism is the
+  **per-connection registry installed as the session's dispatch
+  registry** (option (a)); the overlay registry remains the landing
+  zone for peer-announced ops (F-05) and nested invocation.
+- ADR-022 amendment: the bootstrap-op set (each side may serve
+  `services/list`, `services/schema`, `op/register`), and the
+  direction-semantics note that serving on the connect side is
+  opt-in (F-04).
+- alkhttp ADR-048's bidirectionality promise is re-pointed at these
+  ADRs instead of standing as an unmechanized commitment.
+
+Gate: ADRs recorded; alkhttp OQ-05 and ADR-048 cross-reference them.
+
+## Unit 2 — Fork surface + per-session composition (F-02, F-03, F-06)
+
+- `Clone` (or explicit fork) on `HandlerRegistration` +
+  `OperationRegistry`, carrying handlers and validators.
+- A composition helper (e.g. `OperationRegistry::fork_with(...)`) or
+  documented pattern: fork base → register generic channel ops +
+  openables + bootstrap ops against the fork.
+- Gate: e2e over a live channels connection — open op resolves
+  through the fork; `services/list` shows per-session openables
+  (ACL-filtered); `services/schema` on the fork still validates.
+
+## Unit 3 — Client serving half + bootstrap registration (F-04, F-05)
+
+- Serving-capable single-stream loop (compose the read-pump and
+  dispatch arms); wired into `ChannelClient` opt-in.
+- `op/register` bootstrap op (per-session handler →
+  `register_imported` into the connection overlay), with
+  `AccessControl` gating and reconnect/replace semantics decided in
+  the ADR.
+- Gates: hub→consumer call over an existing session resolves;
+  peer-registered op is discoverable via `services/list-peers` and
+  callable back over channel 0; unregistered/unauthorized
+  registration attempts fail loudly; disconnect drops the overlay
+  and fails both sides' pendings.
+
+## Unit 4 — alkhttp wiring (downstream, after Units 1–3)
+
+Tracked in alkhttp review 003 (Units 2–4 there): openables surface,
+`install_channel_zero` rework over the fork, retained session
+handle, e2e tests, spec reconciliation (OQ-05, ADR-067/048 v1-cut
+notes).
+
+---
+
+## Verification log (this pass)
+
+- All claims verified at tree `c16b069`; the F-01 dispatch claim was
+  checked against both dispatch paths (`run_loop` stream-per-request
+  reads `handle_stream` → same base-registry-only resolution;
+  `run_loop_single_stream` at `dispatch.rs:727-850`).
+- The F-04 claim was verified by reading both loops' frame arms:
+  `read_single_stream_until_closed` → `dispatch_envelope`
+  (`connection.rs:681-705`, no `EVENT_REQUESTED` arm) vs
+  `run_loop_single_stream` (`dispatch.rs:754-850`, no
+  pending-resolution arm).
+- The F-02 clone claims were verified type-by-type, including the
+  third-party `jsonschema::Validator` (0.46.10) and the vendored
+  `Capabilities` (manual `Clone` preserving `Secret`'s clone).
+- The F-05 wire-closure claim was verified against the full kind set
+  (`wire.rs:12-17`) and grep over `src/` for any wire-side
+  registration path (none; `register_imported*` callers are
+  in-process only).
+- `services/list-peers`' overlay-backed discovery verified at
+  `registry/discovery.rs:260-307` (reads `ctx.env.peer_ids()` /
+  `peer_operations()`, populated by `compose_root_env` at
+  `dispatch.rs:211-213`).
+- Baseline gates re-run for this pass: cargo test (565), clippy
+  (all-targets), fmt — all clean. No source changes.
+
+## Remediation log (2026-09-03)
+
+Units 1–3 landed in one pass. Every claim above was re-verified
+against the source before remediation began; all six findings
+confirmed (F-02's member-type list was missing `ScopedPeerEnv`
+(`context.rs:53`) — also Clone, which made the fork surface simpler
+than the review estimated).
+
+**Unit 1 — decision recording (F-01/F-02 direction, F-04/F-05 scope):**
+
+- ADR-047 §4 amendment #2 (2026-09-03): the per-connection registration
+  mechanism is the **per-session fork of the base registry installed as
+  the session's dispatch registry** (option (a)); the overlay registry
+  stays the nested-invocation / peer-announced-ops landing zone.
+- ADR-022 amendment (2026-09-03): the bootstrap-op set (`services/list`,
+  `services/schema`, `op/register`), the opt-in connect-side serving
+  loop, and the `op/register` wire shape (spec in `services/schema`
+  JSON + `replace` flag; forwarding-handler wrap into the connection
+  overlay; `ALREADY_EXISTS` collision gate).
+- alkhttp ADR-048's reconciliation note and OQ-05 re-pointed at these
+  ADRs (Unit 1b).
+
+**Unit 2 — fork surface + per-session composition (F-02, F-03, F-06):**
+
+- `OperationRegistry` gained interior mutability (`parking_lot::RwLock`
+  around the operations and validators maps) so a registry can live
+  behind an `Arc` and be self-referential. `register` now takes
+  `&self`; `registration`/`list_operations` return owned clones
+  (handlers are `Arc` closures — the clone is cheap).
+- `OperationRegistry::fork()` — deep copy of registrations (handlers,
+  provenance, composition authority, capabilities, `scoped_env`) and
+  the cached publish-schema validators. `OperationRegistryBuilder::
+  from_registry` seeds the builder path from an existing registry.
+- `install_bootstrap_discovery(&Arc<OperationRegistry>)` — registers
+  `services/list` / `services/list-peers` / `services/schema` closed
+  over the fork itself, so per-session openables are discoverable
+  (F-06) and `services/schema` answers from the fork.
+- Gate: `fork_registry_open_op_resolves_and_is_discoverable`
+  (`src/channels/client.rs`) — open op registered on the fork resolves
+  over a live channels connection; the openable appears in
+  `services/list` on the fork; `services/schema` on the fork still
+  answers. Unit tests: fork independence both directions, validator
+  carry, post-dispatch registration through a shared `Arc`.
+
+**Unit 3 — client serving half + bootstrap registration (F-04, F-05):**
+
+- `Dispatcher::serve_single_stream` — the full-duplex single-stream
+  loop: `call.requested` → dispatch + response frames (the
+  `run_loop_single_stream` arms); `call.responded`/`completed`/`error`
+  → pending resolution (the read-pump arms); `call.aborted` → both
+  tables (in-flight sink aborts and the pending map's cascade);
+  `call.published` → inbound in-flight sinks. Disconnect fails
+  pendings and drops in-flight sinks (teardown identical to the two
+  half-loops it composes).
+- `ChannelClient::from_connection_with_serving(connection,
+  Option<ServingConfig>)` — opt-in serving; `from_connection`
+  preserves the pure-consumer default (resolution-only read pump).
+  Gate: `serving_loop_hub_to_consumer_call_resolves` — hub→consumer
+  call resolves through the consumer's serving loop, consumer→hub
+  still resolves in the same loop.
+- `registry::op_register` — `OpRegisterRequest` wire DTO,
+  `op_register_spec`, `op_register_handler` (rebuild → forwarding stub
+  → `register_imported` into the connection overlay, forced
+  `Internal`/`FromCall`), `announce_op`. `CallError::already_exists`
+  added for the collision gate. `spec_to_json_pub` made the
+  `services/schema` wire shape public; `rebuild_spec_for` and the
+  forwarding-handler constructors are crate-shared with `from_call`.
+  Gates: `op_register_announce_then_hub_call_routes_back_to_consumer`
+  (announce → overlay → hub call → forwarding stub → consumer serves)
+  plus unit tests (round-trip, collision, replace, forced
+  visibility/provenance).
+
+**Baseline after remediation:** cargo test 581 passed / 0 failed;
+clippy (all-targets, `-D warnings`) clean; `cargo fmt --check` clean;
+wasm target gate re-run below. No wire-format changes: the envelope
+kind set stays closed at six; the bootstrap ops ride channel 0's call
+registry.

docs/reviews/005-serving-loop-concurrency-and-op-register-review.md

@@ -0,0 +1,704 @@
+# Review 005 — Serving-Loop Concurrency and `op/register` Composition (post-remediation review of `f84d214`)
+
+## Status
+
+Units 1–3 remediated and verified (G-01..G-05). All findings closed;
+the review is resolved. See Remediation log.
+
+## Scope
+
+Post-remediation review of commit `f84d214` (review 004 Units 1–3:
+per-session fork registry, connect-side serving loop, `op/register`).
+The remediation landed all six F-findings' mechanisms; this pass
+reviews the landed implementation against the decided spec (ADR-022
+amendment 2026-09-03, ADR-047 §4 amendment #2, review 004's
+acceptance gates) rather than re-litigating the findings.
+
+Every finding below was verified directly in source at tree
+`f84d214`. The headline finding (G-01) was verified **empirically**:
+a probe test was added to the tree, run, observed to reproduce the
+defect, and removed — the tree at this review's baseline carries no
+source changes. Findings continue alkcall review numbering with the
+new prefix `G` (001–004 used P/C/R/A/B/C/D/F — each review numbers
+independently).
+
+Cross-repo context: the alkhttp re-pointing claimed by the remediation
+log was verified in that repo (`5b62307` — `open-questions.md:102-112`,
+`decisions/048…md:19-25`). Unit 4 (alkhttp wiring) remains downstream
+and is not reviewed here.
+
+## Baseline verification (this pass)
+
+```
+cargo test                                        → 581 passed, 0 failed
+cargo test --all-features                         → 598 passed, 0 failed
+cargo clippy --all-targets -- -D warnings         → clean
+cargo clippy --all-features --all-targets -- -D warnings → clean
+cargo fmt --check                                 → clean
+cargo check --target wasm32-unknown-unknown       → clean
+cargo clippy --target wasm32-unknown-unknown -- -D warnings → clean
+cargo doc --no-deps                               → clean
+```
+
+All gates the remediation log claims reproduce. The suite is green —
+and again the green is partial: the gates test the flows they name,
+and the one flow the ADR amendment promises most loudly (nested
+composition of peer-announced ops) is the one no gate exercises. See
+G-02 for why the existing gate cannot catch G-01.
+
+## Verdict
+
+- **Unit 2's fork surface is solid.** Interior mutability, lock
+  discipline, fork independence, validator carry, and the
+  self-referential bootstrap-discovery closure are all correct and
+  tested (see Non-findings).
+- **Unit 3's serving loop resolves the F-04 frame gap but inherits the
+  dispatch loop's serial-inline shape** — and the F-05 flow it exists
+  to enable is exactly the shape that deadlocks under it. The ADR-022
+  amendment's "invocable via nested composition" promise is
+  unmechanized for wire-dispatched handlers (the primary caller
+  shape); a 30s sweeper timeout masquerades as resolution.
+- **The F-05 acceptance gate does not exercise the forwarding stub** —
+  the one component that is genuinely new protocol surface. It
+  verifies announce-lands-in-overlay and the consumer serves, by
+  calling the announced op *directly*, bypassing the stub entirely.
+- **`op/register`'s collision gate is overlay-only**, so a
+  peer-announced op can shadow the serving side's own ops in nested
+  composition (connections resolve before base in `PeerCompositeEnv`).
+
+The architecture decisions (fork as dispatch registry; overlay as
+landing zone; bootstrap-op set; opt-in serving) remain sound. The
+defects are in the serving loop's concurrency model and in what the
+gates measure.
+
+## Severity legend
+
+Same scale as review 004:
+
+- **[critical]** — a decided spec invariant is violated in a way that
+  makes a promised capability unreachable end-to-end; or corrupts
+  data.
+- **[major]** — a core protocol path cannot serve a decided behavior;
+  works only via shapes the spec does not describe.
+- **[minor]** — drift, doc/spec inconsistency, or a missing
+  convenience with no correctness impact.
+
+---
+
+# Part A — The serving loop's concurrency model
+
+## G-01 [major] — Same-connection nested composition deadlocks the serving loop; the forwarding stub's nested call resolves only via the 30s sweeper
+
+**Status: REMEDIATED (Unit 1)** — see Remediation log; both gates
+added and verified load-bearing against the pre-fix loop.
+
+**ADR drift:** ADR-022 amendment (2026-09-03) §`op/register`: the
+announced op is *"invocable via nested composition (`env.invoke`)"*;
+the amendment's whole point is that the hub's handlers compose
+peer-announced ops. Review 004 F-05's recommendation said the same:
+the hub-side handler wraps announcements *"as a forwarding handler
+that issues a nested `call.requested` back over channel 0"*.
+
+**Verified:** YES, by code trace and by an empirical probe (see
+Verification log).
+
+Mechanism, in four steps:
+
+1. `Dispatcher::serve_single_stream` awaits `dispatch()` **inline** in
+   its read loop for `call.requested` frames
+   (`src/protocol/dispatch.rs:1009-1014`) — the same serial-inline
+   shape as the pre-existing accept-side `run_loop_single_stream`
+   (`:765-770`). Only the Sink arm is spawned
+   (`:1031-1053`); Once responses and Sub pumps
+   (`pump_stream_single_stream`, awaited inline at `:1027-1029`)
+   hold the loop.
+2. The F-05 forwarding stub (`make_forwarding_handler`,
+   `src/client/from_call.rs:392-415`) issues its nested call via
+   `connection.call_with_payload` (`:408`) — over **the same
+   connection** whose loop dispatched the parent. In single-stream
+   mode the pending resolves only when *some* loop reads the
+   `call.responded` frame (`connection.rs:274-303` — register
+   pending, write frame, await receiver; resolution is the reader's
+   job).
+3. The only reader of that connection is the serving loop itself —
+   which is blocked in step 1 awaiting the parent dispatch, whose
+   handler is blocked in step 2 awaiting the nested call. Circular
+   wait. The transport buffers the response frame; nobody reads it.
+4. The 30s sweeper (`DEFAULT_CALL_TIMEOUT`, `connection.rs:32`;
+   `evict_expired`, run every 10s per `SWEEPER_INTERVAL`,
+   `dispatch.rs:44`) is the only thing that breaks the cycle —
+   the nested call resolves as `CallError::timeout`.
+
+**Empirical probe (this pass):** hub serves `op/register` + a
+`hub/compose` handler that invokes the connection overlay's
+registration (the forwarding stub — the exact resolution
+`PeerCompositeEnv` → `OverlayOperationEnv` produces for nested
+composition) from inside a wire-dispatched handler; consumer announces
+`consumer/exec`, then calls `hub/compose` over the wire. Result:
+
+```
+PROBE: hub/compose resolved after 30.000761619s:
+  Err(CallError { code: "TIMEOUT", message: "request timed out", retryable: true })
+```
+
+A second run with an 8s timeout confirmed the hang is unconditional
+(not load-dependent): zero progress until the sweeper. The probe was
+removed after evidence capture; the tree is unchanged.
+
+**Consequences (verified):**
+
+- The ADR-022 amendment's flagship flow — a hub handler composing a
+  peer-announced op — fails for every wire-dispatched parent handler.
+  It "works" only for parents dispatched from *in-process* contexts
+  (where the serving loop is idle and free to resolve the nested
+  call's response) — a shape the ADR does not describe. Per the
+  severity legend this is a decided behavior served only via an
+  undescribed shape; it borders [critical] for the composition flow
+  specifically.
+- The same circular wait applies to composition of `from_call`
+  *imports* on the accept side: `run_loop_single_stream` has the same
+  inline shape, and the connection overlay's imported-op stubs ride
+  the same connection. That hazard is **latent pre-`f84d214`** (the
+  accept side had the inline loop and `register_imported` before this
+  commit) — `f84d214` makes it load-bearing by (a) adding the wire
+  path that populates overlays with stubs, (b) putting serving loops
+  on the connect side where both directions are live by design, and
+  (c) writing the composition promise into ADR-022.
+- The same starvation is not limited to the stub: while the loop
+  serves *any* Sub (`pump_stream_single_stream` inline), the side's
+  own outbound pendings queue unresolved. A serving-enabled consumer
+  that calls out while serving a long-lived Sub to the same peer eats
+  30s timeouts on its own calls. The F-04 gate tests the two
+  directions only **sequentially** (hub→consumer resolves, *then*
+  consumer→hub) — never interleaved.
+- Head-of-line blocking: the peer's subsequent requests queue behind
+  each inline dispatch.
+
+**Fix (Unit 1):** spawn the Once dispatch and the Sub pump the way the
+Sink arm already is (the `SharedFrameWriter` already serializes
+concurrent frame writes, so per-request frame ordering is preserved);
+track spawned task handles for teardown on loop exit (the Sink arm's
+`in_flight_sinks` pattern, generalized). Keep read-loop resolution
+unblocked at all times. Apply the same rework to
+`run_loop_single_stream` (or factor the shared loop) — the accept side
+has the identical latent hazard via imported-op composition.
+Acceptance: the probe above (recreated as a permanent gate) resolves
+in < 1s; a concurrent-interleaving gate (outbound call resolving while
+a Sub is being served) passes without sweeper intervention.
+
+## G-02 [major] — The F-05 acceptance gate bypasses the forwarding stub it claims to prove
+
+**Status: REMEDIATED (Unit 1)** — the stub-exercising gate and the
+interleaved-directions gate are added; see Remediation log.
+
+**Verified:** YES. `op_register_announce_then_hub_call_routes_back_to_consumer`
+(`src/channels/client.rs:1412-1543`) asserts the announce resolves and
+the overlay holds the op — then calls
+`accept_conn.call("consumer/exec", …)` (`:1529`). That is a **direct
+wire call**: the hub's dispatcher resolves `consumer/exec`… which is
+not on the hub at all — the frame crosses to the consumer, whose
+serving loop dispatches it from its *local* registry. The forwarding
+stub (`op_register_handler`'s `register_imported` bundle) is never
+invoked. The gate's comment — *"the forwarding stub routes the nested
+call back over channel 0"* — describes a path the test does not take.
+
+**Consequence:** the one behavior that is genuinely new protocol
+surface (stub routing) is untested, and the one defect it hides (G-01)
+is precisely in that path. This is the same failure class review 001
+flagged ("all coverage is unit-level / the integration seams are where
+the real problems live") — the remediation added e2e gates, but the
+F-05 gate measures an adjacent, easier path. The F-04 gate is
+likewise sequential-only (see G-01's third consequence).
+
+**Fix (fold into Unit 1):** add a gate that exercises the stub —
+announce → a wire-dispatched hub handler composes the announced op via
+`ctx.env` (or directly via the overlay registration) → resolves from
+the consumer. This is the removed probe, productized; it is the
+acceptance test G-01's fix must pass. Optionally extend the F-04 gate
+with an interleaved-directions phase.
+
+---
+
+# Part B — The `op/register` composition surface
+
+## G-03 [major] — Announced ops can shadow the serving side's own ops in nested composition; the collision gate checks the overlay only
+
+**Status: REMEDIATED (Unit 2)** — see Remediation log; the ADR-022
+amendment records the collision policy.
+
+**ADR drift:** ADR-018 (composition authority) and ADR-019 (the
+overlay is the *landing zone* for imported ops — a layer beneath the
+deployment's own registry, not a rival to it); ADR-022 amendment
+(silent on name collisions with the serving side's base registry).
+
+**Verified:** YES. `op_register_handler`'s collision gate is
+`connection.overlay_contains(&request.spec.name)`
+(`src/registry/op_register.rs:142`) — it consults **only the
+connection overlay**. The serving registry (the fork the session
+dispatches over) is not consulted. Meanwhile `PeerCompositeEnv`
+resolves **connections before base** (`src/registry/env.rs:230-245` —
+session, then `connection_order`, then base), and `compose_root_env`
+attaches the dispatching connection's overlay
+(`dispatch.rs:207-214`).
+
+Consequence: a peer announces an op named `fs/readFile` (or any name
+the serving side has registered — `services/list` itself is fair
+game). The registration lands in the overlay (forced `Internal`,
+`FromCall` — both per spec). From then on, every handler
+**wire-dispatched from that peer's connection** that nested-composes
+that name via `ctx.env` resolves to the peer's forwarding stub instead
+of the serving side's own op. The forced `Visibility::Internal` does
+not help here — it blocks wire invocation, not `env.invoke`
+(`OverlayOperationEnv` gates on `AccessControl` only,
+`connection.rs:749-836`; the composed child context is `internal:
+true` by design).
+
+Concretely: a hub handler processing peer A's request composes
+`fs/readFile` expecting *its own* registry op (with the deployment's
+`scoped_env`, capabilities, and ACL); it silently gets peer A's stub —
+peer A chooses where the call goes and what it returns. Composition
+authority (ADR-018) is decided by the composing handler's *deployer*;
+a same-name peer registration silently rewrites it. The effect is
+scoped to handlers whose root context is that connection's
+(`compose_root_env` attaches that connection's overlay), which bounds
+the blast radius — but that is exactly the flagship F-05 flow.
+
+**Fix (Unit 2):** decide the collision policy and record it in the
+ADR-022 amendment. Recommended: `op_register_handler` also rejects
+names registered on the serving registry (pass the fork — or a
+name-set closure — into the handler alongside the connection, the
+same way `install_bootstrap_discovery` closes over its registry);
+`ALREADY_EXISTS` for base collisions regardless of `replace` (the
+reconnect path needs to re-announce *peer* ops, not to overwrite the
+deployment's). The alternative — changing `PeerCompositeEnv` to
+resolve base before connections — would break `from_call`'s
+namespace-prefixed imports (prefixed names don't collide) is not
+needed for that, but would change long-decided layering semantics; the
+registration-side gate is the cheaper, narrower door.
+
+## G-04 [minor] — `resource_id_path` does not survive the spec wire round-trip
+
+**Status: REMEDIATED (Unit 3)** — see Remediation log.
+
+**Verified:** YES. `spec_to_json_pub` serializes
+name/namespace/op_type/visibility/schemas/error_schemas/access_control
+(+ `channel_open`/`publish_schema` markers;
+`src/registry/discovery.rs:211-234`) — `resource_id_path` is not
+serialized. `rebuild_spec_for` cannot parse it
+(`src/client/from_call.rs:209-286`), and the field exists on
+`OperationSpec` (`src/registry/spec.rs:190`). An announced (or
+`from_call`-imported) op that declares ownership-scoped resource
+extraction silently loses it; the rebuilt spec's ACL checks run with
+`resource_id: None`.
+
+This is a pre-existing gap in the `from_call` import shape, not
+introduced by `f84d214` — but `op/register` makes it load-bearing for
+a second wire surface (peer-announced specs) that explicitly promises
+"the serializable parts" of a registration. Note `namespace` *is*
+serialized but `rebuild_spec_for` ignores it in favor of the
+`namespace_prefix` parameter — consistent for `op/register`
+(`replace`/re-announce uses `None`), worth a comment either way.
+
+**Fix (Unit 3):** add `resource_id_path` to both halves of the
+round-trip plus a round-trip test. Additive optional field in the
+`services/schema` JSON shape — no existing consumer breaks.
+
+## G-05 [minor] — `install_bootstrap_discovery` registers `services/list-peers`; the ADR-022 amendment's bootstrap set doesn't name it
+
+**Status: REMEDIATED (Unit 3)** — see Remediation log.
+
+**Verified:** YES. `install_bootstrap_discovery` registers
+`services/list`, `services/list-peers`, and `services/schema`
+(`src/registry/discovery.rs:290-313`; `list-peers` at `:300`). The
+ADR-022 amendment's bootstrap set lists exactly three ops —
+`services/list`, `services/schema`, `op/register`
+(`022-…contract.md:380-393`). Review 004's remediation log says
+"list-peers" too; the ADR text was never updated.
+
+Consequence: none at runtime (installing a fourth op is strictly
+additive, and `services/list-peers` is what makes F-05's
+"peer-announced ops are discoverable" true). It is a
+doc/spec-consistency drift on a set the ADR declares *"closed."*
+
+**Fix (Unit 3):** add `services/list-peers` to the ADR-022
+amendment's bootstrap-op list (recommended — `from_call`-import
+discovery already relies on it) or stop installing it; one line
+either way.
+
+---
+
+# Non-findings (verified correct, recorded to bound the re-review)
+
+- **The fork surface (F-02/F-03) is solid.** Lock discipline is
+  correct: `registration()` clones under a read guard and drops it
+  before handler invocation (`registration.rs:190-192`), so
+  self-referential discovery handlers cannot deadlock their own
+  registry; `register` takes the two maps' write locks sequentially
+  and never across an `await`. `fork()` deep-copies both maps
+  (handlers are `Arc` closures — cheap, and the F-03 constraint that
+  closures must carry is met); validators carry; fork independence is
+  tested both directions; `OperationRegistryBuilder::from_registry`
+  seeds the builder path.
+- **`install_bootstrap_discovery` (F-06) is correct** — handlers
+  closed over the same `Arc` they're registered on, so post-install
+  per-session registrations are visible at call time; ACL filtering
+  stays per-caller; the fork gate
+  (`fork_registry_open_op_resolves_and_is_discoverable`) exercises
+  exactly the review-004 acceptance shape.
+- **`serve_single_stream`'s pending-resolution arms match the two
+  half-loops they compose.** Frame by frame against
+  `run_loop_single_stream` (`dispatch.rs:765-905`) and
+  `dispatch_envelope` (`connection.rs:717-741`): responded/completed/
+  error resolve pendings; aborted tries both tables (in-flight sink
+  aborts, then the pending cascade via `handle_abort`); published
+  routes inbound sinks with the same validation/keep semantics; teardown
+  (fail_all + sink drop + sweeper abort) matches. The direction
+  disambiguation by table membership (documented at
+  `dispatch.rs:940-966`) is correct as written — G-01 is *not* a
+  frame-routing bug; the arms are right, the loop's scheduling is the
+  defect.
+- **The pure-consumer default is unchanged.** `from_connection`
+  delegates with `None` and keeps the resolution-only read pump; no
+  behavior change for existing callers.
+- **`op/register`'s DTO and unit semantics are clean** — one spec
+  serialization on the wire (`spec_to_json` shape), forced
+  `Internal`/`FromCall` on landing (unit-tested), `replace` semantics
+  tested, `ALREADY_EXISTS` collision gate tested (against the
+  overlay — see G-03 for the gap).
+- **The alkhttp cross-repo claims check out.** `open-questions.md`
+  OQ-05 and ADR-048's re-point are present in that repo at `5b62307`.
+- **No non-English text reached the tree.** The remediation session's
+  language slip (reported anecdotally) did not land in any code,
+  doc, or metadata: a CJK-range regex sweep over all `.rs`/`.md`/
+  `.toml`/`.json` in the repo returns zero matches at `f84d214`.
+- **Gates reproduce.** All commands in Baseline verification pass at
+  the review tree, matching the remediation log's claims
+  (581 default / 598 all-features).
+
+---
+
+# Remediation plan
+
+Sequenced by dependency. All units are alkcall work; Unit 4 (alkhttp
+wiring) stays downstream and should **not** start before Unit 1 —
+alkhttp's serving consumers would compose over the same connection
+and hit G-01 immediately. (All three units landed — see Remediation
+log.)
+
+## Unit 1 — Concurrent serving loop + a stub-exercising gate (G-01, G-02)
+
+- Rework `serve_single_stream`'s `EVENT_REQUESTED` arm: spawn the Once
+  dispatch and the Sub pump (the Sink arm's existing pattern,
+  generalized; `SharedFrameWriter` serializes frames; track handles
+  for teardown). Apply the same rework to `run_loop_single_stream` or
+  extract the shared loop.
+- Add the G-02 gate: announce → wire-dispatched hub handler composes
+  the announced op via nested composition → resolves from the
+  consumer. This is the removed probe, productized; it is the
+  acceptance test for G-01.
+- Optional: an interleaved-directions phase on the F-04 gate
+  (outbound call resolving while a Sub is being served).
+- Gate: the probe recipe (Verification log) resolves in < 1s; no
+  sweeper evictions in the gates.
+
+## Unit 2 — `op/register` collision policy (G-03)
+
+- `op_register_handler` gains a serving-registry name-set (closure or
+  `Arc<OperationRegistry>`); base-registry name collisions reject
+  with `ALREADY_EXISTS` regardless of `replace`.
+- ADR-022 amendment: record the collision policy (peer-announced ops
+  may collide with peer-announced ops — `replace` governs; never with
+  the serving side's own registrations).
+- Gates: announcing a base-registered name (External *and* Internal)
+  fails loudly; announcing a distinct name still lands; nested
+  composition of a base op is unaffected by an unrelated announce.
+
+## Unit 3 — Round-trip completeness + doc alignment (G-04, G-05)
+
+- `resource_id_path` through `spec_to_json_pub` + `rebuild_spec_for`
+  + round-trip test.
+- ADR-022 bootstrap list gains `services/list-peers` (or drop it from
+  the installer; recommend adding).
+
+---
+
+# Remediation log
+
+## Unit 3 — Round-trip completeness + doc alignment (G-04, G-05) — LANDED
+
+**G-04 fix.** `resource_id_path` now rides both halves of the spec
+wire round-trip: `spec_to_json_pub` serializes it as an optional
+`resource_id_path` string (`src/registry/discovery.rs`), and
+`rebuild_spec_for` parses it into the rebuilt spec's fourth
+constructor argument (`src/client/from_call.rs`). Additive optional
+field — absent stays absent, no existing consumer breaks (verified by
+the companion gate). `rebuild_spec_for`'s doc note from the review
+("namespace *is* serialized but ignored in favor of
+`namespace_prefix`") was addressed by leaving the behavior as-is: for
+`op/register` the parameter is `None` (consistent) and for `from_call`
+the prefix is authoritative — the round-trip tests pin the `name`
+field handling.
+
+**Gates:** `spec_round_trips_resource_id_path` (serialize → parse →
+field intact) and
+`spec_without_resource_id_path_stays_absent_through_round_trip`
+(absent key serializes nothing; rebuilt `None`) in
+`src/client/from_call.rs`.
+
+**G-05 fix.** The ADR-022 amendment's bootstrap-op set gained
+`services/list-peers` with a dated note (review 005 G-05) explaining
+that the installer has registered it since the amendment landed and
+the doc lagged the code. The set remains closed at four.
+
+**Verification (post-fix):**
+
+```
+cargo test                                        → 589 passed, 0 failed
+cargo test --all-features                         → 606 passed, 0 failed
+cargo clippy --all-targets -- -D warnings         → clean
+cargo clippy --all-features --all-targets -- -D warnings → clean
+cargo fmt --check                                 → clean
+cargo clippy --target wasm32-unknown-unknown -- -D warnings → clean
+cargo doc --no-deps                               → clean
+```
+
+---
+
+## Unit 2 — `op/register` collision policy (G-03) — LANDED
+
+**Fix shape.** `op_register_handler` now takes the serving registry
+alongside the connection (`op_register_handler(connection,
+serving_registry)` — the review's "name-set closure" recommendation,
+materialized as the `Arc<OperationRegistry>` itself, mirroring how
+`install_bootstrap_discovery` closes over its registry). The handler
+checks `serving_registry.registration(name)` **before** the overlay
+gate and rejects base collisions with `ALREADY_EXISTS` regardless of
+`replace`; overlay collisions keep the existing `replace` semantics.
+Announced ops may collide with announced ops, never with the serving
+side's own registrations.
+
+**ADR-022 amendment:** the 2026-09-03 amendment's `op/register`
+section gained a "Collision policy (amended 2026-09-04, review 005
+G-03)" paragraph recording the decided policy and the rationale (the
+`PeerCompositeEnv` connections-before-base resolution would let an
+unscreened announce rewrite composition resolution; visibility is
+irrelevant to the gate). The ADR's status line notes the sub-amendment.
+
+**Call sites updated:** both e2e gates pass the fork the session
+dispatches over (`Arc::clone(&accept_registry)` after `fork()`), which
+is the production shape — the collision set is exactly the registry
+the serving loop dispatches against.
+
+**Gates:**
+
+- `handler_rejects_base_registry_collision_even_with_replace` —
+  base-registered `fs/readFile` (External) + `replace: true` →
+  `ALREADY_EXISTS`; the overlay stays clean and the serving
+  registration is untouched.
+- `handler_rejects_collision_with_internal_serving_op` — an
+  `Internal` base op is equally protected (the review's point that
+  forced `Visibility::Internal` on the *announced* spec never helped:
+  `OverlayOperationEnv` gates on `AccessControl`, and the composed
+  child is `internal: true` by design).
+- `handler_overlay_collision_still_governed_by_replace` — announce/
+  announce collisions still follow `replace` (the base gate is
+  scoped to the serving registry, not widened into the overlay).
+- `nested_composition_of_base_op_unaffected_by_unrelated_announce` —
+  after a successful distinct-name announce, composing the base op
+  through the real `compose_root_env` shape (`PeerCompositeEnv` +
+  `attach_peer(conn.overlay_env())`) resolves the serving side's own
+  op, not a peer stub. (Unit-level, using the actual env types rather
+  than a new full-wire fixture — the wire shape is already covered by
+  the Unit-1 gates.)
+
+**Deliberate non-change:** `PeerCompositeEnv` resolution order is
+untouched (connections before base) — the review's recommendation;
+reordering would change long-decided layering semantics
+(ADR-019/ADR-024) for no gain the registration-side gate doesn't
+deliver.
+
+**Verification (post-fix):**
+
+```
+cargo test                                        → 587 passed, 0 failed
+cargo test --all-features                         → 604 passed, 0 failed
+cargo clippy --all-targets -- -D warnings         → clean
+cargo clippy --all-features --all-targets -- -D warnings → clean
+cargo fmt --check                                 → clean
+cargo clippy --target wasm32-unknown-unknown -- -D warnings → clean
+cargo doc --no-deps                               → clean
+```
+
+---
+
+## Unit 1 — Concurrent serving loop + stub-exercising gates (G-01, G-02) — LANDED
+
+**Fix shape.** `dispatch()` was split into a synchronous start half and
+an awaited invocation:
+
+- `Dispatcher::dispatch_start`
+  (`src/protocol/dispatch.rs`) runs the sync prefix (identity
+  resolution, root context, op-type branch) and returns a
+  `StartedDispatch`: `Once` (the invocation as a boxed future), `Stream`
+  (the `ResponseStream`, returned synchronously — `invoke_streaming`
+  is a sync call), or `Sink` (started **inline**, unchanged). `dispatch()`
+  is now a thin wrapper (`Once` = await the boxed future) and keeps its
+  shape for `dispatch_requested`/`handle_stream`/the gateway.
+- The Sink start stays inline **by design**: its `chunk_tx` must be in
+  `in_flight_sinks` before the next `call.published` frame can be
+  routed; the loop insert cannot race the feed. Pub handlers that
+  block forever inside their sink future are a separate (pre-existing,
+  unreported) shape — the read loop no longer waits on any handler
+  except for the few instructions of the sink start itself.
+- Both single-stream loops' `EVENT_REQUESTED` arms spawn the Once
+  invocation (`spawn_once_dispatch`), the Sub pump
+  (`spawn_stream_pump`), and the sink's response writer
+  (`spawn_sink_response_writer`); handles are tracked in a
+  `spawned` list and aborted at loop exit (teardown: sink-map clear →
+  spawn aborts → `fail_all` → sweeper abort). `in_flight_sinks` moved
+  behind `Arc<parking_lot::Mutex>` because the ABORTED/PUBLISHED/ERROR
+  arms must not hold the guard across the `chunk_tx.send().await`
+  (non-`Send` guard across an await — the reason the pre-fix loop
+  couldn't just be `tokio::spawn`ed piecemeal). Guard-dropping
+  (`let entry = ...remove()` before the await) keeps lock discipline:
+  no lock is held across an await anywhere in the loops.
+- **`run_loop_single_stream` (the accept side) got the
+  pending-resolution arms** (`EVENT_RESPONDED`/`COMPLETED`/`ERROR` →
+  the outbound pending map). Previously it served only — an accept-side
+  nested-composing handler (e.g. a `from_call` imported-op stub riding
+  the same connection) had no loop resolving its response frames at
+  all. The G-01 accept-side latent hazard is mechanized shut, not just
+  unblocked: both single-stream loops are now the same shape
+  (dispatch-spawn + pending-resolution), the full-duplex loop
+  `serve_single_stream` composes them and is unchanged in its arm
+  semantics (frame-arm equivalence preserved per the non-findings
+  audit).
+- Write-failure semantics changed from `break` (close the loop) to
+  `warn` (keep reading) in the spawned Once path, matching the Sink
+  arm: a dying transport surfaces on the next read as
+  `ConnectionClosed`; a transient frame-write failure no longer tears
+  down every in-flight request on the connection.
+
+**Gates (G-02, productized from the removed probe):**
+
+- `hub_handler_composes_peer_announced_op_via_nested_composition`
+  (`src/channels/client.rs`): announce → consumer calls `hub/compose`
+  → the hub's serving loop wire-dispatches it → the handler resolves
+  `consumer/exec` via `ctx.env` nested composition → the forwarding
+  stub's nested `call.requested` crosses back to the consumer. This is
+  the stub path the F-05 gate bypassed. Two wiring facts surfaced (both
+  recorded as spec-consistent, both were silent before): (a) the hub's
+  channel-0 `Connection` must carry the peer identity —
+  `compose_root_env` attaches the connection overlay keyed by
+  `identity.id` (ADR-030 §5), and a deployment that skips identity
+  resolution silently gets no peer overlay (the test sets it via the
+  `AuthContext` the adapter passes to `install_channel_zero`);
+  (b) composition reachability is declared on the composing handler's
+  registration (`scoped_env: ScopedPeerEnv::new(["consumer/exec"])`) —
+  the empty `ScopedPeerEnv` is deny-by-default in
+  `PeerCompositeEnv::invoke_with_policy`, so a wire-dispatched handler
+  with no `scoped_env` composes nothing (the reachability gate, not the
+  overlay, was what the probe's first draft tripped over).
+- `outbound_call_resolves_while_inbound_subscription_is_being_served`
+  (the optional interleaved-directions phase on the F-04 shape): the
+  consumer subscribes to the hub (Sub served by the consumer's loop),
+  then calls `hub/interleave` over the wire — its handler issues an
+  outbound hub→consumer call on the same connection while the Sub is
+  live — and asserts the call resolves and the Sub is still live
+  after. Both gates bounded at 5s (vs. the 30s sweeper), so a
+  regression fails fast, not via sweeper eviction.
+
+**Load-bearing verification (both gates, empirical):** each gate was
+run against the pre-fix loop (`git stash push
+src/protocol/dispatch.rs`) and reproduced the G-01 hang:
+
+```
+hub_handler_composes…:  panicked "nested composition through the
+  forwarding stub timed out (G-01 shape): Elapsed(())"  — 5.01s, no progress
+outbound_call_resolves…: panicked "interleaved outbound call starved
+  while Sub served (G-01 shape): Elapsed(())"  — 5.06s, no progress
+```
+
+With the fix, both resolve (0.01s / 0.11s wall including connection
+setup). No sweeper evictions (bounded timeouts ≪ 30s).
+
+**Consequences audited against the fix:**
+
+- Same-connection nested composition of peer-announced ops works for
+  wire-dispatched parents (the flagship ADR-022 amendment flow) — the
+  gate proves it end-to-end.
+- The accept-side imported-op composition hazard (G-01's second
+  consequence, latent pre-`f84d214`) is mechanized shut by the
+  pending-resolution arms in `run_loop_single_stream` — not merely
+  unblocked. There is no dedicated gate for this shape yet (the
+  accept-side import + nested-compose e2e would be a further gate;
+  noted as residual work, not a regression).
+- Head-of-line blocking is gone: spawned arms proceed concurrently;
+  the loop never awaits a handler.
+- Frame ordering: per-request ordering is preserved by the
+  `SharedFrameWriter` (each `write_frame` is atomic under the mutex);
+  responses for two requests may now interleave *at frame
+  granularity*, which single-stream mode always permitted (both
+  directions' frames are multiplexed by design; correlation is by id).
+
+**Verification (post-fix):**
+
+```
+cargo test                                        → 583 passed, 0 failed
+cargo test --all-features                         → 600 passed, 0 failed
+cargo clippy --all-targets -- -D warnings         → clean
+cargo clippy --all-features --all-targets -- -D warnings → clean
+cargo fmt --check                                 → clean (fmt applied)
+cargo clippy --target wasm32-unknown-unknown -- -D warnings → clean
+cargo doc --no-deps                               → clean
+```
+
+**Residual (not blocking Unit 1):** the sink-start-inline shape means a
+`Pub` op whose registration itself blocks (ACL/schema compile are sync
+and fast; `resolve_sink_handler` runs the handler's ACL path) still
+holds the loop briefly — bounded by registry work, not handler work. A
+malicious `AccessControl::check` implementation could stall the loop;
+that is a deployment-provided trait object, the same trust boundary as
+`IdentityProvider`, and unchanged from the pre-existing shape.
+
+---
+
+## Verification log (this pass)
+
+- All gates in Baseline verification reproduced at tree `f84d214`
+  (clean working tree; the probe was added, run, and removed —
+  `git status` verified clean after removal).
+- G-01 verified by code trace (the four steps above, each line
+  read at `f84d214`) **and** empirically: probe
+  `probe_hub_handler_composes_peer_announced_op` (in
+  `src/channels/client.rs#[cfg(test)]`, since removed) — hub registry
+  carries `op/register` + `hub/compose` (whose handler invokes the
+  connection overlay's registration for the announced op, the shape
+  `PeerCompositeEnv` → `OverlayOperationEnv` resolution produces);
+  consumer announces `consumer/exec` then calls `hub/compose`.
+  8s-timeout run: `Elapsed` (no progress). 45s-cap run: resolved at
+  **30.0007s** with `CallError::TIMEOUT` (retryable) — the
+  `DEFAULT_CALL_TIMEOUT` sweeper eviction, not live resolution.
+- The existing gates' blind spot verified by reading both e2e gates'
+  call directions: `serving_loop_hub_to_consumer_call_resolves`
+  (`client.rs:1373-1399`) and
+  `op_register_announce_then_hub_call_routes_back_to_consumer`
+  (`client.rs:1529`) — the latter's `accept_conn.call("consumer/exec")`
+  crosses the wire to the consumer's local registry; the forwarding
+  stub is unreachable from it.
+- G-03 verified by reading the collision gate (`op_register.rs:142`)
+  against `PeerCompositeEnv::invoke_with_policy`'s resolution order
+  (`env.rs:230-245`) and `compose_root_env`'s attach
+  (`dispatch.rs:207-214`).
+- G-04 verified by field audit: `OperationSpec`'s fields
+  (`spec.rs:175-206`) vs `spec_to_json_pub` (`discovery.rs:211-234`)
+  vs `rebuild_spec_for` (`from_call.rs:209-286`).
+- G-05 verified against `discovery.rs:290-313` and ADR-022
+  amendment's set (`022-…contract.md:380-393`).
+- Frame-arm equivalence of `serve_single_stream` vs the two half-loops
+  verified read-through (Non-findings).
+- alkhttp cross-repo claims verified in `/workspace/@alkdev/alkhttp`
+  at `5b62307` (OQ-05 `open-questions.md:102-112`; ADR-048
+  `decisions/048-…md:19-25`).
+- CJK sweep: no CJK-range codepoints in any `.rs`/`.md`/`.toml`/
+  `.json` under the repo at `f84d214`.

docs/reviews/consumer-findings-ledger.md

@@ -0,0 +1,111 @@
+# Consumer findings ledger — alkhttp consuming alkcall
+
+Findings from writing the first real consumer (alkhttp) against alkcall.
+alkcall is deliberately minimal; expect missing features and edge-case
+bugs to surface here first. Extract items into alkcall tasks/reviews as
+the project sees fit.
+
+Format: date | found-in (alkhttp context) | severity | status.
+
+---
+
+## Open
+
+(none)
+
+---
+
+## Resolved
+
+### CF-004 — `services_schema_handler` discloses Internal/ACL-restricted op specs — no visibility or AccessControl check (2026-08-30) — RESOLVED 2026-08-31
+
+- **Found in:** alkhttp Review 002, finding PRJ-16
+  (`alkhttp/docs/reviews/002-post-remediation-review.md`, Part F',
+  verified against tree `91483a7` + alkcall source). Filed to this
+  ledger 2026-08-30.
+  `src/registry/discovery.rs:327-343` (`services_schema_handler`) —
+  the handler did a bare `registry.registration(&name)` and returned
+  `spec_to_json(&reg.spec)` verbatim, with **no `Visibility` check and
+  no `AccessControl::check(peer_identity)`**.
+- **Fix:** the handler now applies the same two gates as
+  `OperationRegistry::invoke` — the Internal-visibility rejection
+  (`!ctx.internal` → spec-404) and `AccessControl::check` with the same
+  identity resolution invoke() uses (`handler_identity` when internal,
+  `identity` otherwise). Restricted ops return `NOT_FOUND` (spec-404)
+  rather than `FORBIDDEN` — no information leaks about the restricted
+  surface's existence or shape. The gate is in the handler itself, so
+  every transport (wire `/call`, HTTP routes, MCP tools) is covered.
+- **Status:** resolved — 2026-08-30.
+
+### CF-003 — `publish_schema` compile failure is fail-open on the wire dispatch path (2026-08-30) — RESOLVED 2026-08-30
+
+- **Found in:** alkhttp Review 001 post-remediation task
+  `review-001-publish-schema-validation-robust` (the GW-01/#1 follow-up
+  work). While fixing alkhttp's HTTP-side instance of the pattern, the
+  identical instance was verified on the wire dispatch path:
+  `src/protocol/dispatch.rs:352-366` (`Dispatcher::dispatch`,
+  `OperationType::Pub` arm) — when a Pub op's `publish_schema` failed to
+  **compile**, the pump logged a warn and proceeded with
+  `validator: None` (silent unvalidated ingest, plus per-request
+  recompile cost).
+- **Fix:** fail-closed at the source — `publish_schema` is now compiled
+  at **registration time** in both insertion points
+  (`OperationRegistry::register` and `OperationRegistryBuilder::store`);
+  an un-compilable schema is a registration error, so an
+  unvalidated-ingest path can never be constructed. The compiled
+  validator is cached per-op (`OperationRegistry::publish_validator`)
+  and the dispatch path consumes the cache — the per-request compile is
+  gone. The `from_call` import path inherits the guarantee (imports
+  register through the same choke point); forwarded-chunk validation on
+  the producer side is the remote peer's dispatch path and now inherits
+  the same fail-closed property.
+- **Behavior change (semver-relevant):** `register`/builder methods
+  return `Err` for an un-compilable `publish_schema` (previously:
+  accepted). Code that registered garbage schemas will now get a
+  registration error instead of a silently-unvalidated op.
+- **Status:** resolved — 2026-08-30.
+
+### CF-002 — demux `TooLarge` skip allocates the full peer-declared length up front (up to ~4 GiB from an 8-byte header) (2026-08-30) — RESOLVED 2026-08-30
+
+- **Found in:** alkhttp Review 001, finding WS-12 (cross-crate;
+  `docs/reviews/001-initial-implementation-review.md`, Part B,
+  verified against tree `4a825d3`). Filed to this ledger 2026-08-30.
+  `src/channels/adapter.rs:144` (`run_demux_loop_for_client`) — the
+  `ChunkError::TooLarge` arm allocated
+  `let mut discard = vec![0u8; length as usize];` **before** reading:
+  the buffer was sized from the peer's untrusted 8-byte header (`u32`
+  `length`, up to ~4 GiB).
+- **Fix:** stream-skip with a fixed 64 KiB buffer + `read_exact` loop
+  consuming `length` bytes — no allocation sized from peer input.
+  Plus a cumulative skipped-bytes budget (256 MiB) that tears down the
+  connection when a peer loops `TooLarge` headers to burn bandwidth/CPU.
+  The budget resets on each valid chunk, so legitimate isolated
+  oversized chunks (the resync path, covered by the existing test) are
+  unaffected. The existing `demux_resyncs_after_oversized_chunk` test
+  passes unchanged.
+- **Status:** resolved — 2026-08-30.
+
+### CF-001 — `call_single_stream` write-failure maps to non-retryable `INTERNAL` (2026-08-29) — RESOLVED 2026-08-30
+
+- **Found in:** alkhttp `from_wss` drop-monitor race tests (WS-02/CON-02,
+  review-001-ws-eof-signal). When the transport mux died mid-call, the
+  write path failed fast and the write-failure mapping produced a
+  non-retryable `INTERNAL: failed to write request frame` — even though
+  the call never reached the producer and a reconnect/retry would be
+  safe.
+- **Fix:** new retryable `CallError::connection_closed` constructor
+  (`CONNECTION_CLOSED` code, `retryable: true`), applied **only** to
+  write failures where the call is provably undelivered:
+  `call.requested` frame write failures on all consumer paths
+  (`call`/`subscribe`/`publish`, single-stream and stream-per-request;
+  the publish pump tags its write stages so only the request-frame
+  stage is retryable). Mid-publish and completed-frame write failures
+  stay `INTERNAL` (delivery ambiguous — retry unsafe), and the
+  producer-side `fail_all(...)` on connection close stays `INTERNAL`.
+  The new code string is additive to the wire error vocabulary;
+  `CONNECTION_CLOSED` is new — consumers should treat unknown codes per
+  their existing policy (the `retryable` flag is the machine-readable
+  signal).
+- **Status:** resolved — 2026-08-30. alkhttp follow-up: the
+  `review-001-ws-eof-signal` race test can now tighten back to
+  retryable-only asserts.

src/channels/adapter.rs

@@ -128,6 +128,15 @@ impl ChannelsAdapter {
         reader: Box<dyn tokio::io::AsyncRead + Send + Unpin>,
         policy: Option<&Arc<dyn ChannelLifecyclePolicy>>,
     ) {
+        // Stack buffer for skipping oversized payloads — sized once, never
+        // from peer-declared lengths (a peer-declared `length` is untrusted
+        // up to u32::MAX ≈ 4 GiB; pre-sizing off it is an allocation-shape
+        // DoS). A cumulative cap tears down dribbling peers that loop
+        // TooLarge headers to burn bandwidth indefinitely.
+        const SKIP_BUF_LEN: usize = 64 * 1024;
+        const MAX_CONSECUTIVE_SKIPPED_BYTES: u64 = 256 * 1024 * 1024;
+        let mut skip_buf = vec![0u8; SKIP_BUF_LEN];
+        let mut skipped_since_last_valid: u64 = 0;
         let mut reader = reader;
         let mut header_buf = [0u8; CHUNK_HEADER_LEN];
         loop {
@@ -141,10 +150,23 @@ impl ChannelsAdapter {
                                 max = super::wire::MAX_CHUNK_LEN,
                                 "demux: chunk too large, skipping payload bytes"
                             );
-                            let mut discard = vec![0u8; length as usize];
-                            if let Err(e) = reader.read_exact(&mut discard).await {
-                                warn!(error = %e, "demux: failed to skip oversized payload");
-                                break;
+                            let mut remaining = length as u64;
+                            while remaining > 0 {
+                                let take = remaining.min(SKIP_BUF_LEN as u64) as usize;
+                                if let Err(e) = reader.read_exact(&mut skip_buf[..take]).await {
+                                    warn!(error = %e, "demux: failed to skip oversized payload");
+                                    return;
+                                }
+                                remaining -= take as u64;
+                            }
+                            skipped_since_last_valid += length as u64;
+                            if skipped_since_last_valid > MAX_CONSECUTIVE_SKIPPED_BYTES {
+                                warn!(
+                                    skipped_since_last_valid,
+                                    max = MAX_CONSECUTIVE_SKIPPED_BYTES,
+                                    "demux: skipped-bytes budget exceeded; tearing down connection"
+                                );
+                                return;
                             }
                             continue;
                         }
@@ -170,6 +192,7 @@ impl ChannelsAdapter {
                         }
                     };
                     manager.route_payload(header.channel_id, payload).await;
+                    skipped_since_last_valid = 0;
                 }
                 Err(e) => {
                     if e.kind() == std::io::ErrorKind::UnexpectedEof {
@@ -529,4 +552,142 @@ mod tests {
             assert_eq!(buf, [i; 4], "channel A chunk {i} intact — no data loss");
         }
     }
+
+    /// CF-002 #1 — the skipped-bytes budget tears down a dribbling peer.
+    /// Sixteen back-to-back `TooLarge` headers each declare 16 MiB + 1 and
+    /// their payloads are genuinely on the wire (the demux must consume
+    /// every byte to reach the next header, so the test cannot avoid
+    /// moving ~268 MiB through the duplex). The 16th skip pushes the
+    /// cumulative total past the 256 MiB budget: a working demux tears
+    /// down and *stops reading there*, while a peer that keeps dribbling
+    /// (a 17th header parked mid-payload) must not wedge it. Without the
+    /// budget the demux blocks skipping the 17th payload forever and the
+    /// timeout fires.
+    #[tokio::test]
+    async fn demux_tears_down_after_cumulative_skipped_bytes_budget_exceeded() {
+        let oversized_len = super::super::wire::MAX_CHUNK_LEN as usize + 1;
+        let burst_count = 16;
+        let (client, server) = tokio::io::duplex(1024 * 1024);
+        let (server_read, server_write) = tokio::io::split(server);
+        let (mux_handle, mux_runner) = MuxRunner::new(Box::new(server_write));
+        let _mux_task = tokio::spawn(async move {
+            let _ = mux_runner.run().await;
+        });
+        let manager = ChannelManager::with_defaults(mux_handle, None);
+
+        let (_id, _send, _recv) = manager
+            .open_channel("alk/tty", "alice", None)
+            .await
+            .expect("open");
+
+        let demux_manager = manager.clone();
+        let demux_task = tokio::spawn(async move {
+            ChannelsAdapter::run_demux_loop_for_client(&demux_manager, Box::new(server_read), None)
+                .await;
+        });
+
+        // Writer task: 16 full oversized payloads (15 × ~16 MiB ≈ 240 MiB
+        // skip cleanly under the budget; the 16th crosses it), then a 17th
+        // header + partial payload — and park instead of finishing (the
+        // peer keeps dribbling, never EOFs).
+        let writer_task = tokio::spawn(async move {
+            use tokio::io::AsyncWriteExt;
+            let mut client_write = client;
+            let mut header = [0u8; 8];
+            let chunk = vec![0u8; 64 * 1024];
+            let full = oversized_len / chunk.len();
+            let tail = oversized_len % chunk.len();
+            for _ in 0..burst_count {
+                super::super::wire::write_header(1, oversized_len as u32, &mut header)
+                    .expect("write header");
+                client_write.write_all(&header).await.expect("write header");
+                for _ in 0..full {
+                    client_write.write_all(&chunk).await.expect("write payload");
+                }
+                if tail > 0 {
+                    client_write
+                        .write_all(&chunk[..tail])
+                        .await
+                        .expect("write payload tail");
+                }
+            }
+            // The budget is already blown by the 16th skip; this header is
+            // the dribble that a budgetless demux would park on forever.
+            super::super::wire::write_header(1, oversized_len as u32, &mut header)
+                .expect("write header");
+            client_write.write_all(&header).await.expect("write header");
+            client_write
+                .write_all(&chunk[..8192])
+                .await
+                .expect("write partial payload");
+            std::future::pending::<()>().await;
+        });
+
+        let demux_done =
+            tokio::time::timeout(std::time::Duration::from_secs(120), demux_task).await;
+        assert!(
+            demux_done.is_ok(),
+            "demux must tear down on the skipped-bytes budget, not block in the skip loop"
+        );
+        writer_task.abort();
+    }
+
+    /// CF-002 #2 — a skip that hits EOF mid-payload (peer declared more
+    /// bytes than it sent) ends the demux loop; nothing routes to the
+    /// channel and the loop does not spin back to header parsing on a
+    /// truncated stream.
+    #[tokio::test]
+    async fn demux_ends_when_oversized_payload_skip_hits_eof() {
+        let (client, server) = tokio::io::duplex(1024 * 1024);
+        let (server_read, server_write) = tokio::io::split(server);
+        let (mux_handle, mux_runner) = MuxRunner::new(Box::new(server_write));
+        let _mux_task = tokio::spawn(async move {
+            let _ = mux_runner.run().await;
+        });
+        let manager = ChannelManager::with_defaults(mux_handle, None);
+
+        let (_id, _send, mut recv) = manager
+            .open_channel("alk/tty", "alice", None)
+            .await
+            .expect("open");
+
+        let demux_manager = manager.clone();
+        let demux_task = tokio::spawn(async move {
+            ChannelsAdapter::run_demux_loop_for_client(&demux_manager, Box::new(server_read), None)
+                .await;
+        });
+
+        use tokio::io::AsyncWriteExt;
+        let oversized_len = super::super::wire::MAX_CHUNK_LEN as usize + 1;
+        let mut header = [0u8; 8];
+        super::super::wire::write_header(1, oversized_len as u32, &mut header).expect("header");
+        let mut client_write = client;
+        client_write.write_all(&header).await.expect("write header");
+        client_write
+            .write_all(&[0u8; 4096])
+            .await
+            .expect("write partial payload");
+        drop(client_write);
+
+        let demux_done = tokio::time::timeout(std::time::Duration::from_secs(10), demux_task).await;
+        assert!(
+            demux_done.is_ok(),
+            "demux must end when the skip hits EOF instead of looping"
+        );
+
+        // The oversized chunk was never routed — nothing to assert on the
+        // channel beyond the loop ending; reading would surface EOF since
+        // the demux dropped its sender without delivering a payload.
+        use tokio::io::AsyncReadExt;
+        let mut buf = [0u8; 4];
+        let read = tokio::time::timeout(
+            std::time::Duration::from_millis(500),
+            recv.read_exact(&mut buf),
+        )
+        .await;
+        assert!(
+            read.is_err() || read.unwrap().is_err(),
+            "no payload must arrive after the truncated skip"
+        );
+    }
 }

src/channels/env.rs

@@ -83,6 +83,10 @@ impl OperationEnv for ChannelsSessionEnv {
         self.base.peer_operations(peer)
     }
 
+    fn list_operation_names(&self) -> Vec<String> {
+        self.base.list_operation_names()
+    }
+
     async fn invoke_peer(
         &self,
         peer: &crate::registry::env::PeerRef,

src/channels/manager.rs

@@ -7,7 +7,7 @@
 //! `ChannelsAdapter`, the open-op wrapper, and relay logic can all
 //! hold a handle.
 
-use std::collections::HashMap;
+use std::collections::{HashMap, VecDeque};
 use std::net::SocketAddr;
 use std::sync::atomic::{AtomicU32, AtomicU64, Ordering};
 use std::sync::Arc;
@@ -93,8 +93,23 @@ struct Inner {
     remote_addr: Option<SocketAddr>,
     side: ChannelSide,
     dropped_unknown_chunks: AtomicU64,
+    /// Chunks that arrived for a channel_id this side has not
+    /// adopted yet (the open-op response carrying `channel_id` and the
+    /// producer's first data-plane write race — the handler can start
+    /// pumping before the connect side's `adopt_channel` runs). Each
+    /// channel buffers up to `early_arrival_cap` payloads; `adopt_channel`
+    /// drains them into the new receiver. `clear_all` drops the map.
+    early_arrivals: Mutex<HashMap<u32, VecDeque<Bytes>>>,
+    early_arrival_count: AtomicU64,
 }
 
+/// The per-channel cap on parked early-arrival chunks (`route_payload`
+/// parks chunks for not-yet-adopted channels instead of dropping them —
+/// the open-op response/first-data race). A channel whose adopter never
+/// arrives leaks its parked chunks until `clear_all`; the cap bounds
+/// that leak per channel.
+const EARLY_ARRIVAL_CAP: usize = 64;
+
 impl ChannelManager {
     /// Construct a new manager with the given `MuxHandle`, max
     /// channels, buffer cap, and side. The `remote_addr` is
@@ -124,6 +139,8 @@ impl ChannelManager {
                 remote_addr,
                 side,
                 dropped_unknown_chunks: AtomicU64::new(0),
+                early_arrivals: Mutex::new(HashMap::new()),
+                early_arrival_count: AtomicU64::new(0),
             }),
         }
     }
@@ -232,7 +249,7 @@ impl ChannelManager {
         let (demux_sender, recv) = MpscRecvStream::channel(self.inner.buffer_cap);
 
         let state = ChannelState {
-            demux_sender,
+            demux_sender: demux_sender.clone(),
             handler_task,
             alpn,
         };
@@ -288,7 +305,7 @@ impl ChannelManager {
         let (demux_sender, recv) = MpscRecvStream::channel(self.inner.buffer_cap);
 
         let state = ChannelState {
-            demux_sender,
+            demux_sender: demux_sender.clone(),
             handler_task,
             alpn,
         };
@@ -305,6 +322,10 @@ impl ChannelManager {
             }
         }
 
+        // Route any chunks that arrived between the open-op response and
+        // this adopt (the producer may have started pumping already).
+        self.drain_early_arrivals(channel_id, &demux_sender).await;
+
         Ok((send, recv))
     }
 
@@ -377,8 +398,15 @@ impl ChannelManager {
     /// Route a chunk payload to the reassembled stream for
     /// `channel_id` (the demux's per-chunk route). A zero-length
     /// payload is the EOF sentinel — the reassembled stream interprets
-    /// it as EOF (REQ-CH-01). An unknown `channel_id` is dropped with
-    /// a debug log (REQ-CH-04 — lenient handling).
+    /// it as EOF (REQ-CH-01). An unknown `channel_id` parks the payload
+    /// in the early-arrival buffer (up to `EARLY_ARRIVAL_CAP` payloads
+    /// per channel) — the open-op response carrying `channel_id` and
+    /// the producer's first data-plane write race, and a push-first
+    /// producer (a TTY backend's banner, a sub protocol's greeting)
+    /// must not lose its first chunks. `adopt_channel` drains the
+    /// parked payloads into the new receiver. Chunks beyond the cap
+    /// drop with a debug log (the adopted-side REQ-CH-04 leniency for
+    /// genuinely unknown channels).
     ///
     /// Awaits the bounded channel sender — if the handler's read half
     /// is slow, the demux stalls here (ADR-040 REQ-CH-05: lossless
@@ -400,17 +428,67 @@ impl ChannelManager {
                 }
             }
             None => {
-                self.inner
-                    .dropped_unknown_chunks
-                    .fetch_add(1, Ordering::Relaxed);
-                debug!(
-                    channel_id,
-                    "demux: unknown channel_id, dropping chunk (lenient)"
-                );
+                self.park_early_arrival(channel_id, payload);
             }
         }
     }
 
+    /// Park an early-arrival chunk for `channel_id` (the adopter has
+    /// not run yet). FIFO per channel, bounded by `EARLY_ARRIVAL_CAP`;
+    /// beyond the cap the chunk is dropped with a debug log and the
+    /// dropped counter increments (observability only).
+    fn park_early_arrival(&self, channel_id: u32, payload: Bytes) {
+        let mut arrivals = self.inner.early_arrivals.lock();
+        let queue = arrivals.entry(channel_id).or_default();
+        if queue.len() >= EARLY_ARRIVAL_CAP {
+            drop(arrivals);
+            self.inner
+                .dropped_unknown_chunks
+                .fetch_add(1, Ordering::Relaxed);
+            debug!(
+                channel_id,
+                "demux: unknown channel_id, early-arrival buffer full, dropping chunk"
+            );
+            return;
+        }
+        queue.push_back(payload);
+        self.inner
+            .early_arrival_count
+            .fetch_add(1, Ordering::Relaxed);
+        debug!(
+            channel_id,
+            "demux: unknown channel_id, parking chunk (early arrival, not yet adopted)"
+        );
+    }
+
+    /// Drain the early-arrival buffer for `channel_id` into `recv`'s
+    /// sender (called by `adopt_channel` after the channel state is
+    /// installed). Returns the drained payload count. Payloads past the
+    /// receiver's buffer capacity await normally (backpressure).
+    async fn drain_early_arrivals(
+        &self,
+        channel_id: u32,
+        demux_sender: &mpsc::Sender<Bytes>,
+    ) -> usize {
+        let payloads = self
+            .inner
+            .early_arrivals
+            .lock()
+            .remove(&channel_id)
+            .unwrap_or_default();
+        let count = payloads.len();
+        for payload in payloads {
+            if demux_sender.send(payload).await.is_err() {
+                debug!(
+                    channel_id,
+                    "demux: channel receiver dropped while draining early arrivals"
+                );
+                break;
+            }
+        }
+        count
+    }
+
     /// Take the sender for `channel_id` — used on close / teardown to
     /// drop the sender (which signals EOF to the handler, REQ-CH-02).
     /// Returns the handler task (if any) so the caller can abort it
@@ -447,6 +525,8 @@ impl ChannelManager {
                 task.abort();
             }
         }
+        // The early-arrival buffers die with the connection too.
+        self.inner.early_arrivals.lock().clear();
         // The opener ledger has the opener PeerIds.
         self.inner.opener_ledger.drain()
     }
@@ -582,20 +662,53 @@ mod tests {
             .await;
     }
 
+    /// Early arrivals for a not-yet-adopted channel are parked, not
+    /// dropped: `adopt_channel` drains them into the new receiver. This
+    /// closes the open-op-response/first-data race (a push-first
+    /// producer's first chunks must not be lost).
+    #[tokio::test]
+    async fn route_payload_to_unknown_channel_parks_until_adopt() {
+        let manager = make_manager_with_runner().await;
+        manager.route_payload(7, Bytes::from_static(b"first")).await;
+        manager
+            .route_payload(7, Bytes::from_static(b"second"))
+            .await;
+
+        let (_send, mut recv) = manager
+            .adopt_channel(7, "alk/tty", None)
+            .await
+            .expect("adopt");
+
+        use tokio::io::AsyncReadExt;
+        let mut buf = [0u8; 11];
+        recv.read_exact(&mut buf).await.expect("read first+second");
+        assert_eq!(&buf, b"firstsecond");
+    }
+
     #[tokio::test]
     async fn route_payload_to_unknown_channel_increments_dropped_counter() {
         let manager = make_manager_with_runner().await;
         assert_eq!(manager.dropped_unknown_chunks(), 0);
+        // Parked arrivals don't increment the counter...
         manager
-            .route_payload(999, Bytes::from_static(b"data"))
-            .await;
-        manager
-            .route_payload(998, Bytes::from_static(b"more"))
+            .route_payload(998, Bytes::from_static(b"kept"))
             .await;
         assert_eq!(
             manager.dropped_unknown_chunks(),
-            2,
-            "dropped_unknown_chunks counter increments per unknown-channel drop"
+            0,
+            "parked early arrivals are not drops"
+        );
+        // ...but chunks beyond the per-channel cap do (the adopted-side
+        // REQ-CH-04 leniency for genuinely unknown channels).
+        for i in 0..(super::EARLY_ARRIVAL_CAP + 1) {
+            let n = i;
+            let payload = vec![b'x'; n % 9 + 1];
+            manager.route_payload(999, Bytes::from(payload)).await;
+        }
+        assert_eq!(
+            manager.dropped_unknown_chunks(),
+            1,
+            "dropped_unknown_chunks counter increments when the early-arrival buffer overflows"
         );
     }
 

src/channels/operations.rs

@@ -62,7 +62,7 @@ impl ChannelOperations {
     /// Register the three generic ops on the call `OperationRegistry`.
     /// The per-ALPN open ops are registered separately by the ALPN
     /// crates via [`ChannelCore::register_openable`] (ADR-047 §3).
-    pub fn register_on(&self, registry: &mut OperationRegistry) -> Result<(), String> {
+    pub fn register_on(&self, registry: &OperationRegistry) -> Result<(), String> {
         let manager = self.manager.clone();
         let policy = Arc::clone(&self.policy);
         registry.register(HandlerRegistration::new(
@@ -400,7 +400,7 @@ impl ChannelCore {
         &self,
         spec: OperationSpec,
         open_handler: OpenHandler,
-        registry: &mut OperationRegistry,
+        registry: &OperationRegistry,
         auth: AuthContext,
     ) -> Result<(), String> {
         let op_type = spec.op_type;
@@ -765,8 +765,8 @@ mod tests {
     async fn register_on_registers_three_ops() {
         let manager = make_manager().await;
         let ops = ChannelOperations::with_default_policy(manager);
-        let mut registry = OperationRegistry::new();
-        ops.register_on(&mut registry).expect("register");
+        let registry = OperationRegistry::new();
+        ops.register_on(&registry).expect("register");
         assert!(registry.registration(OP_CHANNEL_CLOSE).is_some());
         assert!(registry.registration(OP_CHANNEL_CONTROL).is_some());
         assert!(registry
@@ -899,8 +899,8 @@ mod tests {
         let manager = make_manager().await;
         let policy = super::super::policy::default_policy();
         let ops = ChannelOperations::new(manager, policy);
-        let mut registry = OperationRegistry::new();
-        ops.register_on(&mut registry).expect("register");
+        let registry = OperationRegistry::new();
+        ops.register_on(&registry).expect("register");
         assert!(registry.registration(OP_CHANNEL_CLOSE).is_some());
     }
 
@@ -1129,11 +1129,11 @@ mod tests {
             spawned_clone.store(true, Ordering::SeqCst);
             tokio::spawn(async {})
         });
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         core.register_openable(
             spec,
             open_handler,
-            &mut registry,
+            &registry,
             AuthContext::anonymous(b"alk/call"),
         )
         .expect("register");
@@ -1172,11 +1172,11 @@ mod tests {
             spawned_clone.store(true, Ordering::SeqCst);
             tokio::spawn(async {})
         });
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         core.register_openable(
             spec,
             open_handler,
-            &mut registry,
+            &registry,
             AuthContext::anonymous(b"alk/call"),
         )
         .expect("register");
@@ -1209,11 +1209,11 @@ mod tests {
         )
         .with_channel_open(ChannelOpenSpec::new("alk/tty"));
         let open_handler: OpenHandler = Arc::new(|_input, _conn, _auth| tokio::spawn(async {}));
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         core.register_openable(
             spec,
             open_handler,
-            &mut registry,
+            &registry,
             AuthContext::anonymous(b"alk/call"),
         )
         .expect("register");
@@ -1239,11 +1239,11 @@ mod tests {
             None,
         );
         let open_handler: OpenHandler = Arc::new(|_input, _conn, _auth| tokio::spawn(async {}));
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let result = core.register_openable(
             spec,
             open_handler,
-            &mut registry,
+            &registry,
             AuthContext::anonymous(b"alk/call"),
         );
         assert!(result.is_err());

src/channels/source.rs

@@ -4,7 +4,7 @@
 //!
 //! The handler for a channel receives a `Connection` constructed via
 //! `Connection::from_source(ChannelBidiStreamSource, alpn)`. The
-//! handler calls `accept_bi()` once (yield-once per channel, ADR-065)
+//! handler calls `accept_bi()` once (yield-once per channel, ADR-007)
 //! and gets a `BiStream` — identical to how it works on a top-level
 //! QUIC connection. The `BiStream` is the reassembled read half joined
 //! to the mux write half via `BiStream::from_joined`.
@@ -25,8 +25,8 @@ use crate::core::types::{BiStream, BidiStreamSource, StreamError};
 ///
 /// The `BiStream` is constructed once, in [`ChannelBidiStreamSource::new`],
 /// and held in an `Option` — `accept_bi` takes it. This preserves the
-/// yield-once contract (ADR-065) and the "split never crosses a crate
-/// boundary as part of a constructor" rule (ADR-092) — the join happens
+/// yield-once contract (ADR-007) and the "split never crosses a crate
+/// boundary as part of a constructor" rule (ADR-009) — the join happens
 /// here, in the `BidiStreamSource` impl, not per-handler.
 pub struct ChannelBidiStreamSource {
     stream: Mutex<Option<BiStream>>,
@@ -65,9 +65,9 @@ impl BidiStreamSource for ChannelBidiStreamSource {
 
     /// `code`/`reason` are ignored: a single channel has no
     /// QUIC-shaped application-level close codes. The drop is the
-    /// close (ADR-065 §"Negative"). The `_` prefix is intentional —
+    /// close (ADR-007 §"Negative"). The `_` prefix is intentional —
     /// the signature matches the public `Connection::close` API
-    /// (ADR-070 §"REQ-CORE-02").
+    /// (ADR-008 §"REQ-CORE-02").
     fn close(&self, _code: u32, _reason: &str) {
         let _ = self.stream.lock().take();
     }

src/channels/wire.rs

@@ -23,7 +23,7 @@ use thiserror::Error;
 pub const CHUNK_HEADER_LEN: usize = 8;
 
 /// The maximum chunk payload length (16 MiB, matching TTY's cap —
-/// ADR-052 §5). A chunk with `length > MAX_CHUNK_LEN` returns
+/// ADR-034). A chunk with `length > MAX_CHUNK_LEN` returns
 /// [`ChunkError::TooLarge`] and does not corrupt the stream — the demux
 /// drops the chunk and continues. The header is always exactly 8 bytes,
 /// so the demux can always resync by reading the next 8-byte header.

src/client/call_client.rs

@@ -122,7 +122,7 @@ mod tests {
     }
 
     fn registry_with_caps() -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 external_spec("pub/run"),

src/client/from_call.rs

@@ -202,7 +202,11 @@ async fn fetch_schema(connection: &CallConnection, name: &str) -> Result<Value,
 /// Rebuild an `OperationSpec` from the `services/schema` JSON, applying the
 /// optional namespace prefix. The spec JSON shape matches `spec_to_json` in
 /// `registry/discovery.rs`.
-fn rebuild_spec_for(
+///
+/// `pub(crate)` so the `op/register` bootstrap path (review 004 F-05) can
+/// rebuild peer-announced specs from the same wire shape — one parser, two
+/// consumers.
+pub(crate) fn rebuild_spec_for(
     schema_json: &Value,
     remote_name: &str,
     namespace_prefix: &Option<String>,
@@ -252,7 +256,10 @@ fn rebuild_spec_for(
         output_schema,
         error_schemas,
         access_control,
-        None,
+        schema_json
+            .get("resource_id_path")
+            .and_then(|v| v.as_str())
+            .map(String::from),
     );
 
     // ADR-047 §2: the `channel_open` marker survives discovery
@@ -385,7 +392,10 @@ fn parse_access_control(v: &Value) -> AccessControl {
 /// If `context.identity` is `None` (the hub chose not to disclose, or has not
 /// authenticated an originator), `forwarded_for` is omitted — the spoke
 /// receives only the hub's identity.
-fn make_forwarding_handler(connection: Arc<CallConnection>, remote_name: String) -> Handler {
+pub(crate) fn make_forwarding_handler(
+    connection: Arc<CallConnection>,
+    remote_name: String,
+) -> Handler {
     use crate::registry::registration::make_handler;
     make_handler(move |input, context| {
         let connection = Arc::clone(&connection);
@@ -411,7 +421,7 @@ fn make_forwarding_handler(connection: Arc<CallConnection>, remote_name: String)
 /// leaf: on invocation, calls `CallConnection::subscribe_with_payload()` and
 /// forwards the remote stream end-to-end. Each `call.responded` from the
 /// remote becomes a stream item, `call.completed` ends the stream, and
-/// `call.aborted` drops it (ADR-049 §8). No truncation, no first-value
+/// `call.aborted` drops it (ADR-021 §8). No truncation, no first-value
 /// fallback.
 ///
 /// `forwarded_for` is populated from `context.identity` (ADR-032 §3), exactly
@@ -421,7 +431,7 @@ fn make_forwarding_handler(connection: Arc<CallConnection>, remote_name: String)
 /// `PendingRequestMap`, so the abort cascade (ADR-016 §6) is already wired:
 /// a parent abort drops the `SubscriptionStream`, which sends `call.aborted`
 /// to the remote node.
-fn make_streaming_forwarding_handler(
+pub(crate) fn make_streaming_forwarding_handler(
     connection: Arc<CallConnection>,
     remote_name: String,
 ) -> StreamingHandler {
@@ -455,7 +465,7 @@ fn make_streaming_forwarding_handler(
 /// `forwarded_for` is populated from `context.identity` (ADR-032 §3),
 /// exactly as the request/response and streaming forwarding handlers
 /// do — both via `build_forwarded_payload`.
-fn make_sink_forwarding_handler(
+pub(crate) fn make_sink_forwarding_handler(
     connection: Arc<CallConnection>,
     remote_name: String,
 ) -> SinkHandler {
@@ -480,7 +490,11 @@ fn make_sink_forwarding_handler(
 /// `forwarded_for` from the hub's `OperationContext.identity` (ADR-032 §3).
 /// `forwarded_for` is omitted when `context.identity` is `None` (the hub
 /// chooses not to disclose the originator).
-fn build_forwarded_payload(operation_id: &str, input: Value, context: &OperationContext) -> Value {
+pub(crate) fn build_forwarded_payload(
+    operation_id: &str,
+    input: Value,
+    context: &OperationContext,
+) -> Value {
     let mut payload = serde_json::Map::new();
     payload.insert(
         "operationId".to_string(),
@@ -591,6 +605,67 @@ mod tests {
         assert_eq!(spec.access_control.resource_type.as_deref(), Some("fs"));
     }
 
+    // --- review 005 Unit 3 gate (G-04 spec wire round-trip) ----------------
+
+    /// G-04 gate: `resource_id_path` survives the spec wire round-trip —
+    /// `spec_to_json_pub` serializes it, `rebuild_spec_for` parses it.
+    /// The review found the field silently dropped: an announced (or
+    /// `from_call`-imported) op with ownership-scoped resource
+    /// extraction rebuilt with `resource_id: None`, so the rebuilt
+    /// spec's ACL checks ran without the resource ID.
+    #[test]
+    fn spec_round_trips_resource_id_path() {
+        use crate::registry::discovery::spec_to_json_pub;
+
+        let spec = OperationSpec::new(
+            "fs/readFile",
+            OperationType::Query,
+            Visibility::External,
+            json!({}),
+            json!({}),
+            vec![],
+            crate::registry::spec::AccessControl::default(),
+            Some("/path".to_string()),
+        );
+        let wire = spec_to_json_pub(&spec);
+        assert_eq!(
+            wire.get("resource_id_path").and_then(|v| v.as_str()),
+            Some("/path"),
+            "resource_id_path serialized"
+        );
+
+        let rebuilt = rebuild_spec_for(&wire, "fs/readFile", &None).expect("rebuild");
+        assert_eq!(
+            rebuilt.resource_id_path.as_deref(),
+            Some("/path"),
+            "resource_id_path survives the round-trip"
+        );
+    }
+
+    /// G-04 companion: a spec without `resource_id_path` serializes no
+    /// `resource_id_path` key and rebuilds with `None` (additive
+    /// optional field — absent stays absent).
+    #[test]
+    fn spec_without_resource_id_path_stays_absent_through_round_trip() {
+        use crate::registry::discovery::spec_to_json_pub;
+
+        let spec = OperationSpec::new(
+            "fs/readFile",
+            OperationType::Query,
+            Visibility::External,
+            json!({}),
+            json!({}),
+            vec![],
+            crate::registry::spec::AccessControl::default(),
+            None,
+        );
+        let wire = spec_to_json_pub(&spec);
+        assert!(wire.get("resource_id_path").is_none());
+
+        let rebuilt = rebuild_spec_for(&wire, "fs/readFile", &None).expect("rebuild");
+        assert_eq!(rebuilt.resource_id_path, None);
+    }
+
     #[test]
     fn rebuild_spec_channel_open_marker_set_for_channels_alpn_op() {
         let mut schema = sample_schema_json("channels/tty/sub", "sub");

src/client/mod.rs

@@ -10,6 +10,11 @@ mod from_call;
 pub use call_client::CallClient;
 pub use from_call::{from_call, FromCallConfig};
 
+// crate-internal surface for the `op/register` bootstrap path (review
+// 004 F-05): the forwarding-handler constructor and the spec wire
+// parser are shared with `registry::op_register`.
+pub(crate) use from_call::{make_forwarding_handler, rebuild_spec_for};
+
 use crate::registry::registration::HandlerRegistration;
 
 /// Errors produced by [`OperationAdapter::import`].

src/core/types.rs

@@ -227,14 +227,14 @@ pub trait ProtocolHandler: Send + Sync + 'static {
     async fn handle(&self, connection: Connection, auth: &AuthContext) -> Result<(), HandlerError>;
 }
 
-// --- BiStream: the handler leaf (ADR-092) ---------------------------------
+// --- BiStream: the handler leaf (ADR-009) ---------------------------------
 
 /// Internal helper trait — the union of `AsyncRead + AsyncWrite + Send`.
 /// Not public; exists only to give `BiStream` a single boxed field.
 trait AsyncReadWrite: AsyncRead + AsyncWrite + Send {}
 impl<T: AsyncRead + AsyncWrite + Send> AsyncReadWrite for T {}
 
-/// The handler leaf — a bidirectional byte stream (ADR-092).
+/// The handler leaf — a bidirectional byte stream (ADR-005, ADR-009).
 ///
 /// `accept_bi`/`open_bi` return a `BiStream`, not a split
 /// `(SendStream, RecvStream)` pair. Handlers that want the split halves call
@@ -316,7 +316,7 @@ impl AsyncWrite for BiStream {
     }
 }
 
-// --- SendStream / RecvStream: thin newtypes (ADR-092) ---------------------
+// --- SendStream / RecvStream: thin newtypes (ADR-009) ---------------------
 
 pub struct SendStream {
     inner: Box<dyn AsyncWrite + Send + Unpin>,
@@ -327,10 +327,11 @@ pub struct RecvStream {
 }
 
 impl SendStream {
-    /// Box a write half into the thin `SendStream` newtype. Used by
-    /// `into_sub_streams()` (ADR-074) and the channels reassembly path.
-    /// Not a constructor that feeds `Connection` — the split never crosses
-    /// a crate boundary as part of a constructor (ADR-092).
+    /// Box a write half into the thin `SendStream` newtype. Used by tests
+    /// and any consumer that holds pre-split halves (the channels layer
+    /// has its own `MpscSendStream`; `into_sub_streams()` was removed by
+    /// ADR-035). Not a constructor that feeds `Connection` — the split
+    /// never crosses a crate boundary as part of a constructor (ADR-009).
     pub fn from_stream(stream: impl AsyncWrite + Send + Unpin + 'static) -> Self {
         Self {
             inner: Box::new(stream),
@@ -339,10 +340,11 @@ impl SendStream {
 }
 
 impl RecvStream {
-    /// Box a read half into the thin `RecvStream` newtype. Used by
-    /// `into_sub_streams()` (ADR-074) and the channels reassembly path.
-    /// Not a constructor that feeds `Connection` — the split never crosses
-    /// a crate boundary as part of a constructor (ADR-092).
+    /// Box a read half into the thin `RecvStream` newtype. Used by tests
+    /// and any consumer that holds pre-split halves (the channels layer
+    /// has its own `MpscRecvStream`; `into_sub_streams()` was removed by
+    /// ADR-035). Not a constructor that feeds `Connection` — the split
+    /// never crosses a crate boundary as part of a constructor (ADR-009).
     pub fn from_stream(stream: impl AsyncRead + Send + Unpin + 'static) -> Self {
         Self {
             inner: Box::new(stream),
@@ -386,15 +388,15 @@ impl AsyncRead for RecvStream {
 
 /// Yield bidirectional streams to a `Connection`. Downstream crates implement
 /// this trait to add connection shapes (channels, a future transport, a test
-/// double beyond the single-stream case) without editing core. See ADR-070
-/// for the full rationale and ADR-065 for the yield-once contract the
+/// double beyond the single-stream case) without editing core. See ADR-008
+/// for the full rationale and ADR-007 for the yield-once contract the
 /// `StreamBidiStreamSource` impl preserves. The return type is `BiStream`
-/// (ADR-092) — the join happens once, in the impl, not per-handler.
+/// (ADR-009) — the join happens once, in the impl, not per-handler.
 #[async_trait]
 pub trait BidiStreamSource: Send + Sync + 'static {
     /// Yield the next bidirectional stream this connection provides.
     ///
-    /// Transport semantics (carried from ADR-065):
+    /// Transport semantics (carried from ADR-007):
     /// - QUIC (quinn/iroh): returns a new bidi stream on each call,
     ///   `ConnectionClosed` when the underlying connection closes.
     /// - Single-stream (TCP+TLS, SSH channel, WebTransport stream, wasm):
@@ -407,7 +409,7 @@ pub trait BidiStreamSource: Send + Sync + 'static {
     /// Open a bidirectional stream to the peer.
     ///
     /// Single-stream sources return `StreamClosed` (a single stream cannot
-    /// open new application streams — ADR-065). QUIC and channels sources
+    /// open new application streams — ADR-007). QUIC and channels sources
     /// open new streams.
     async fn open_bi(&self) -> Result<BiStream, StreamError>;
 
@@ -416,13 +418,13 @@ pub trait BidiStreamSource: Send + Sync + 'static {
 
     /// Close the connection. The `code`/`reason` args are QUIC application-
     /// level close codes; non-QUIC sources ignore them (the drop is the
-    /// close — ADR-065 §"Negative"). See ADR-070 §"REQ-CORE-02" for the
+    /// close — ADR-007 §"Negative"). See ADR-008 §"REQ-CORE-02" for the
     /// rationale for keeping the QUIC-shaped signature on the trait.
     fn close(&self, code: u32, reason: &str);
 }
 
 /// Single-stream `BidiStreamSource` (TCP+TLS, SSH channel, WebTransport
-/// stream, wasm stream — ADR-065). Crate-private; constructed via
+/// stream, wasm stream — ADR-007). Crate-private; constructed via
 /// `Connection::from_bidi`. `accept_bi` yields the underlying `BiStream`
 /// once, then `ConnectionClosed`; `open_bi` returns `StreamClosed`.
 struct StreamBidiStreamSource {
@@ -449,9 +451,9 @@ impl BidiStreamSource for StreamBidiStreamSource {
     }
 
     /// `code`/`reason` are ignored: a single stream has no QUIC-shaped
-    /// application-level close codes. The drop is the close (ADR-065
+    /// application-level close codes. The drop is the close (ADR-007
     /// §"Negative"). The `_` prefix is intentional — the signature matches
-    /// the public `Connection::close` API (ADR-070 §"REQ-CORE-02").
+    /// the public `Connection::close` API (ADR-008 §"REQ-CORE-02").
     fn close(&self, _code: u32, _reason: &str) {
         let _ = self.stream.lock().unwrap_or_else(|e| e.into_inner()).take();
     }
@@ -467,11 +469,11 @@ impl Connection {
     /// Construct a `Connection` from a single bidirectional stream (e.g.
     /// `tokio::io::DuplexStream`, `TlsStream<TcpStream>`,
     /// `russh::Channel::into_stream()`). The stream is wrapped in a
-    /// `BiStream` (ADR-092) and yielded by `accept_bi` once, then
+    /// `BiStream` (ADR-009) and yielded by `accept_bi` once, then
     /// `ConnectionClosed`. `open_bi` returns `StreamClosed` (a single
-    /// stream can't open new application streams — ADR-065).
+    /// stream can't open new application streams — ADR-007).
     ///
-    /// This is the only public stream constructor (ADR-092): the split
+    /// This is the only public stream constructor (ADR-009): the split
     /// never crosses a crate boundary as part of a constructor. Handlers
     /// that want the split halves call `tokio::io::split(&mut *stream)` on
     /// the `BiStream` they receive from `accept_bi`.
@@ -492,7 +494,7 @@ impl Connection {
 
     /// Construct from a caller-supplied `BidiStreamSource` impl. The
     /// extension point for downstream crates — implement the trait and
-    /// construct a `Connection` from it without editing core. See ADR-070.
+    /// construct a `Connection` from it without editing core. See ADR-008.
     pub fn from_source(source: impl BidiStreamSource, alpn: Vec<u8>) -> Self {
         Self {
             source: Box::new(source),
@@ -514,7 +516,7 @@ impl Connection {
     /// Handlers that loop `accept_bi` (e.g. `TtyAdapter`) get one session
     /// per single-stream connection; handlers that call once (e.g.
     /// `HttpAdapter`) get the stream directly. Both are correct. The
-    /// return type is `BiStream` (ADR-092); handlers that want the split
+    /// return type is `BiStream` (ADR-009); handlers that want the split
     /// halves call `tokio::io::split` on the `BiStream`.
     pub async fn accept_bi(&self) -> Result<BiStream, StreamError> {
         self.source.accept_bi().await
@@ -656,7 +658,7 @@ mod tests {
     /// `tokio::io::sink()` + `tokio::io::empty()`: reads yield EOF
     /// immediately (zero bytes), writes discard. Exists because
     /// `Connection::from_bidi` requires a single value that implements
-    /// both traits (ADR-092). Used only to construct a `Connection` for
+    /// both traits (ADR-009). Used only to construct a `Connection` for
     /// tests that exercise `Connection`-level state (alpn, addr, identity)
     /// without ever reading or writing the stream.
     pub(crate) struct SinkEmpty;

src/gateway/dispatch.rs

@@ -0,0 +1,787 @@
+//! The transport-neutral dispatch spine ([`GatewayDispatch`]) and the
+//! shared `services/schema` disclosure guard
+//! ([`schema_disclosure_denial`]) — the non-HTTP half of alkhttp's
+//! gateway, promoted for reuse by hubs, relays, and protocol crates
+//! that expose a call surface to a less-trusted in-transport caller.
+//!
+//! The spine constructs the root [`OperationContext`] identically for
+//! every transport (`internal: false` — ACL runs against the caller's
+//! identity, not a handler's composition authority; `forwarded_for:
+//! None` — wire-ingress only), resolves the registration's
+//! composition authority / capabilities / scoped env into it, and
+//! bounds Once-ops and sink dispatch with a configurable deadline
+//! while leaving streaming subscriptions unbounded (ADR-021:
+//! subscriptions are long-lived).
+//!
+//! There are no HTTP concepts here: no statuses, no headers, no body
+//! framing. CallError → HTTP mapping, NDJSON/SSE framing, body caps,
+//! and batch envelopes are the consumer's (see alkhttp's gateway).
+//!
+//! See ADR-048 for the promotion decision and the divergence note on
+//! ACL-denial codes (`FORBIDDEN` here, spec-404 in the wire handler).
+
+use std::collections::HashMap;
+use std::sync::atomic::{AtomicUsize, Ordering};
+use std::sync::Arc;
+use std::time::{Duration, Instant};
+
+use crate::core::auth::Identity;
+use crate::core::types::Capabilities;
+use crate::protocol::wire::{CallError, ResponseEnvelope};
+use crate::registry::context::{AbortPolicy, OperationContext, ScopedPeerEnv};
+use crate::registry::env::LocalOperationEnv;
+use crate::registry::registration::OperationRegistry;
+use crate::registry::spec::{AccessResult, Visibility};
+use futures::stream::BoxStream;
+use serde_json::Value;
+
+const SERVICES_SCHEMA: &str = "services/schema";
+
+/// The default handler deadline for Once-ops and sink dispatch: 30 s.
+/// Override with [`GatewayDispatch::with_deadline`].
+pub const DEFAULT_DEADLINE: Duration = Duration::from_secs(30);
+
+/// The transport-neutral dispatch spine over an
+/// [`OperationRegistry`](crate::registry::registration::OperationRegistry):
+/// invokes operations for the neutral `ResponseEnvelope` result shape
+/// any transport maps to its own wire format. Identity arrives
+/// per-call as `Option<Identity>` — bearer resolution, transport
+/// framing, and error presentation happen upstream in the consumer.
+///
+/// See the [module docs](crate::gateway) and ADR-048.
+pub struct GatewayDispatch {
+    registry: Arc<OperationRegistry>,
+    deadline: Option<Duration>,
+    invoke_count: AtomicUsize,
+}
+
+impl GatewayDispatch {
+    /// Assemble a dispatch spine over a registry with the 30 s default
+    /// deadline ([`DEFAULT_DEADLINE`]).
+    pub fn new(registry: Arc<OperationRegistry>) -> Self {
+        Self {
+            registry,
+            deadline: Some(DEFAULT_DEADLINE),
+            invoke_count: AtomicUsize::new(0),
+        }
+    }
+
+    /// Override the Once/sink deadline: `Some(duration)` bounds every
+    /// [`GatewayDispatch::invoke`] and [`GatewayDispatch::invoke_sink`]
+    /// dispatch (a hung handler surfaces as a retryable `TIMEOUT` error
+    /// envelope); `None` removes the bound entirely (the wire path's
+    /// Pub dispatch behavior). Streaming subscriptions are unbounded
+    /// either way (ADR-021).
+    pub fn with_deadline(mut self, deadline: Option<Duration>) -> Self {
+        self.deadline = deadline;
+        self
+    }
+
+    /// The registry operations resolve against.
+    pub fn registry(&self) -> &Arc<OperationRegistry> {
+        &self.registry
+    }
+
+    /// How many [`GatewayDispatch::invoke`] calls this spine has
+    /// served. A test-spy accessor: consumers' over-cap batch tests
+    /// assert it stays at zero to prove no dispatch happened before the
+    /// cap rejection.
+    pub fn invoke_count(&self) -> usize {
+        self.invoke_count.load(Ordering::Relaxed)
+    }
+
+    /// Invoke a Query/Mutation op under the configured deadline; a hung
+    /// handler surfaces as a retryable `TIMEOUT` error envelope.
+    pub async fn invoke(
+        &self,
+        identity: Option<Identity>,
+        op: &str,
+        input: Value,
+    ) -> ResponseEnvelope {
+        self.invoke_count.fetch_add(1, Ordering::Relaxed);
+        let operation_name = strip_leading_slash(op).to_string();
+        if let Some(error) =
+            schema_via_call_denial(&self.registry, &operation_name, &input, identity.as_ref())
+        {
+            return ResponseEnvelope::error(uuid::Uuid::new_v4().to_string(), error);
+        }
+        let request_id = uuid::Uuid::new_v4().to_string();
+        let context = self.build_root_context(&request_id, &operation_name, identity);
+        let fut = self.registry.invoke(&operation_name, input, context);
+        match self.deadline {
+            Some(deadline) => match tokio::time::timeout(deadline, fut).await {
+                Ok(envelope) => envelope,
+                Err(_elapsed) => ResponseEnvelope::error(
+                    request_id,
+                    CallError::timeout(format!(
+                        "operation did not complete within the {deadline:?} dispatch deadline"
+                    )),
+                ),
+            },
+            None => fut.await,
+        }
+    }
+
+    /// Dispatch a Sub op: the returned stream of envelopes is unbounded
+    /// by the deadline (subscriptions are long-lived per ADR-021);
+    /// pre-handler failures surface as one error envelope.
+    pub fn invoke_streaming(
+        &self,
+        identity: Option<Identity>,
+        op: &str,
+        input: Value,
+    ) -> BoxStream<'static, ResponseEnvelope> {
+        let operation_name = strip_leading_slash(op).to_string();
+        if let Some(error) =
+            schema_via_call_denial(&self.registry, &operation_name, &input, identity.as_ref())
+        {
+            let request_id = uuid::Uuid::new_v4().to_string();
+            return Box::pin(futures::stream::once(async move {
+                ResponseEnvelope::error(request_id, error)
+            }));
+        }
+        let request_id = uuid::Uuid::new_v4().to_string();
+        let context = self.build_root_context_streaming(&request_id, &operation_name, identity);
+        self.registry
+            .invoke_streaming(&operation_name, input, context)
+    }
+
+    /// Dispatch a `Pub` operation (ADR-046). The `publish_stream` is
+    /// the initiator's chunk stream (each item one published chunk, or
+    /// an initiator-side error). The dispatch — chunk pacing and the
+    /// sink handler's final completion alike — is bounded by the
+    /// configured deadline: a hung sink handler surfaces as a retryable
+    /// `TIMEOUT` error envelope, not an indefinitely-held initiator.
+    pub async fn invoke_sink(
+        &self,
+        identity: Option<Identity>,
+        op: &str,
+        input: Value,
+        publish_stream: crate::registry::registration::PublishStream,
+    ) -> ResponseEnvelope {
+        let operation_name = strip_leading_slash(op).to_string();
+        let request_id = uuid::Uuid::new_v4().to_string();
+        let context = self.build_root_context_sink(&request_id, &operation_name, identity);
+        let fut = self
+            .registry
+            .invoke_sink(&operation_name, input, publish_stream, context);
+        match self.deadline {
+            Some(deadline) => match tokio::time::timeout(deadline, fut).await {
+                Ok(envelope) => envelope,
+                Err(_elapsed) => ResponseEnvelope::error(
+                    request_id,
+                    CallError::timeout(format!(
+                        "operation did not complete within the {deadline:?} dispatch deadline"
+                    )),
+                ),
+            },
+            None => fut.await,
+        }
+    }
+
+    fn build_root_context_sink(
+        &self,
+        request_id: &str,
+        operation_name: &str,
+        identity: Option<Identity>,
+    ) -> OperationContext {
+        self.build_root_context_inner(request_id, operation_name, identity, false)
+    }
+
+    fn build_root_context(
+        &self,
+        request_id: &str,
+        operation_name: &str,
+        identity: Option<Identity>,
+    ) -> OperationContext {
+        self.build_root_context_inner(request_id, operation_name, identity, true)
+    }
+
+    fn build_root_context_streaming(
+        &self,
+        request_id: &str,
+        operation_name: &str,
+        identity: Option<Identity>,
+    ) -> OperationContext {
+        self.build_root_context_inner(request_id, operation_name, identity, false)
+    }
+
+    fn build_root_context_inner(
+        &self,
+        request_id: &str,
+        operation_name: &str,
+        identity: Option<Identity>,
+        bounded: bool,
+    ) -> OperationContext {
+        let registration = self.registry.registration(operation_name);
+        let (composition_authority, capabilities, scoped_env) = match registration {
+            Some(r) => (
+                r.composition_authority.clone(),
+                r.capabilities.clone(),
+                r.scoped_env.clone().unwrap_or_else(ScopedPeerEnv::empty),
+            ),
+            None => (None, Capabilities::new(), ScopedPeerEnv::empty()),
+        };
+
+        let env: Arc<dyn crate::registry::env::OperationEnv + Send + Sync> =
+            Arc::new(LocalOperationEnv::new(Arc::clone(&self.registry)));
+
+        OperationContext {
+            request_id: request_id.to_string(),
+            parent_request_id: None,
+            identity,
+            handler_identity: composition_authority,
+            forwarded_for: None,
+            capabilities,
+            metadata: HashMap::new(),
+            deadline: bounded
+                .then_some(self.deadline)
+                .flatten()
+                .map(|deadline| Instant::now() + deadline),
+            scoped_env,
+            env,
+            abort_policy: AbortPolicy::default(),
+            internal: false,
+            ownership: None,
+        }
+    }
+}
+
+fn strip_leading_slash(operation_id: &str) -> &str {
+    operation_id.strip_prefix('/').unwrap_or(operation_id)
+}
+
+/// The `services/schema` op-path guard (ADR-048): when the dispatched
+/// operation is the schema meta-op, apply the disclosure checks to the
+/// **inner** `name` input before dispatching, so no transport through
+/// this spine can fetch a spec that would be denied for the same
+/// identity elsewhere.
+fn schema_via_call_denial(
+    registry: &OperationRegistry,
+    operation: &str,
+    input: &Value,
+    identity: Option<&Identity>,
+) -> Option<CallError> {
+    let name = input.get("name").and_then(Value::as_str)?;
+    if !matches!(
+        registry.registration(operation),
+        Some(registration) if registration.spec.name == SERVICES_SCHEMA
+    ) {
+        return None;
+    }
+    schema_disclosure_denial(registry, name, identity)
+}
+
+/// The schema-disclosure check shared by every transport surface:
+/// Internal ops are invisible (`NOT_FOUND` regardless of caller);
+/// ACL-forbidden ops are denied with `FORBIDDEN`. One implementation so
+/// the transports cannot drift (CF-004's shared-implementation fix; the
+/// wire-path handler stays the conservative spec-404 outer bound).
+pub fn schema_disclosure_denial(
+    registry: &OperationRegistry,
+    operation: &str,
+    identity: Option<&Identity>,
+) -> Option<CallError> {
+    let name = strip_leading_slash(operation);
+    let registration = registry.registration(name)?;
+    if registration.spec.visibility == Visibility::Internal {
+        return Some(CallError::not_found(operation));
+    }
+    if let AccessResult::Forbidden(message) =
+        registration.spec.access_control.check(identity, None, None)
+    {
+        return Some(CallError::forbidden(message));
+    }
+    None
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::registry::registration::{
+        make_handler, make_sink_handler, make_streaming_handler, HandlerKind, HandlerRegistration,
+        OperationProvenance,
+    };
+    use crate::registry::spec::{AccessControl, OperationSpec, OperationType, Visibility};
+    use futures::StreamExt;
+
+    // The deadline tests use a short custom deadline instead of the
+    // module default, so no wall-clock assertion by hand is needed.
+
+    fn spec(name: &str, visibility: Visibility, op_type: OperationType) -> OperationSpec {
+        OperationSpec::new(
+            name,
+            op_type,
+            visibility,
+            serde_json::json!({}),
+            serde_json::json!({}),
+            vec![],
+            AccessControl::default(),
+            None,
+        )
+    }
+
+    fn echo_registry() -> Arc<OperationRegistry> {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("echo/run", Visibility::External, OperationType::Query),
+                HandlerKind::Once(make_handler(|input, ctx| async move {
+                    ResponseEnvelope::ok(ctx.request_id, input)
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        Arc::new(registry)
+    }
+
+    #[tokio::test]
+    async fn invoke_external_op_round_trips() {
+        let dispatch = GatewayDispatch::new(echo_registry());
+        let envelope = dispatch
+            .invoke(None, "/echo/run", serde_json::json!({ "x": 1 }))
+            .await;
+        assert!(envelope.result.is_ok(), "expected ok, got {envelope:?}");
+    }
+
+    #[tokio::test]
+    async fn invoke_unknown_op_returns_not_found() {
+        let dispatch = GatewayDispatch::new(echo_registry());
+        let envelope = dispatch
+            .invoke(None, "/missing/op", serde_json::json!({}))
+            .await;
+        match envelope.result {
+            Err(error) => assert_eq!(error.code, "NOT_FOUND"),
+            Ok(v) => panic!("expected error, got {v:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn invoke_internal_op_returns_not_found() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("internal/op", Visibility::Internal, OperationType::Query),
+                HandlerKind::Once(make_handler(|_input, ctx| async move {
+                    ResponseEnvelope::ok(ctx.request_id, serde_json::json!({}))
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let dispatch = GatewayDispatch::new(Arc::new(registry));
+        let envelope = dispatch
+            .invoke(None, "/internal/op", serde_json::json!({}))
+            .await;
+        match envelope.result {
+            Err(error) => assert_eq!(error.code, "NOT_FOUND"),
+            Ok(v) => panic!("expected error, got {v:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn invoke_enforces_the_configured_deadline_on_a_hung_handler() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("hung/op", Visibility::External, OperationType::Query),
+                HandlerKind::Once(make_handler(|_input, ctx| async move {
+                    tokio::time::sleep(Duration::from_secs(120)).await;
+                    ResponseEnvelope::ok(ctx.request_id, serde_json::json!({}))
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let dispatch =
+            GatewayDispatch::new(Arc::new(registry)).with_deadline(Some(Duration::from_millis(50)));
+        let started = std::time::Instant::now();
+        let envelope = dispatch
+            .invoke(None, "/hung/op", serde_json::json!({}))
+            .await;
+        assert!(
+            started.elapsed() < Duration::from_secs(10),
+            "the 50 ms deadline must fire well before the 120 s handler sleep"
+        );
+        match envelope.result {
+            Err(error) => {
+                assert_eq!(error.code, "TIMEOUT");
+                assert!(error.retryable, "the deadline error is retryable");
+            }
+            Ok(v) => panic!("expected a TIMEOUT error, got {v:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn invoke_completes_within_the_deadline_for_a_fast_handler() {
+        let dispatch = GatewayDispatch::new(echo_registry());
+        let envelope = dispatch
+            .invoke(None, "/echo/run", serde_json::json!({}))
+            .await;
+        assert!(envelope.result.is_ok(), "a fast handler must not time out");
+    }
+
+    #[tokio::test]
+    async fn invoke_with_no_deadline_completes_a_slow_handler() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("slow/op", Visibility::External, OperationType::Query),
+                HandlerKind::Once(make_handler(|_input, ctx| async move {
+                    tokio::time::sleep(Duration::from_millis(80)).await;
+                    ResponseEnvelope::ok(ctx.request_id, serde_json::json!({}))
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let dispatch = GatewayDispatch::new(Arc::new(registry)).with_deadline(None);
+        let envelope = dispatch
+            .invoke(None, "/slow/op", serde_json::json!({}))
+            .await;
+        assert!(
+            envelope.result.is_ok(),
+            "a no-deadline spine must not time out, got {envelope:?}"
+        );
+    }
+
+    #[tokio::test]
+    async fn invoke_sink_enforces_the_configured_deadline_on_a_hung_sink_handler() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("hung/sink", Visibility::External, OperationType::Pub),
+                HandlerKind::Sink(make_sink_handler(|_input, ctx, _chunks| async move {
+                    tokio::time::sleep(Duration::from_secs(120)).await;
+                    ResponseEnvelope::ok(ctx.request_id, serde_json::json!({}))
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let dispatch =
+            GatewayDispatch::new(Arc::new(registry)).with_deadline(Some(Duration::from_millis(50)));
+        let chunks: crate::registry::registration::PublishStream =
+            Box::pin(futures::stream::empty());
+        let started = std::time::Instant::now();
+        let envelope = dispatch
+            .invoke_sink(None, "/hung/sink", serde_json::json!({}), chunks)
+            .await;
+        assert!(
+            started.elapsed() < Duration::from_secs(10),
+            "the 50 ms deadline must fire well before the 120 s handler sleep"
+        );
+        match envelope.result {
+            Err(error) => {
+                assert_eq!(error.code, "TIMEOUT");
+                assert!(error.retryable, "the deadline error is retryable");
+            }
+            Ok(v) => panic!("expected a TIMEOUT error, got {v:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn invoke_sink_completes_within_the_deadline_for_a_fast_sink_handler() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("fast/sink", Visibility::External, OperationType::Pub),
+                HandlerKind::Sink(make_sink_handler(|_input, ctx, mut chunks| async move {
+                    let mut count = 0u64;
+                    while let Some(chunk) = chunks.next().await {
+                        if chunk.is_ok() {
+                            count += 1;
+                        }
+                    }
+                    ResponseEnvelope::ok(ctx.request_id, serde_json::json!({ "count": count }))
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let dispatch = GatewayDispatch::new(Arc::new(registry));
+        let chunks: crate::registry::registration::PublishStream =
+            Box::pin(futures::stream::iter(vec![
+                Ok(serde_json::json!({ "n": 1 })),
+                Ok(serde_json::json!({ "n": 2 })),
+            ]));
+        let envelope = dispatch
+            .invoke_sink(None, "/fast/sink", serde_json::json!({}), chunks)
+            .await;
+        assert!(
+            envelope.result.is_ok(),
+            "a fast sink handler must not time out, got {envelope:?}"
+        );
+    }
+
+    #[tokio::test]
+    async fn invoke_sink_with_no_deadline_completes_a_slow_sink_handler() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("slow/sink", Visibility::External, OperationType::Pub),
+                HandlerKind::Sink(make_sink_handler(|_input, ctx, mut chunks| async move {
+                    while let Some(chunk) = chunks.next().await {
+                        let _ = chunk;
+                    }
+                    tokio::time::sleep(Duration::from_millis(80)).await;
+                    ResponseEnvelope::ok(ctx.request_id, serde_json::json!({}))
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let dispatch = GatewayDispatch::new(Arc::new(registry)).with_deadline(None);
+        let chunks: crate::registry::registration::PublishStream = Box::pin(futures::stream::iter(
+            vec![Ok(serde_json::json!({ "n": 1 }))],
+        ));
+        let envelope = dispatch
+            .invoke_sink(None, "/slow/sink", serde_json::json!({}), chunks)
+            .await;
+        assert!(
+            envelope.result.is_ok(),
+            "a no-deadline spine must not time out a slow sink, got {envelope:?}"
+        );
+    }
+
+    #[tokio::test]
+    async fn streaming_sub_op_streams_envelopes() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("tick/stream", Visibility::External, OperationType::Sub),
+                HandlerKind::Stream(make_streaming_handler(|input, ctx| {
+                    let count = input.get("count").and_then(|v| v.as_u64()).unwrap_or(2);
+                    let request_id = ctx.request_id.clone();
+                    futures::stream::iter(0..count).map(move |i| {
+                        ResponseEnvelope::ok(request_id.clone(), serde_json::json!({ "tick": i }))
+                    })
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let dispatch = GatewayDispatch::new(Arc::new(registry));
+        let mut stream =
+            dispatch.invoke_streaming(None, "/tick/stream", serde_json::json!({ "count": 3 }));
+        let mut ticks = Vec::new();
+        while let Some(envelope) = stream.next().await {
+            ticks.push(envelope);
+        }
+        assert_eq!(ticks.len(), 3);
+    }
+
+    #[tokio::test]
+    async fn invoke_count_counts_invokes_only() {
+        let dispatch = GatewayDispatch::new(echo_registry());
+        let _ = dispatch
+            .invoke(None, "/echo/run", serde_json::json!({}))
+            .await;
+        let _ = dispatch.invoke_streaming(None, "/echo/run", serde_json::json!({}));
+        assert_eq!(dispatch.invoke_count(), 1);
+    }
+
+    fn registry_with_services_schema_over(inner_ops: Vec<OperationSpec>) -> Arc<OperationRegistry> {
+        use crate::registry::discovery::{services_schema_handler, services_schema_spec};
+
+        let inner = Arc::new({
+            let registry = OperationRegistry::new();
+            for op_spec in &inner_ops {
+                registry
+                    .register(HandlerRegistration::new(
+                        op_spec.clone(),
+                        HandlerKind::Once(make_handler(|input, ctx| async move {
+                            ResponseEnvelope::ok(ctx.request_id, input)
+                        })),
+                        OperationProvenance::Local,
+                        None,
+                        None,
+                        Capabilities::new(),
+                    ))
+                    .unwrap();
+            }
+            registry
+        });
+        let registry = OperationRegistry::new();
+        for op_spec in &inner_ops {
+            registry
+                .register(HandlerRegistration::new(
+                    op_spec.clone(),
+                    HandlerKind::Once(make_handler(|input, ctx| async move {
+                        ResponseEnvelope::ok(ctx.request_id, input)
+                    })),
+                    OperationProvenance::Local,
+                    None,
+                    None,
+                    Capabilities::new(),
+                ))
+                .unwrap();
+        }
+        registry
+            .register(HandlerRegistration::new(
+                services_schema_spec(),
+                HandlerKind::Once(services_schema_handler(Arc::clone(&inner))),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        Arc::new(registry)
+    }
+
+    #[tokio::test]
+    async fn invoke_of_services_schema_with_internal_inner_name_is_blocked_pre_dispatch() {
+        let registry = registry_with_services_schema_over(vec![spec(
+            "secret/op",
+            Visibility::Internal,
+            OperationType::Query,
+        )]);
+        let dispatch = GatewayDispatch::new(registry);
+        let envelope = dispatch
+            .invoke(
+                None,
+                "services/schema",
+                serde_json::json!({ "name": "secret/op" }),
+            )
+            .await;
+        match envelope.result {
+            Err(error) => {
+                assert_eq!(error.code, "NOT_FOUND");
+                assert!(error.message.contains("secret/op"));
+            }
+            Ok(v) => panic!("the internal op spec must not be returned, got {v:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn invoke_of_services_schema_with_authorized_inner_name_still_projects() {
+        let registry = registry_with_services_schema_over(vec![spec(
+            "public/op",
+            Visibility::External,
+            OperationType::Query,
+        )]);
+        let dispatch = GatewayDispatch::new(registry);
+        let envelope = dispatch
+            .invoke(
+                None,
+                "services/schema",
+                serde_json::json!({ "name": "public/op" }),
+            )
+            .await;
+        assert!(
+            envelope.result.is_ok(),
+            "an allowed inner name must still project, got {envelope:?}"
+        );
+        assert_eq!(
+            envelope
+                .result
+                .as_ref()
+                .ok()
+                .and_then(|v| v.get("name"))
+                .and_then(Value::as_str),
+            Some("public/op")
+        );
+    }
+
+    #[tokio::test]
+    async fn invoke_of_services_schema_with_forbidden_inner_name_is_denied() {
+        let restricted = OperationSpec::new(
+            "admin/op",
+            OperationType::Query,
+            Visibility::External,
+            serde_json::json!({}),
+            serde_json::json!({}),
+            vec![],
+            AccessControl {
+                required_scopes: vec!["admin".to_string()],
+                ..Default::default()
+            },
+            None,
+        );
+        let registry = registry_with_services_schema_over(vec![restricted]);
+        let dispatch = GatewayDispatch::new(registry);
+        let envelope = dispatch
+            .invoke(
+                Some(Identity {
+                    id: "user".to_string(),
+                    scopes: vec!["user".to_string()],
+                    resources: HashMap::new(),
+                }),
+                "services/schema",
+                serde_json::json!({ "name": "admin/op" }),
+            )
+            .await;
+        match envelope.result {
+            Err(error) => assert_eq!(error.code, "FORBIDDEN"),
+            Ok(v) => panic!("the ACL-denied spec must not be returned, got {v:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn invoke_streaming_of_services_schema_with_internal_inner_name_is_blocked() {
+        let registry = registry_with_services_schema_over(vec![spec(
+            "secret/op",
+            Visibility::Internal,
+            OperationType::Query,
+        )]);
+        let dispatch = GatewayDispatch::new(registry);
+        let mut stream = dispatch.invoke_streaming(
+            None,
+            "services/schema",
+            serde_json::json!({ "name": "secret/op" }),
+        );
+        let envelopes: Vec<ResponseEnvelope> = stream.by_ref().collect().await;
+        assert_eq!(envelopes.len(), 1, "the guard error is the only item");
+        match &envelopes[0].result {
+            Err(error) => assert_eq!(error.code, "NOT_FOUND"),
+            Ok(v) => panic!("the internal op spec must not be returned, got {v:?}"),
+        }
+    }
+
+    #[test]
+    fn schema_disclosure_denial_hides_internal_ops() {
+        let registry = OperationRegistry::new();
+        registry
+            .register(HandlerRegistration::new(
+                spec("secret/op", Visibility::Internal, OperationType::Query),
+                HandlerKind::Once(make_handler(|_input, ctx| async move {
+                    ResponseEnvelope::ok(ctx.request_id, serde_json::json!({}))
+                })),
+                OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let error = schema_disclosure_denial(&registry, "secret/op", None).unwrap();
+        assert_eq!(error.code, "NOT_FOUND");
+        assert!(schema_disclosure_denial(&registry, "missing/op", None).is_none());
+    }
+
+    #[test]
+    fn schema_disclosure_denial_allows_unrestricted_ops_without_identity() {
+        let error = schema_disclosure_denial(&echo_registry(), "echo/run", None);
+        assert!(
+            error.is_none(),
+            "default-ACL ops are fetchable unauthenticated"
+        );
+    }
+}

src/gateway/mod.rs

@@ -0,0 +1,6 @@
+//! The transport-neutral dispatch spine and the shared
+//! `services/schema` disclosure guard (ADR-048).
+
+pub(crate) mod dispatch;
+
+pub use dispatch::{schema_disclosure_denial, GatewayDispatch, DEFAULT_DEADLINE};

src/lib.rs

@@ -17,6 +17,10 @@
 //!   going forward.
 //! - **Registry** ([`registry`]): operation specs, context, dispatch, and
 //!   the operation registry — the call half's dispatch core.
+//! - **Gateway** (module `gateway`, feature `gateway`): the transport-neutral
+//!   dispatch spine and `services/schema` disclosure guard — the
+//!   deadline-bounded, re-rooted-context invoke surface for HTTP
+//!   gateways, hub relays, and other transport front-ends (ADR-048).
 //! - **Protocol** ([`protocol`]): wire format, streams, adapter, dispatch
 //!   loop, pending requests, abort cascade — the call half's wire layer.
 //! - **Client** ([`client`]): `CallClient`, `from_call`, `OperationAdapter`
@@ -71,6 +75,8 @@
 pub mod channels;
 pub mod client;
 pub mod core;
+#[cfg(feature = "gateway")]
+pub mod gateway;
 pub mod protocol;
 pub mod registry;
 

src/protocol/adapter.rs

@@ -253,7 +253,7 @@ mod tests {
         acl: AccessControl,
         handler: crate::registry::registration::Handler,
     ) -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 OperationSpec::new(
@@ -404,7 +404,7 @@ mod tests {
 
     #[tokio::test]
     async fn build_root_context_carries_capabilities_and_scoped_env() {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let scoped = ScopedPeerEnv::new(["fs/readFile"]);
         let caps = Capabilities::new().with_api_key("google", "k".to_string());
         registry

src/protocol/connection.rs

@@ -164,6 +164,18 @@ impl CallConnection {
         self.imported_operations.write().insert(name, registration);
     }
 
+    /// `true` when the connection overlay holds a registration for
+    /// `name` — the collision check `op/register`'s replace semantics
+    /// need (review 004 F-05).
+    pub fn overlay_contains(&self, name: &str) -> bool {
+        self.imported_operations.read().contains_key(name)
+    }
+
+    /// The connection overlay's registration for `name`, if present.
+    pub fn overlay_registration(&self, name: &str) -> Option<HandlerRegistration> {
+        self.imported_operations.read().get(name).cloned()
+    }
+
     pub fn register_imported_all(&self, registrations: Vec<HandlerRegistration>) {
         let mut overlay = self.imported_operations.write();
         for reg in registrations {
@@ -213,7 +225,7 @@ impl CallConnection {
             }
         };
 
-        // `open_bi` returns a `BiStream` (ADR-092); split it into halves for
+        // `open_bi` returns a `BiStream` (ADR-009); split it into halves for
         // the call protocol's separate write (request) and read (response)
         // pumps. The split is the stdlib idiom; no per-handler wrapper.
         let stream = match connection.open_bi().await {
@@ -235,7 +247,7 @@ impl CallConnection {
         };
 
         if let Err(err) = self.write_request(send, &request_id, payload).await {
-            let call_error = CallError::internal(err);
+            let call_error = CallError::connection_closed(err);
             self.pending
                 .lock()
                 .handle_error(&request_id, call_error.clone());
@@ -276,7 +288,8 @@ impl CallConnection {
 
         let envelope = EventEnvelope::requested(&request_id, payload);
         if let Err(err) = writer.write_frame(&envelope).await {
-            let call_error = CallError::internal(format!("failed to write request frame: {err}"));
+            let call_error =
+                CallError::connection_closed(format!("failed to write request frame: {err}"));
             self.pending
                 .lock()
                 .handle_error(&request_id, call_error.clone());
@@ -336,7 +349,7 @@ impl CallConnection {
             }
         };
 
-        // `open_bi` returns a `BiStream` (ADR-092); split for the separate
+        // `open_bi` returns a `BiStream` (ADR-009); split for the separate
         // write (request) and read (subscription events) pumps.
         let stream = match connection.open_bi().await {
             Ok(s) => s,
@@ -353,7 +366,7 @@ impl CallConnection {
         };
 
         if let Err(err) = self.write_request(send, &request_id, payload).await {
-            let call_error = CallError::internal(err);
+            let call_error = CallError::connection_closed(err);
             self.pending
                 .lock()
                 .handle_error(&request_id, call_error.clone());
@@ -385,7 +398,8 @@ impl CallConnection {
 
         let envelope = EventEnvelope::requested(&request_id, payload);
         if let Err(err) = writer.write_frame(&envelope).await {
-            let call_error = CallError::internal(format!("failed to write request frame: {err}"));
+            let call_error =
+                CallError::connection_closed(format!("failed to write request frame: {err}"));
             self.pending
                 .lock()
                 .handle_error(&request_id, call_error.clone());
@@ -498,8 +512,17 @@ impl CallConnection {
         });
 
         let write_result = pump_publish_to_wire(send, &request_id, payload, stream).await;
-        if let Err(err) = write_result {
-            let call_error = CallError::internal(err);
+        if let Err((stage, err)) = write_result {
+            // A write failure before the request frame could be delivered is
+            // provably undelivered — retry is safe. Mid-publish failures leave
+            // delivery ambiguous; the INTERNAL mapping (non-retryable) is the
+            // safe default there.
+            let call_error = match stage {
+                PublishWriteStage::Request => CallError::connection_closed(err),
+                PublishWriteStage::Published | PublishWriteStage::Completed => {
+                    CallError::internal(err)
+                }
+            };
             self.pending
                 .lock()
                 .handle_error(&request_id, call_error.clone());
@@ -533,7 +556,8 @@ impl CallConnection {
 
         let requested = EventEnvelope::requested(&request_id, payload);
         if let Err(err) = writer.write_frame(&requested).await {
-            let call_error = CallError::internal(format!("failed to write request frame: {err}"));
+            let call_error =
+                CallError::connection_closed(format!("failed to write request frame: {err}"));
             self.pending
                 .lock()
                 .handle_error(&request_id, call_error.clone());
@@ -588,7 +612,7 @@ impl CallConnection {
             .connection
             .as_ref()
             .ok_or_else(|| "no underlying connection (overlay-only)".to_string())?;
-        // `open_bi` returns a `BiStream` (ADR-092). We only need the write
+        // `open_bi` returns a `BiStream` (ADR-009). We only need the write
         // half to send the envelope; split and drop the read half.
         let stream = connection
             .open_bi()
@@ -603,33 +627,48 @@ impl CallConnection {
     }
 }
 
+/// Which frame of a publish pump failed to write. The `Request` stage is
+/// provably undelivered (the producer never received `call.requested`, so a
+/// reconnect/retry is safe); the later stages leave delivery ambiguous.
+enum PublishWriteStage {
+    Request,
+    Published,
+    Completed,
+}
+
 async fn pump_publish_to_wire<W>(
     send: W,
     request_id: &str,
     payload: Value,
     mut stream: Pin<Box<dyn Stream<Item = Value> + Send>>,
-) -> Result<(), String>
+) -> Result<(), (PublishWriteStage, String)>
 where
     W: AsyncWrite + Unpin,
 {
     let mut writer = FrameFramedWriter::new(send);
     let requested = EventEnvelope::requested(request_id, payload);
-    writer
-        .write_frame(&requested)
-        .await
-        .map_err(|e| format!("failed to write request frame: {e}"))?;
+    writer.write_frame(&requested).await.map_err(|e| {
+        (
+            PublishWriteStage::Request,
+            format!("failed to write request frame: {e}"),
+        )
+    })?;
     while let Some(chunk) = stream.next().await {
         let envelope = EventEnvelope::published(request_id, chunk);
-        writer
-            .write_frame(&envelope)
-            .await
-            .map_err(|e| format!("failed to write published frame: {e}"))?;
+        writer.write_frame(&envelope).await.map_err(|e| {
+            (
+                PublishWriteStage::Published,
+                format!("failed to write published frame: {e}"),
+            )
+        })?;
     }
     let completed = EventEnvelope::completed(request_id);
-    writer
-        .write_frame(&completed)
-        .await
-        .map_err(|e| format!("failed to write completed frame: {e}"))?;
+    writer.write_frame(&completed).await.map_err(|e| {
+        (
+            PublishWriteStage::Completed,
+            format!("failed to write completed frame: {e}"),
+        )
+    })?;
     Ok(())
 }
 
@@ -799,6 +838,10 @@ impl OperationEnv for OverlayOperationEnv {
     fn contains(&self, name: &str) -> bool {
         self.overlay.read().contains_key(name)
     }
+
+    fn list_operation_names(&self) -> Vec<String> {
+        self.overlay.read().keys().cloned().collect()
+    }
 }
 
 pub struct SubscriptionStream {
@@ -1476,7 +1519,7 @@ mod tests {
             )
         });
 
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 pub_spec_e2e("fs/upload"),
@@ -1880,6 +1923,346 @@ mod tests {
         assert_eq!(read, envelope);
     }
 
+    // CF-001 — a transport write failure before the request frame could be
+    // delivered maps to retryable CONNECTION_CLOSED (the call provably
+    // never reached the producer). Mid-publish failures stay INTERNAL.
+
+    /// AsyncWrite that always errors — simulates a dead mux mid-call.
+    struct BrokenWrite;
+    impl AsyncWrite for BrokenWrite {
+        fn poll_write(
+            self: Pin<&mut Self>,
+            _cx: &mut Context<'_>,
+            _buf: &[u8],
+        ) -> Poll<std::io::Result<usize>> {
+            Poll::Ready(Err(std::io::Error::new(
+                std::io::ErrorKind::BrokenPipe,
+                "mux dead",
+            )))
+        }
+        fn poll_flush(self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
+            Poll::Ready(Ok(()))
+        }
+        fn poll_shutdown(self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
+            Poll::Ready(Ok(()))
+        }
+    }
+
+    fn broken_write_bi_stream() -> crate::core::types::BiStream {
+        // The joined stream holds the read end; the write half always errors.
+        let (read_end, _write_end) = tokio::io::duplex(1024);
+        crate::core::types::BiStream::from_joined(read_end, BrokenWrite)
+    }
+
+    /// AsyncWrite whose writes succeed but whose Nth `poll_flush` fails.
+    /// `FrameFramedWriter::write_frame` flushes exactly once per frame, so
+    /// `remaining_ok_flushes = k` lets frames 1..=k land and fails inside
+    /// frame k+1 — a deterministic mid-stream write failure.
+    struct FailOnFlushN {
+        remaining_ok_flushes: std::sync::atomic::AtomicUsize,
+    }
+    impl AsyncWrite for FailOnFlushN {
+        fn poll_write(
+            self: Pin<&mut Self>,
+            _cx: &mut Context<'_>,
+            buf: &[u8],
+        ) -> Poll<std::io::Result<usize>> {
+            Poll::Ready(Ok(buf.len()))
+        }
+        fn poll_flush(self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
+            let prev = self
+                .remaining_ok_flushes
+                .fetch_sub(1, std::sync::atomic::Ordering::SeqCst);
+            if prev == 0 {
+                Poll::Ready(Err(std::io::Error::new(
+                    std::io::ErrorKind::BrokenPipe,
+                    "mux dead",
+                )))
+            } else {
+                Poll::Ready(Ok(()))
+            }
+        }
+        fn poll_shutdown(self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
+            Poll::Ready(Ok(()))
+        }
+    }
+
+    #[tokio::test]
+    async fn single_stream_call_write_failure_is_retryable_connection_closed() {
+        let bi_stream = broken_write_bi_stream();
+        let (writer, reader) = split_single_stream(bi_stream);
+        let conn = CallConnection::new_single_stream(stub_connection(), writer);
+        drop(reader);
+
+        let response = conn.call("test/echo", serde_json::json!({})).await;
+        match response.result {
+            Err(e) => {
+                assert_eq!(e.code, "CONNECTION_CLOSED");
+                assert!(e.retryable);
+            }
+            other => panic!("expected retryable CONNECTION_CLOSED, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn multi_stream_call_write_request_failure_is_retryable() {
+        // A `Connection` whose open_bi succeeds but whose write half is
+        // dead: the request write fails, the call provably undelivered.
+        use crate::core::types::StreamError;
+        use crate::core::types::{BiStream, BidiStreamSource, Connection};
+        use std::net::SocketAddr;
+
+        struct OneBiStream(std::sync::Mutex<Option<crate::core::types::BiStream>>);
+        #[async_trait::async_trait]
+        impl BidiStreamSource for OneBiStream {
+            async fn accept_bi(&self) -> Result<BiStream, StreamError> {
+                Err(StreamError::ConnectionClosed)
+            }
+            async fn open_bi(&self) -> Result<BiStream, StreamError> {
+                let stream = self.0.lock().expect("open once").take();
+                stream.ok_or(StreamError::StreamClosed)
+            }
+            fn remote_addr(&self) -> Option<SocketAddr> {
+                None
+            }
+            fn close(&self, _code: u32, _reason: &str) {}
+        }
+
+        let conn = Connection::from_source(
+            OneBiStream(std::sync::Mutex::new(Some(broken_write_bi_stream()))),
+            b"alk/call".to_vec(),
+        );
+        let call_conn = CallConnection::new(conn);
+        let response = call_conn.call("test/echo", serde_json::json!({})).await;
+        match response.result {
+            Err(e) => {
+                assert_eq!(e.code, "CONNECTION_CLOSED");
+                assert!(e.retryable);
+            }
+            other => panic!("expected retryable CONNECTION_CLOSED, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn multi_stream_publish_request_frame_write_failure_is_retryable() {
+        use crate::core::types::StreamError;
+        use crate::core::types::{BiStream, BidiStreamSource, Connection};
+        use std::net::SocketAddr;
+
+        struct OneBiStream(std::sync::Mutex<Option<crate::core::types::BiStream>>);
+        #[async_trait::async_trait]
+        impl BidiStreamSource for OneBiStream {
+            async fn accept_bi(&self) -> Result<BiStream, StreamError> {
+                Err(StreamError::ConnectionClosed)
+            }
+            async fn open_bi(&self) -> Result<BiStream, StreamError> {
+                let stream = self.0.lock().expect("open once").take();
+                stream.ok_or(StreamError::StreamClosed)
+            }
+            fn remote_addr(&self) -> Option<SocketAddr> {
+                None
+            }
+            fn close(&self, _code: u32, _reason: &str) {}
+        }
+
+        let conn = Connection::from_source(
+            OneBiStream(std::sync::Mutex::new(Some(broken_write_bi_stream()))),
+            b"alk/call".to_vec(),
+        );
+        let call_conn = CallConnection::new(conn);
+        let response = call_conn
+            .publish(
+                "fs/upload",
+                serde_json::json!({}),
+                Box::pin(futures::stream::iter(vec![serde_json::json!({"c": 1})])),
+            )
+            .await;
+        match response.result {
+            Err(e) => {
+                // The pump died on the *request* frame — nothing delivered.
+                assert_eq!(e.code, "CONNECTION_CLOSED");
+                assert!(e.retryable);
+            }
+            other => panic!("expected retryable CONNECTION_CLOSED, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn connection_closed_error_is_retryable() {
+        let err = CallError::connection_closed("test");
+        assert_eq!(err.code, "CONNECTION_CLOSED");
+        assert!(err.retryable);
+    }
+
+    #[tokio::test]
+    async fn single_stream_subscribe_write_failure_is_retryable_connection_closed() {
+        use futures::stream::StreamExt;
+
+        let bi_stream = broken_write_bi_stream();
+        let (writer, reader) = split_single_stream(bi_stream);
+        let conn = CallConnection::new_single_stream(stub_connection(), writer);
+        drop(reader);
+
+        let mut stream = conn.subscribe("test/events", serde_json::json!({})).await;
+        let item = stream.next().await.unwrap();
+        match item.result {
+            Err(e) => {
+                assert_eq!(e.code, "CONNECTION_CLOSED");
+                assert!(e.retryable);
+            }
+            other => panic!("expected retryable CONNECTION_CLOSED, got {other:?}"),
+        }
+        assert!(stream.next().await.is_none());
+    }
+
+    #[tokio::test]
+    async fn multi_stream_subscribe_write_failure_is_retryable_connection_closed() {
+        use crate::core::types::StreamError;
+        use crate::core::types::{BiStream, BidiStreamSource, Connection};
+        use futures::stream::StreamExt;
+        use std::net::SocketAddr;
+
+        struct OneBiStream(std::sync::Mutex<Option<crate::core::types::BiStream>>);
+        #[async_trait::async_trait]
+        impl BidiStreamSource for OneBiStream {
+            async fn accept_bi(&self) -> Result<BiStream, StreamError> {
+                Err(StreamError::ConnectionClosed)
+            }
+            async fn open_bi(&self) -> Result<BiStream, StreamError> {
+                let stream = self.0.lock().expect("open once").take();
+                stream.ok_or(StreamError::StreamClosed)
+            }
+            fn remote_addr(&self) -> Option<SocketAddr> {
+                None
+            }
+            fn close(&self, _code: u32, _reason: &str) {}
+        }
+
+        let conn = Connection::from_source(
+            OneBiStream(std::sync::Mutex::new(Some(broken_write_bi_stream()))),
+            b"alk/call".to_vec(),
+        );
+        let call_conn = CallConnection::new(conn);
+
+        let mut stream = call_conn
+            .subscribe("test/events", serde_json::json!({}))
+            .await;
+        let item = stream.next().await.unwrap();
+        match item.result {
+            Err(e) => {
+                assert_eq!(e.code, "CONNECTION_CLOSED");
+                assert!(e.retryable);
+            }
+            other => panic!("expected retryable CONNECTION_CLOSED, got {other:?}"),
+        }
+        assert!(stream.next().await.is_none());
+    }
+
+    #[tokio::test]
+    async fn single_stream_publish_request_frame_write_failure_is_retryable() {
+        let bi_stream = broken_write_bi_stream();
+        let (writer, reader) = split_single_stream(bi_stream);
+        let conn = CallConnection::new_single_stream(stub_connection(), writer);
+        drop(reader);
+
+        let response = conn
+            .publish(
+                "fs/upload",
+                serde_json::json!({}),
+                Box::pin(futures::stream::iter(vec![serde_json::json!({"c": 1})])),
+            )
+            .await;
+        match response.result {
+            Err(e) => {
+                // The request frame never landed — nothing delivered, retry safe.
+                assert_eq!(e.code, "CONNECTION_CLOSED");
+                assert!(e.retryable);
+            }
+            other => panic!("expected retryable CONNECTION_CLOSED, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn single_stream_publish_midpublish_write_failure_stays_internal() {
+        // Flush succeeds for the request frame (1) + first chunk (2), then
+        // fails inside the second chunk — delivery is ambiguous from there.
+        let bi_stream = crate::core::types::BiStream::from_joined(
+            tokio::io::empty(),
+            FailOnFlushN {
+                remaining_ok_flushes: std::sync::atomic::AtomicUsize::new(2),
+            },
+        );
+        let (writer, reader) = split_single_stream(bi_stream);
+        let conn = CallConnection::new_single_stream(stub_connection(), writer);
+        drop(reader);
+
+        let chunks = vec![serde_json::json!({"c": 1}), serde_json::json!({"c": 2})];
+        let stream: Pin<Box<dyn Stream<Item = Value> + Send>> =
+            Box::pin(futures::stream::iter(chunks));
+        let response = conn
+            .publish("fs/upload", serde_json::json!({}), stream)
+            .await;
+        match response.result {
+            Err(e) => {
+                assert_eq!(e.code, "INTERNAL");
+                assert!(!e.retryable, "mid-publish failure must not be retryable");
+            }
+            other => panic!("expected non-retryable INTERNAL, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn multi_stream_publish_midpublish_write_failure_stays_internal() {
+        use crate::core::types::StreamError;
+        use crate::core::types::{BiStream, BidiStreamSource, Connection};
+        use std::net::SocketAddr;
+
+        struct OneBiStream(std::sync::Mutex<Option<crate::core::types::BiStream>>);
+        #[async_trait::async_trait]
+        impl BidiStreamSource for OneBiStream {
+            async fn accept_bi(&self) -> Result<BiStream, StreamError> {
+                Err(StreamError::ConnectionClosed)
+            }
+            async fn open_bi(&self) -> Result<BiStream, StreamError> {
+                let stream = self.0.lock().expect("open once").take();
+                stream.ok_or(StreamError::StreamClosed)
+            }
+            fn remote_addr(&self) -> Option<SocketAddr> {
+                None
+            }
+            fn close(&self, _code: u32, _reason: &str) {}
+        }
+
+        let conn = Connection::from_source(
+            OneBiStream(std::sync::Mutex::new(Some(
+                crate::core::types::BiStream::from_joined(
+                    tokio::io::empty(),
+                    FailOnFlushN {
+                        remaining_ok_flushes: std::sync::atomic::AtomicUsize::new(2),
+                    },
+                ),
+            ))),
+            b"alk/call".to_vec(),
+        );
+        let call_conn = CallConnection::new(conn);
+
+        let chunks = vec![serde_json::json!({"c": 1}), serde_json::json!({"c": 2})];
+        let stream: Pin<Box<dyn Stream<Item = Value> + Send>> =
+            Box::pin(futures::stream::iter(chunks));
+        let response = call_conn
+            .publish("fs/upload", serde_json::json!({}), stream)
+            .await;
+        match response.result {
+            Err(e) => {
+                // The pump died on a *chunk* frame — the request landed,
+                // delivery is ambiguous, INTERNAL (non-retryable) is correct.
+                assert_eq!(e.code, "INTERNAL");
+                assert!(!e.retryable);
+            }
+            other => panic!("expected non-retryable INTERNAL, got {other:?}"),
+        }
+    }
+
     #[tokio::test]
     async fn shared_frame_writer_concurrent_writes_do_not_interleave() {
         let (client_end, server_end) = tokio::io::duplex(64 * 1024);
@@ -1993,7 +2376,7 @@ mod tests {
             }
         }
 
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 external_spec("test/echo"),
@@ -2078,7 +2461,7 @@ mod tests {
             }
         }
 
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 OperationSpec::new(

src/protocol/dispatch.rs

@@ -32,7 +32,7 @@ use super::abort::AbortCascade;
 use super::connection::CallConnection;
 use super::wire::{
     CallError, EventEnvelope, FrameFramedReader, FrameFramedWriter, ResponseEnvelope,
-    EVENT_ABORTED, EVENT_COMPLETED, EVENT_ERROR, EVENT_PUBLISHED, EVENT_REQUESTED,
+    EVENT_ABORTED, EVENT_COMPLETED, EVENT_ERROR, EVENT_PUBLISHED, EVENT_REQUESTED, EVENT_RESPONDED,
 };
 use crate::protocol::adapter::SessionOverlaySource;
 use crate::registry::context::{AbortPolicy, OperationContext, ScopedPeerEnv};
@@ -114,6 +114,18 @@ impl std::fmt::Debug for DispatchResult {
     }
 }
 
+/// The started half of a dispatched `call.requested` event, from
+/// `Dispatcher::dispatch_start`. `Once`/`Stream` are returned ready
+/// for the caller to spawn (review 005 G-01); `Sink` was started
+/// inline (its `chunk_tx` must be registered before the next
+/// `call.published` frame is read) and carries the started handler
+/// future.
+pub enum StartedDispatch {
+    Once(Pin<Box<dyn Future<Output = ResponseEnvelope> + Send>>),
+    Stream(ResponseStream),
+    Sink(SinkDispatch),
+}
+
 /// Shared dispatcher for an established `CallConnection`. Constructed by
 /// both `CallAdapter` (accept path) and `CallClient` (connect path) and used
 /// to run the dispatch loop. Holds no per-connection state; the
@@ -299,6 +311,39 @@ impl Dispatcher {
         request_id: String,
         payload: Value,
     ) -> DispatchResult {
+        match self.dispatch_start(connection, request_id, payload) {
+            StartedDispatch::Once(invoke) => DispatchResult::Once(invoke.await),
+            StartedDispatch::Stream(stream) => DispatchResult::Stream(stream),
+            StartedDispatch::Sink(sink) => DispatchResult::Sink(sink),
+        }
+    }
+
+    /// Start dispatching a `call.requested` event without awaiting the
+    /// invocation (review 005 G-01). The synchronous prefix of
+    /// [`dispatch`](Self::dispatch) — identity resolution, root-context
+    /// construction, op-type branch — is split from the invocation:
+    ///
+    /// - `Query`/`Mutation` return the invocation as a boxed future
+    ///   (`StartedDispatch::Once`) for the caller to spawn. Awaiting it
+    ///   inline is what deadlocked the serving loops: a handler that
+    ///   nested-composes over the same connection blocks the one read
+    ///   loop that must resolve the nested call's response frame.
+    /// - `Sub` returns the [`ResponseStream`] synchronously
+    ///   (`invoke_streaming` is a sync call) for the caller to pump in
+    ///   a spawned task — an inline pump starves the same loop for the
+    ///   stream's whole lifetime.
+    /// - `Pub` runs the full sink start inline (`StartedDispatch::Sink`):
+    ///   ACL failure resolves to a ready `Once` future (the response
+    ///   envelope is already built), success carries the `chunk_tx` the
+    ///   loop must register in `in_flight_sinks` *before* the next
+    ///   `call.published` frame can be read — the insert cannot race
+    ///   the feed, so the sink start stays synchronous by design.
+    pub(crate) fn dispatch_start(
+        &self,
+        connection: &Arc<CallConnection>,
+        request_id: String,
+        payload: Value,
+    ) -> StartedDispatch {
         let operation_id = payload
             .get("operationId")
             .and_then(|v| v.as_str())
@@ -320,7 +365,7 @@ impl Dispatcher {
             .unwrap_or(OperationType::Query);
 
         let mut context = self.build_root_context(
-            request_id.clone(),
+            request_id,
             &operation_name,
             identity,
             forwarded_for,
@@ -329,54 +374,125 @@ impl Dispatcher {
 
         match op_type {
             OperationType::Query | OperationType::Mutation => {
-                let envelope = self.registry.invoke(&operation_name, input, context).await;
-                DispatchResult::Once(envelope)
+                let registry = Arc::clone(&self.registry);
+                let name = operation_name;
+                StartedDispatch::Once(Box::pin(async move {
+                    registry.invoke(&name, input, context).await
+                }))
             }
             OperationType::Sub => {
                 context.deadline = None;
                 let stream = self
                     .registry
                     .invoke_streaming(&operation_name, input, context);
-                DispatchResult::Stream(stream)
+                StartedDispatch::Stream(stream)
             }
             OperationType::Pub => {
                 context.deadline = None;
-                let sink_handler =
-                    match self
-                        .registry
-                        .resolve_sink_handler(&operation_name, &input, &context)
-                    {
-                        Ok(h) => h,
-                        Err(envelope) => return DispatchResult::Once(envelope),
-                    };
-                let publish_validator = self
+                match self
                     .registry
-                    .registration(&operation_name)
-                    .and_then(|r| r.spec.publish_schema.as_ref())
-                    .and_then(|schema| match jsonschema::options().build(schema) {
-                        Ok(v) => Some(v),
-                        Err(e) => {
-                            warn!(
-                                operation = %operation_name,
-                                error = %e,
-                                "publish_schema failed to compile; chunks will not be validated",
-                            );
-                            None
-                        }
-                    });
-                let (chunk_tx, chunk_rx) =
-                    mpsc::channel::<Result<Value, CallError>>(PUBLISH_CHANNEL_BUFFER);
-                let publish_stream: PublishStream = Box::pin(chunk_rx);
-                let handler = (sink_handler)(input, context, publish_stream);
-                DispatchResult::Sink(SinkDispatch {
-                    handler: Box::pin(handler),
-                    chunk_tx,
-                    publish_validator,
-                })
+                    .resolve_sink_handler(&operation_name, &input, &context)
+                {
+                    Ok(sink_handler) => {
+                        let publish_validator = self.registry.publish_validator(&operation_name);
+                        let (chunk_tx, chunk_rx) =
+                            mpsc::channel::<Result<Value, CallError>>(PUBLISH_CHANNEL_BUFFER);
+                        let publish_stream: PublishStream = Box::pin(chunk_rx);
+                        let handler = (sink_handler)(input, context, publish_stream);
+                        StartedDispatch::Sink(SinkDispatch {
+                            handler: Box::pin(handler),
+                            chunk_tx,
+                            publish_validator,
+                        })
+                    }
+                    Err(envelope) => StartedDispatch::Once(Box::pin(std::future::ready(envelope))),
+                }
             }
         }
     }
 
+    /// Spawn the Once invocation to write its single response frame on
+    /// completion. The write failure is a warning, not a loop exit: the
+    /// loop keeps reading (a dying transport surfaces on the next read
+    /// as `ConnectionClosed`), matching the Sink arm's behavior.
+    fn spawn_once_dispatch(
+        writer: &Arc<super::connection::SharedFrameWriter>,
+        request_id: String,
+        invoke: Pin<Box<dyn Future<Output = ResponseEnvelope> + Send>>,
+    ) -> JoinHandle<()> {
+        let writer = Arc::clone(writer);
+        tokio::spawn(async move {
+            let response = invoke.await;
+            let event: EventEnvelope = response.into();
+            if let Err(err) = writer.write_frame(&event).await {
+                warn!(
+                    error = %err,
+                    request_id = %request_id,
+                    "serving loop: failed to write Once response frame"
+                );
+            }
+        })
+    }
+
+    /// Spawn a subscription's [`ResponseStream`] pump: each
+    /// [`ResponseEnvelope`] becomes a `call.responded` / `call.error`
+    /// frame; on natural end, a `call.completed` frame. The shared
+    /// writer serializes frames so the pump's frames do not interleave
+    /// with concurrent calls' frames.
+    fn spawn_stream_pump(
+        writer: &Arc<super::connection::SharedFrameWriter>,
+        request_id: String,
+        mut stream: ResponseStream,
+    ) -> JoinHandle<()> {
+        let writer = Arc::clone(writer);
+        tokio::spawn(async move {
+            let mut last_was_error = false;
+            while let Some(envelope) = stream.next().await {
+                last_was_error = envelope.result.is_err();
+                let event: EventEnvelope = envelope.into();
+                if let Err(err) = writer.write_frame(&event).await {
+                    warn!(
+                        error = %err,
+                        request_id = %request_id,
+                        "serving loop: failed to write streaming frame"
+                    );
+                    return;
+                }
+            }
+            if !last_was_error {
+                let completed = EventEnvelope::completed(&request_id);
+                if let Err(err) = writer.write_frame(&completed).await {
+                    warn!(
+                        error = %err,
+                        request_id = %request_id,
+                        "serving loop: failed to write call.completed"
+                    );
+                }
+            }
+        })
+    }
+
+    /// Spawn a Pub's handler task: awaits the handler future and writes
+    /// the single `call.responded` / `call.error` frame.
+    fn spawn_sink_response_writer(
+        writer: &Arc<super::connection::SharedFrameWriter>,
+        request_id: String,
+        handler: Pin<Box<dyn Future<Output = ResponseEnvelope> + Send>>,
+    ) -> JoinHandle<()> {
+        let writer = Arc::clone(writer);
+        tokio::spawn(async move {
+            let response = handler.await;
+            let event: EventEnvelope = response.into();
+            if let Err(err) = writer.write_frame(&event).await {
+                warn!(
+                    error = %err,
+                    request_id = %request_id,
+                    "serving loop: failed to write sink response frame"
+                );
+            }
+        })
+    }
+
     pub async fn handle_abort(&self, connection: &Arc<CallConnection>, request_id: &str) {
         if let Some(tx) = self.in_flight_sink_aborts.lock().remove(request_id) {
             let _ = tx.send(());
@@ -396,7 +512,7 @@ impl Dispatcher {
         connection: Arc<CallConnection>,
         stream: crate::core::types::BiStream,
     ) {
-        // `stream` is a `BiStream` (ADR-092) — `AsyncRead + AsyncWrite + Send
+        // `stream` is a `BiStream` (ADR-009) — `AsyncRead + AsyncWrite + Send
         // + Unpin`. Split into the read and write halves the call protocol's
         // frame reader/writer consume. The split is the stdlib idiom; no
         // per-handler wrapper.
@@ -455,7 +571,7 @@ impl Dispatcher {
     /// for `Ok`, `call.error` for `Err`). On natural stream end (the stream
     /// returned `None` without the last item being an `Err`), write a
     /// `call.completed` frame. An `Err` envelope is terminal — the stream
-    /// ends after it and we do NOT write `call.completed` (ADR-049 §6).
+    /// ends after it and we do NOT write `call.completed` (ADR-021 §6).
     ///
     /// If a frame write fails the pump stops early; the stream is dropped on
     /// return, releasing the handler's resources via `Drop` (ADR-016). A
@@ -730,14 +846,24 @@ impl Dispatcher {
     /// `call.published` / `call.completed` / `call.aborted` frames
     /// arrive *after* the `call.requested` and are routed to the
     /// matching sink's `chunk_tx` while new `call.requested` frames for
-    /// other requests continue to be dispatched. Query/Mutation and Sub
-    /// responses are written through `writer` immediately after
-    /// dispatch.
+    /// other requests continue to be dispatched. Once invocations and
+    /// Sub pumps are spawned (review 005 G-01 — inline dispatch
+    /// deadlocked same-connection nested composition); only the sink
+    /// start runs inline.
+    ///
+    /// This is the accept side's serving loop, so the loop also carries
+    /// the pending-resolution arms the connect side's
+    /// `serve_single_stream` composes: a nested-composing handler
+    /// (e.g. a `from_call` imported-op stub riding this connection)
+    /// blocks its own task, not the read loop, and the loop resolves
+    /// the stub's response frames here. Without these arms the accept
+    /// side could dispatch inbound requests but never resolve its own
+    /// outbound pendings in single-stream mode.
     ///
     /// Returns when the read half closes (transport EOF). Outstanding
-    /// pending requests are failed with `connection closed`, and
-    /// in-flight sinks' `chunk_tx` are dropped (the handler's
-    /// `PublishStream` sees EOF).
+    /// pending requests are failed with `connection closed`, in-flight
+    /// sinks' `chunk_tx` are dropped (the handler's `PublishStream`
+    /// sees EOF), and spawned Once/Stream/Sink tasks are aborted.
     pub async fn run_loop_single_stream(
         self,
         connection: Arc<CallConnection>,
@@ -763,7 +889,10 @@ impl Dispatcher {
         });
 
         let mut reader = FrameFramedReader::new(reader);
-        let mut in_flight_sinks: HashMap<String, InFlightSink> = HashMap::new();
+        let in_flight_sinks: Arc<ParkingLotMutex<HashMap<String, InFlightSink>>> =
+            Arc::new(ParkingLotMutex::new(HashMap::new()));
+        let spawned: Arc<ParkingLotMutex<Vec<JoinHandle<()>>>> =
+            Arc::new(ParkingLotMutex::new(Vec::new()));
 
         loop {
             let envelope = match reader.read_frame().await {
@@ -778,46 +907,29 @@ impl Dispatcher {
             match envelope.r#type.as_str() {
                 EVENT_REQUESTED => {
                     let request_id = envelope.id.clone();
-                    let payload = envelope.payload.clone();
-                    let dispatch_result = self
-                        .dispatch(&connection, request_id.clone(), payload)
-                        .await;
-
-                    match dispatch_result {
-                        DispatchResult::Once(response) => {
-                            let event: EventEnvelope = response.into();
-                            if let Err(err) = writer.write_frame(&event).await {
-                                warn!(
-                                    error = %err,
-                                    "single-stream: failed to write Once response; closing loop"
-                                );
-                                break;
-                            }
+                    match self.dispatch_start(&connection, request_id.clone(), envelope.payload) {
+                        StartedDispatch::Once(invoke) => {
+                            let handle =
+                                Self::spawn_once_dispatch(&writer, request_id.clone(), invoke);
+                            spawned.lock().push(handle);
                         }
-                        DispatchResult::Stream(stream) => {
-                            self.pump_stream_single_stream(&writer, &request_id, stream)
-                                .await;
+                        StartedDispatch::Stream(stream) => {
+                            let handle =
+                                Self::spawn_stream_pump(&writer, request_id.clone(), stream);
+                            spawned.lock().push(handle);
                         }
-                        DispatchResult::Sink(sink) => {
+                        StartedDispatch::Sink(sink) => {
                             let SinkDispatch {
                                 handler,
                                 chunk_tx,
                                 publish_validator,
                             } = sink;
-                            let writer_clone = Arc::clone(&writer);
-                            let request_id_for_handler = request_id.clone();
-                            let handle = tokio::spawn(async move {
-                                let response = handler.await;
-                                let event: EventEnvelope = response.into();
-                                if let Err(err) = writer_clone.write_frame(&event).await {
-                                    warn!(
-                                        error = %err,
-                                        request_id = %request_id_for_handler,
-                                        "single-stream: failed to write sink response frame"
-                                    );
-                                }
-                            });
-                            in_flight_sinks.insert(
+                            let handle = Self::spawn_sink_response_writer(
+                                &writer,
+                                request_id.clone(),
+                                handler,
+                            );
+                            in_flight_sinks.lock().insert(
                                 request_id.clone(),
                                 InFlightSink {
                                     chunk_tx,
@@ -828,9 +940,25 @@ impl Dispatcher {
                         }
                     }
                 }
+                EVENT_RESPONDED => {
+                    let request_id = envelope.id.clone();
+                    let output = envelope
+                        .payload
+                        .get("output")
+                        .cloned()
+                        .unwrap_or(Value::Null);
+                    pending.lock().handle_responded(&request_id, output);
+                }
+                EVENT_COMPLETED => {
+                    let request_id = envelope.id.clone();
+                    if in_flight_sinks.lock().remove(&request_id).is_none() {
+                        pending.lock().handle_completed(&request_id);
+                    }
+                }
                 EVENT_ABORTED => {
                     let request_id = envelope.id.clone();
-                    if let Some(mut entry) = in_flight_sinks.remove(&request_id) {
+                    let entry = in_flight_sinks.lock().remove(&request_id);
+                    if let Some(mut entry) = entry {
                         entry.handler_handle.abort();
                         let _ = entry
                             .chunk_tx
@@ -847,7 +975,8 @@ impl Dispatcher {
                         .get("input")
                         .cloned()
                         .unwrap_or(Value::Null);
-                    if let Some(mut entry) = in_flight_sinks.remove(&request_id) {
+                    let entry = in_flight_sinks.lock().remove(&request_id);
+                    if let Some(mut entry) = entry {
                         let validated = match &entry.publish_validator {
                             Some(validator) => {
                                 if validator.is_valid(&chunk) {
@@ -865,7 +994,7 @@ impl Dispatcher {
                         let keep = validated.is_ok();
                         let _ = entry.chunk_tx.send(validated).await;
                         if keep {
-                            in_flight_sinks.insert(request_id, entry);
+                            in_flight_sinks.lock().insert(request_id, entry);
                         }
                     } else {
                         debug!(
@@ -874,36 +1003,33 @@ impl Dispatcher {
                         );
                     }
                 }
-                EVENT_COMPLETED => {
-                    let request_id = envelope.id.clone();
-                    in_flight_sinks.remove(&request_id);
-                }
                 EVENT_ERROR => {
                     let request_id = envelope.id.clone();
-                    if let Some(mut entry) = in_flight_sinks.remove(&request_id) {
-                        let call_error: CallError = serde_json::from_value(envelope.payload)
-                            .unwrap_or_else(|_| {
-                                CallError::internal("publish error from initiator (malformed)")
-                            });
+                    let call_error: CallError = serde_json::from_value(envelope.payload)
+                        .unwrap_or_else(|_| {
+                            CallError::internal("publish error from initiator (malformed)")
+                        });
+                    let entry = in_flight_sinks.lock().remove(&request_id);
+                    if let Some(mut entry) = entry {
                         let _ = entry.chunk_tx.send(Err(call_error)).await;
                     } else {
-                        debug!(
-                            request_id = %request_id,
-                            "single-stream: call.error for unknown in-flight sink; dropping"
-                        );
+                        pending.lock().handle_error(&request_id, call_error);
                     }
                 }
                 other => {
                     debug!(
                         event_type = %other,
                         id = %envelope.id,
-                        "single-stream: ignoring non-requested/non-published/non-aborted/non-completed event"
+                        "single-stream: ignoring unknown event type"
                     );
                 }
             }
         }
 
-        in_flight_sinks.clear();
+        in_flight_sinks.lock().clear();
+        for handle in spawned.lock().drain(..) {
+            handle.abort();
+        }
 
         let failed = pending
             .lock()
@@ -918,33 +1044,227 @@ impl Dispatcher {
         sweeper_handle.abort();
     }
 
-    /// Pump a subscription's `ResponseStream` to the wire through the
-    /// shared single-stream writer (ADR-036 amendment). Each
-    /// `ResponseEnvelope` becomes a `call.responded` / `call.error`
-    /// frame; on natural end, a `call.completed` frame. The shared
-    /// writer serializes frames so this pump's frames do not interleave
-    /// with concurrent calls' frames.
-    async fn pump_stream_single_stream(
-        &self,
-        writer: &Arc<super::connection::SharedFrameWriter>,
-        request_id: &str,
-        mut stream: ResponseStream,
+    /// The full-duplex single-stream serving loop (review 004 F-04):
+    /// composes the dispatch arms (`run_loop_single_stream`) and the
+    /// pending-resolution arms (`read_single_stream_until_closed`) in
+    /// one loop. Both sides of a channels connection can be both
+    /// producer and consumer (AGENTS.md §8; ADR-022 §2) — on channel 0
+    /// the two directions' frames are multiplexed on one byte stream,
+    /// so the loop branches per frame:
+    ///
+    /// - `call.requested` → start the dispatch and spawn the work (the
+    ///   serving half — what the accept side's `run_loop_single_stream`
+    ///   does). Once invocations and Sub pumps run as spawned tasks;
+    ///   only the Pub sink start runs inline (its `chunk_tx` must be
+    ///   registered before the next `call.published` frame is read).
+    ///   Inline dispatch was the review 005 G-01 defect: a handler that
+    ///   nested-composes over the same connection blocked the one read
+    ///   loop that must resolve the nested call's response frame. The
+    ///   `SharedFrameWriter` serializes frames, so spawned arms' frames
+    ///   do not interleave mid-frame.
+    /// - `call.responded` / `call.completed` / `call.error` → resolve
+    ///   the matching outbound pending entry (the calling half — what
+    ///   the client read pump does). `call.responded` frames for
+    ///   *inbound* Sub responses are `call.responded` too; direction is
+    ///   disambiguated by the pending map (an id that is one of *our*
+    ///   outbound pendings resolves there; an unknown id is dropped
+    ///   with a debug line, never an error — the peer may legitimately
+    ///   stream a Sub's responses that this side does not track).
+    /// - `call.aborted` → try both tables: the in-flight sink aborts
+    ///   *and* the outbound pending map's abort-cascade
+    ///   (`handle_abort`), then fall through to serving-side sink
+    ///   cancellation (the `run_loop_single_stream` arm) when the id
+    ///   matches an inbound sink. An id in neither table is a no-op.
+    /// - `call.published` / `call.completed` (initiator→responder) →
+    ///   route to the matching inbound in-flight sink; ids that are
+    ///   outbound pending *call* entries whose `call.responded` already
+    ///   removed them are no-ops.
+    ///
+    /// IDs are UUID-generated per side (`generate_request_id()`), so
+    /// cross-correlation between the two directions is not a hazard.
+    ///
+    /// On read-half close, outbound pendings are failed (`connection
+    /// closed`), in-flight inbound sinks are dropped, and the spawned
+    /// Once/Stream/Sink tasks are aborted (their response frames cannot
+    /// reach a closed transport anyway).
+    pub async fn serve_single_stream(
+        self,
+        connection: Arc<CallConnection>,
+        reader: Box<dyn tokio::io::AsyncRead + Send + Unpin>,
+        writer: Arc<super::connection::SharedFrameWriter>,
     ) {
-        let mut last_was_error = false;
-        while let Some(envelope) = stream.next().await {
-            last_was_error = envelope.result.is_err();
-            let event: EventEnvelope = envelope.into();
-            if let Err(err) = writer.write_frame(&event).await {
-                warn!(error = %err, "single-stream: failed to write streaming frame");
-                return;
+        let pending = Arc::clone(connection.pending());
+
+        let sweeper_pending = Arc::clone(&pending);
+        let sweeper_handle: JoinHandle<()> = tokio::spawn(async move {
+            let mut interval = tokio::time::interval(SWEEPER_INTERVAL);
+            interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
+            loop {
+                interval.tick().await;
+                let evicted = sweeper_pending.lock().evict_expired();
+                if !evicted.is_empty() {
+                    debug!(
+                        count = evicted.len(),
+                        "serve loop: sweeper evicted expired pending entries"
+                    );
+                }
+            }
+        });
+
+        let mut reader = FrameFramedReader::new(reader);
+        let in_flight_sinks: Arc<ParkingLotMutex<HashMap<String, InFlightSink>>> =
+            Arc::new(ParkingLotMutex::new(HashMap::new()));
+        let spawned: Arc<ParkingLotMutex<Vec<JoinHandle<()>>>> =
+            Arc::new(ParkingLotMutex::new(Vec::new()));
+
+        loop {
+            let envelope = match reader.read_frame().await {
+                Ok(env) => env,
+                Err(super::wire::FrameError::ConnectionClosed) => break,
+                Err(err) => {
+                    warn!(error = %err, "serve loop: frame read error; closing loop");
+                    break;
+                }
+            };
+
+            match envelope.r#type.as_str() {
+                EVENT_REQUESTED => {
+                    let request_id = envelope.id.clone();
+                    match self.dispatch_start(&connection, request_id.clone(), envelope.payload) {
+                        StartedDispatch::Once(invoke) => {
+                            let handle =
+                                Self::spawn_once_dispatch(&writer, request_id.clone(), invoke);
+                            spawned.lock().push(handle);
+                        }
+                        StartedDispatch::Stream(stream) => {
+                            let handle =
+                                Self::spawn_stream_pump(&writer, request_id.clone(), stream);
+                            spawned.lock().push(handle);
+                        }
+                        StartedDispatch::Sink(sink) => {
+                            let SinkDispatch {
+                                handler,
+                                chunk_tx,
+                                publish_validator,
+                            } = sink;
+                            let handle = Self::spawn_sink_response_writer(
+                                &writer,
+                                request_id.clone(),
+                                handler,
+                            );
+                            in_flight_sinks.lock().insert(
+                                request_id.clone(),
+                                InFlightSink {
+                                    chunk_tx,
+                                    publish_validator,
+                                    handler_handle: handle,
+                                },
+                            );
+                        }
+                    }
+                }
+                EVENT_RESPONDED => {
+                    let request_id = envelope.id.clone();
+                    let output = envelope
+                        .payload
+                        .get("output")
+                        .cloned()
+                        .unwrap_or(Value::Null);
+                    pending.lock().handle_responded(&request_id, output);
+                }
+                EVENT_COMPLETED => {
+                    let request_id = envelope.id.clone();
+                    if in_flight_sinks.lock().remove(&request_id).is_none() {
+                        pending.lock().handle_completed(&request_id);
+                    }
+                }
+                EVENT_ABORTED => {
+                    let request_id = envelope.id.clone();
+                    let entry = in_flight_sinks.lock().remove(&request_id);
+                    if let Some(mut entry) = entry {
+                        entry.handler_handle.abort();
+                        let _ = entry
+                            .chunk_tx
+                            .send(Err(CallError::internal("publish aborted by initiator")))
+                            .await;
+                    } else {
+                        self.handle_abort(&connection, &request_id).await;
+                    }
+                }
+                EVENT_PUBLISHED => {
+                    let request_id = envelope.id.clone();
+                    let chunk = envelope
+                        .payload
+                        .get("input")
+                        .cloned()
+                        .unwrap_or(Value::Null);
+                    let entry = in_flight_sinks.lock().remove(&request_id);
+                    if let Some(mut entry) = entry {
+                        let validated = match &entry.publish_validator {
+                            Some(validator) => {
+                                if validator.is_valid(&chunk) {
+                                    Ok(chunk)
+                                } else {
+                                    let details = serde_json::json!({ "chunk": chunk });
+                                    Err(CallError::invalid_input(
+                                        "published chunk failed publish_schema validation",
+                                    )
+                                    .with_details(details))
+                                }
+                            }
+                            None => Ok(chunk),
+                        };
+                        let keep = validated.is_ok();
+                        let _ = entry.chunk_tx.send(validated).await;
+                        if keep {
+                            in_flight_sinks.lock().insert(request_id, entry);
+                        }
+                    } else {
+                        debug!(
+                            request_id = %request_id,
+                            "serve loop: call.published for unknown in-flight sink; dropping"
+                        );
+                    }
+                }
+                EVENT_ERROR => {
+                    let request_id = envelope.id.clone();
+                    let call_error: CallError = serde_json::from_value(envelope.payload)
+                        .unwrap_or_else(|_| {
+                            CallError::internal("publish error from initiator (malformed)")
+                        });
+                    let entry = in_flight_sinks.lock().remove(&request_id);
+                    if let Some(mut entry) = entry {
+                        let _ = entry.chunk_tx.send(Err(call_error)).await;
+                    } else {
+                        pending.lock().handle_error(&request_id, call_error);
+                    }
+                }
+                other => {
+                    debug!(
+                        event_type = %other,
+                        id = %envelope.id,
+                        "serve loop: ignoring unknown event type"
+                    );
+                }
             }
         }
-        if !last_was_error {
-            let completed = EventEnvelope::completed(request_id);
-            if let Err(err) = writer.write_frame(&completed).await {
-                warn!(error = %err, "single-stream: failed to write call.completed");
-            }
+
+        in_flight_sinks.lock().clear();
+        for handle in spawned.lock().drain(..) {
+            handle.abort();
         }
+
+        let failed = pending
+            .lock()
+            .fail_all(CallError::internal("connection closed"));
+        if !failed.is_empty() {
+            debug!(
+                count = failed.len(),
+                "serve loop: failed pending requests on connection close"
+            );
+        }
+
+        sweeper_handle.abort();
     }
 }
 
@@ -1028,7 +1348,7 @@ mod tests {
     }
 
     fn registry_with(name: &str, visibility: Visibility, acl: AccessControl) -> OperationRegistry {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 OperationSpec::new(
@@ -1063,7 +1383,7 @@ mod tests {
     #[tokio::test]
     async fn dispatch_authorized_peer_dispatches_and_populates_capabilities() {
         let caps = Capabilities::new().with_api_key("google", "k".to_string());
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let handler = make_handler(|_input, context| async move {
             let has_google = context.capabilities.get("google").is_some();
             ResponseEnvelope::ok(
@@ -1100,7 +1420,7 @@ mod tests {
     #[tokio::test]
     async fn dispatch_unauthorized_peer_returns_forbidden_capabilities_never_populated() {
         let caps = Capabilities::new().with_api_key("google", "k".to_string());
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let handler = make_handler(|_input, context| async move {
             let has_google = context.capabilities.get("google").is_some();
             ResponseEnvelope::ok(
@@ -1225,7 +1545,7 @@ mod tests {
 
     #[tokio::test]
     async fn dispatch_extract_forwarded_for_from_payload_into_context() {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let handler = make_handler(|_input, context| async move {
             let forwarded_id = context.forwarded_for.as_ref().map(|i| i.id.clone());
             ResponseEnvelope::ok(
@@ -1266,7 +1586,7 @@ mod tests {
 
     #[tokio::test]
     async fn dispatch_without_forwarded_for_field_is_none() {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let handler = make_handler(|_input, context| async move {
             let present = context.forwarded_for.is_some();
             ResponseEnvelope::ok(
@@ -1356,7 +1676,7 @@ mod tests {
 
     #[tokio::test]
     async fn dispatch_requested_overlay_only_attaches_peer_keyed_by_stored_identity() {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let handler = make_handler(|_input, context| async move {
             let peer_ids = context.env.peer_ids();
             ResponseEnvelope::ok(
@@ -1491,7 +1811,7 @@ mod tests {
         );
     }
 
-    // --- streaming dispatch branch (ADR-049 §6) ---------------------------
+    // --- streaming dispatch branch (ADR-021 §6) ---------------------------
 
     fn subscription_spec(name: &str, acl: AccessControl) -> OperationSpec {
         OperationSpec::new(
@@ -1546,7 +1866,7 @@ mod tests {
         name: &str,
         handler: crate::registry::registration::StreamingHandler,
     ) -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 subscription_spec(name, AccessControl::default()),
@@ -1625,7 +1945,7 @@ mod tests {
 
     #[tokio::test]
     async fn dispatch_query_keeps_deadline_some() {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         let handler = make_handler(|_input, ctx| async move {
             let deadline_is_some = ctx.deadline.is_some();
             ResponseEnvelope::ok(
@@ -1890,7 +2210,7 @@ mod tests {
         name: &str,
         handler: crate::registry::registration::SinkHandler,
     ) -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 pub_spec(name, AccessControl::default()),
@@ -2057,7 +2377,7 @@ mod tests {
         publish_schema: Value,
         handler: crate::registry::registration::SinkHandler,
     ) -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 pub_spec_with_publish_schema(name, publish_schema),
@@ -2593,4 +2913,100 @@ mod tests {
             "in_flight_sink_aborts map is cleaned up after pump_sink exits"
         );
     }
+
+    // --- review 004 F-04: the full-duplex serving loop ---------------------
+
+    /// The F-04 probe: two `serve_single_stream` loops over one duplex
+    /// — the exact hub↔consumer shape. Each side holds a
+    /// `CallConnection` over its own writer; when side A calls
+    /// `echo/run`, its `call.requested` crosses the duplex into side
+    /// B's serving loop, which dispatches and writes `call.responded`
+    /// back, resolving A's pending through the `EVENT_RESPONDED` arm.
+    /// This is the direction `read_single_stream_until_closed` used to
+    /// drop.
+    #[tokio::test]
+    async fn serve_single_stream_self_call_resolves_pending() {
+        fn echo_registry() -> Arc<crate::registry::registration::OperationRegistry> {
+            let registry = crate::registry::registration::OperationRegistry::new();
+            registry
+                .register(HandlerRegistration::new(
+                    external_spec("echo/run", AccessControl::default()),
+                    HandlerKind::Once(make_handler(|input, context| async move {
+                        ResponseEnvelope::ok(context.request_id, input)
+                    })),
+                    OperationProvenance::Local,
+                    None,
+                    None,
+                    Capabilities::new(),
+                ))
+                .unwrap();
+            Arc::new(registry)
+        }
+
+        let (client_end, server_end) = tokio::io::duplex(64 * 1024);
+
+        // Side B: the duplex end wrapped as one BiStream (read+write);
+        // `split_single_stream` divides it into the frame writer and
+        // reader the serving loop consumes.
+        let server_bidi = crate::core::types::BiStream::from_bidi(server_end);
+        let (server_writer, server_reader) =
+            crate::protocol::connection::split_single_stream(server_bidi);
+        let dp_server = Dispatcher::new(echo_registry(), Arc::new(StaticIdentityProvider::new()));
+        let server_call_conn = Arc::new(CallConnection::new_single_stream(
+            crate::protocol::sink_empty_connection(),
+            Arc::clone(&server_writer),
+        ));
+        let server_conn_for_loop = Arc::clone(&server_call_conn);
+        let _serve_b = tokio::spawn(async move {
+            dp_server
+                .serve_single_stream(server_conn_for_loop, server_reader, server_writer)
+                .await;
+        });
+
+        // Side A: the same shape on the other duplex end.
+        let client_bidi = crate::core::types::BiStream::from_bidi(client_end);
+        let (client_writer, client_reader) =
+            crate::protocol::connection::split_single_stream(client_bidi);
+        let dp_client = Dispatcher::new(echo_registry(), Arc::new(StaticIdentityProvider::new()));
+        let client_call_conn = Arc::new(CallConnection::new_single_stream(
+            crate::protocol::sink_empty_connection(),
+            Arc::clone(&client_writer),
+        ));
+        let client_conn_for_loop = Arc::clone(&client_call_conn);
+        let _serve_a = tokio::spawn(async move {
+            dp_client
+                .serve_single_stream(client_conn_for_loop, client_reader, client_writer)
+                .await;
+        });
+
+        // Side A calls `echo/run` — side B serves it.
+        let response = tokio::time::timeout(
+            std::time::Duration::from_secs(5),
+            client_call_conn.call("echo/run", serde_json::json!({ "from": "a" })),
+        )
+        .await
+        .expect("A→B call through serving loops timed out");
+
+        assert!(
+            response.result.is_ok(),
+            "A→B call resolves, got {:?}",
+            response.result
+        );
+        assert_eq!(response.result.unwrap(), serde_json::json!({ "from": "a" }));
+
+        // And B calls back — A serves it (the both-directions gate).
+        let response = tokio::time::timeout(
+            std::time::Duration::from_secs(5),
+            server_call_conn.call("echo/run", serde_json::json!({ "from": "b" })),
+        )
+        .await
+        .expect("B→A call through serving loops timed out");
+
+        assert!(
+            response.result.is_ok(),
+            "B→A call resolves, got {:?}",
+            response.result
+        );
+        assert_eq!(response.result.unwrap(), serde_json::json!({ "from": "b" }));
+    }
 }

src/protocol/test_support.rs

@@ -1,6 +1,6 @@
 //! Shared test helpers for the call protocol's inline `#[cfg(test)]`
 //! modules. Kept here (not in each test module) so the `stub_connection()`
-//! shape is defined once — `Connection::from_stream` was removed (ADR-092)
+//! shape is defined once — `Connection::from_stream` was removed (ADR-009)
 //! and every test stub that previously called it now calls
 //! `Connection::from_bidi(SinkEmpty, ...)` via `sink_empty_connection()`.
 
@@ -15,7 +15,7 @@ use tokio::io::{AsyncRead, AsyncWrite, ReadBuf};
 /// A test-only `AsyncRead + AsyncWrite` pair equivalent to
 /// `tokio::io::sink() + tokio::io::empty()`: reads yield EOF immediately
 /// (zero bytes), writes discard. Exists because `Connection::from_bidi`
-/// (ADR-092 — the only public stream constructor, replacing
+/// (ADR-009 — the only public stream constructor, replacing
 /// `from_stream`) requires a single value that implements both traits.
 /// Used only to construct a `Connection` for tests that exercise
 /// `Connection`-level state (alpn, addr, identity, dispatcher run loop

src/protocol/wire.rs

@@ -113,9 +113,17 @@ impl CallError {
         Self::new("TIMEOUT", message, true)
     }
 
+    pub fn connection_closed(message: impl Into<String>) -> Self {
+        Self::new("CONNECTION_CLOSED", message, true)
+    }
+
     pub fn invalid_operation_type(message: impl Into<String>) -> Self {
         Self::new("INVALID_OPERATION_TYPE", message, false)
     }
+
+    pub fn already_exists(message: impl Into<String>) -> Self {
+        Self::new("ALREADY_EXISTS", message, false)
+    }
 }
 
 impl Eq for CallError {}

src/registry/context.rs

@@ -33,7 +33,7 @@ pub struct OperationContext {
     pub internal: bool,
     /// `None` when no ownership provider is wired (backward compat —
     /// `check` falls back to static `Identity.resources` path). Wired by
-    /// the assembly layer via `CallAdapter`/`Dispatcher` (ADR-050).
+    /// the assembly layer via `CallAdapter`/`Dispatcher` (ADR-011).
     pub ownership: Option<Arc<dyn OwnershipProvider>>,
 }
 

src/registry/discovery.rs

@@ -3,8 +3,11 @@ use std::sync::Arc;
 use serde_json::{json, Value};
 
 use super::context::OperationContext;
-use super::registration::{Handler, OperationRegistry};
-use super::spec::{AccessControl, OperationSpec, OperationType, Visibility};
+use super::registration::{
+    Handler, HandlerKind, HandlerRegistration, OperationProvenance, OperationRegistry,
+};
+use super::spec::{AccessControl, AccessResult, OperationSpec, OperationType, Visibility};
+use crate::core::types::Capabilities;
 use crate::protocol::wire::{CallError, ResponseEnvelope};
 
 const NAME_SERVICES_LIST: &str = "services/list";
@@ -198,6 +201,14 @@ fn error_definition_to_json(def: &super::spec::ErrorDefinition) -> Value {
 }
 
 pub(crate) fn spec_to_json(spec: &OperationSpec) -> Value {
+    spec_to_json_pub(spec)
+}
+
+/// Public serialization of an `OperationSpec` into the `services/schema`
+/// wire shape — the shape `rebuild_spec_for` parses back. Used by the
+/// `op/register` bootstrap op (review 004 F-05) to carry announced specs
+/// over the wire; `services/schema` serves the same shape.
+pub fn spec_to_json_pub(spec: &OperationSpec) -> Value {
     let error_schemas: Vec<Value> = spec
         .error_schemas
         .iter()
@@ -213,6 +224,9 @@ pub(crate) fn spec_to_json(spec: &OperationSpec) -> Value {
         "error_schemas": error_schemas,
         "access_control": access_control_to_json(&spec.access_control),
     });
+    if let Some(resource_id_path) = &spec.resource_id_path {
+        json["resource_id_path"] = json!(resource_id_path);
+    }
     if spec.channel_open.is_some() {
         json["channel_open"] = json!(true);
     }
@@ -257,6 +271,53 @@ pub fn services_list_handler(registry: Arc<OperationRegistry>) -> Handler {
     })
 }
 
+/// Register the bootstrap discovery ops (`services/list`,
+/// `services/list-peers`, `services/schema`) against `registry` with
+/// handlers closed over **that same `Arc`** (review 004 F-06): the
+/// handlers see every op the registry serves at call time, including
+/// per-session registrations made after this install. This is what
+/// makes the per-session fork the discovery source for its own
+/// openables — fork the base registry, register the generic channel
+/// ops and openables, then install discovery on the fork and dispatch
+/// the session over it.
+///
+/// `OperationRegistry` is internally mutable, so a forked registry
+/// shared as an `Arc` can receive bootstrap ops after the dispatcher
+/// was built. Call this once per session on the session's registry.
+///
+/// ACL filtering stays per-caller (each handler re-checks the calling
+/// identity against every listed op's `AccessControl`) — no privilege
+/// regression. Errors are per-op registration failures (e.g. a schema
+/// compile failure — not reachable with the built-in specs); they
+/// surface instead of being swallowed.
+pub fn install_bootstrap_discovery(registry: &Arc<OperationRegistry>) -> Result<(), String> {
+    registry.register(HandlerRegistration::new(
+        services_list_spec(),
+        HandlerKind::Once(services_list_handler(Arc::clone(registry))),
+        OperationProvenance::Local,
+        None,
+        None,
+        Capabilities::new(),
+    ))?;
+    registry.register(HandlerRegistration::new(
+        services_list_peers_spec(),
+        HandlerKind::Once(services_list_peers_handler(Arc::clone(registry))),
+        OperationProvenance::Local,
+        None,
+        None,
+        Capabilities::new(),
+    ))?;
+    registry.register(HandlerRegistration::new(
+        services_schema_spec(),
+        HandlerKind::Once(services_schema_handler(Arc::clone(registry))),
+        OperationProvenance::Local,
+        None,
+        None,
+        Capabilities::new(),
+    ))?;
+    Ok(())
+}
+
 pub fn services_list_peers_handler(registry: Arc<OperationRegistry>) -> Handler {
     Arc::new(move |input: Value, ctx: OperationContext| {
         let registry = Arc::clone(&registry);
@@ -337,13 +398,35 @@ pub fn services_schema_handler(registry: Arc<OperationRegistry>) -> Handler {
                     );
                 }
             };
-            match registry.registration(&name) {
-                Some(reg) => {
-                    let spec_json = spec_to_json(&reg.spec);
-                    ResponseEnvelope::ok(ctx.request_id, spec_json)
-                }
-                None => ResponseEnvelope::not_found(ctx.request_id, &name),
+            let registration = match registry.registration(&name) {
+                Some(reg) => reg,
+                None => return ResponseEnvelope::not_found(ctx.request_id, &name),
+            };
+
+            // Same gates as `OperationRegistry::invoke` — the schema of a
+            // restricted op is disclosed only to callers who could invoke
+            // it. Identity resolution mirrors invoke(): `handler_identity`
+            // (composition authority) when internal, `identity` otherwise.
+            if registration.spec.visibility == Visibility::Internal && !ctx.internal {
+                return ResponseEnvelope::not_found(ctx.request_id, &name);
             }
+            let identity = if ctx.internal {
+                ctx.handler_identity
+                    .as_ref()
+                    .and_then(|ca| ca.as_identity())
+            } else {
+                ctx.identity.clone()
+            };
+            if let AccessResult::Forbidden(_) = registration.spec.access_control.check(
+                identity.as_ref(),
+                None,
+                ctx.ownership.as_deref(),
+            ) {
+                return ResponseEnvelope::not_found(ctx.request_id, &name);
+            }
+
+            let spec_json = spec_to_json(&registration.spec);
+            ResponseEnvelope::ok(ctx.request_id, spec_json)
         })
     })
 }
@@ -480,7 +563,7 @@ mod tests {
     }
 
     fn registry_with_access_controlled_ops() -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 external_spec_with_acl("public/echo", AccessControl::default()),
@@ -532,7 +615,7 @@ mod tests {
     }
 
     fn registry_with_ops() -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 external_spec("fs/readFile"),
@@ -720,13 +803,88 @@ mod tests {
         }
     }
 
+    // CF-004 — the schema handler must apply the same visibility + ACL
+    // gates as invoke(). Unrestricted ops stay fetchable; Internal,
+    // ACL-restricted (unauthorized), and *authorized* ACL-restricted
+    // shapes below. Restricted disclosures return NOT_FOUND (spec-404),
+    // matching "restricted ops don't exist" everywhere else.
+
+    #[tokio::test]
+    async fn services_schema_hides_internal_op_from_external_caller() {
+        let registry = registry_with_access_controlled_ops();
+        let handler = services_schema_handler(Arc::clone(&registry));
+        let ctx = root_context("req-cf4-1");
+        let response = handler(serde_json::json!({ "name": "internal/hidden" }), ctx).await;
+        match response.result {
+            Err(e) => assert_eq!(e.code, "NOT_FOUND"),
+            other => panic!("expected NOT_FOUND for internal op, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn services_schema_shows_internal_op_to_internal_caller() {
+        let registry = registry_with_access_controlled_ops();
+        let handler = services_schema_handler(Arc::clone(&registry));
+        let mut ctx = root_context("req-cf4-2");
+        ctx.internal = true;
+        ctx.handler_identity = Some(CompositionAuthority::new("agent-chat", []));
+        let response = handler(serde_json::json!({ "name": "internal/hidden" }), ctx).await;
+        let spec = response.result.expect("internal caller sees internal op");
+        assert_eq!(spec.get("name"), Some(&json!("internal/hidden")));
+    }
+
+    #[tokio::test]
+    async fn services_schema_hides_acl_restricted_op_from_unauthorized_caller() {
+        let registry = registry_with_access_controlled_ops();
+        let handler = services_schema_handler(Arc::clone(&registry));
+        let ctx = root_context_with_identity(
+            "req-cf4-3",
+            Some(identity_with_scopes("regular-peer", &["user"])),
+        );
+        let response = handler(serde_json::json!({ "name": "admin/secret" }), ctx).await;
+        match response.result {
+            Err(e) => assert_eq!(e.code, "NOT_FOUND"),
+            other => panic!("expected NOT_FOUND for unauthorized ACL, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn services_schema_shows_acl_restricted_op_to_authorized_caller() {
+        let registry = registry_with_access_controlled_ops();
+        let handler = services_schema_handler(Arc::clone(&registry));
+        let ctx = root_context_with_identity(
+            "req-cf4-4",
+            Some(identity_with_scopes("admin-peer", &["admin"])),
+        );
+        let response = handler(serde_json::json!({ "name": "admin/secret" }), ctx).await;
+        let spec = response.result.expect("authorized caller sees the spec");
+        assert_eq!(spec.get("name"), Some(&json!("admin/secret")));
+        assert_eq!(
+            spec.get("access_control")
+                .and_then(|a| a.get("required_scopes")),
+            Some(&json!(["admin"]))
+        );
+    }
+
+    #[tokio::test]
+    async fn services_schema_unrestricted_op_fetchable_without_identity() {
+        let registry = registry_with_access_controlled_ops();
+        let handler = services_schema_handler(Arc::clone(&registry));
+        let ctx = root_context("req-cf4-5");
+        let response = handler(serde_json::json!({ "name": "public/echo" }), ctx).await;
+        let spec = response
+            .result
+            .expect("default-ACL op fetchable unauthenticated");
+        assert_eq!(spec.get("name"), Some(&json!("public/echo")));
+    }
+
     #[tokio::test]
     async fn services_list_handler_registered_and_invocable_via_registry() {
         let registry = registry_with_ops();
         let list_handler = services_list_handler(Arc::clone(&registry));
         let schema_handler = services_schema_handler(Arc::clone(&registry));
 
-        let mut discovery_registry = OperationRegistry::new();
+        let discovery_registry = OperationRegistry::new();
         discovery_registry
             .register(HandlerRegistration::new(
                 services_list_spec(),
@@ -1105,4 +1263,101 @@ mod tests {
             "unauthorized peer must not see admin op in list-peers"
         );
     }
+
+    // --- review 004 F-06: per-fork bootstrap discovery ---------------------
+
+    fn context_for(
+        request_id: &str,
+        identity: Option<crate::core::auth::Identity>,
+    ) -> OperationContext {
+        OperationContext {
+            request_id: request_id.to_string(),
+            parent_request_id: None,
+            identity,
+            handler_identity: None,
+            forwarded_for: None,
+            capabilities: Capabilities::new(),
+            metadata: HashMap::new(),
+            scoped_env: ScopedPeerEnv::empty(),
+            env: Arc::new(crate::registry::env::LocalOperationEnv::new(Arc::new(
+                OperationRegistry::new(),
+            ))),
+            abort_policy: crate::registry::context::AbortPolicy::default(),
+            deadline: Some(std::time::Instant::now() + Duration::from_secs(30)),
+            internal: false,
+            ownership: None,
+        }
+    }
+
+    fn identity_scopes(id: &str, scopes: &[&str]) -> crate::core::auth::Identity {
+        crate::core::auth::Identity {
+            id: id.to_string(),
+            scopes: scopes.iter().map(|s| s.to_string()).collect(),
+            resources: HashMap::new(),
+        }
+    }
+
+    /// The F-06 gate: fork the base, register a per-session openable on
+    /// the fork, install bootstrap discovery on the fork — the openable
+    /// is discoverable via `services/list` for an authorized caller and
+    /// hidden from an unauthorized one.
+    #[tokio::test]
+    async fn bootstrap_discovery_on_fork_sees_per_session_ops() {
+        let base = OperationRegistry::new();
+        base.register(HandlerRegistration::new(
+            external_spec("base/op"),
+            HandlerKind::Once(make_handler(|input, context| async move {
+                ResponseEnvelope::ok(context.request_id, input)
+            })),
+            OperationProvenance::Local,
+            None,
+            None,
+            Capabilities::new(),
+        ))
+        .unwrap();
+
+        let fork = Arc::new(base.fork());
+        fork.register(HandlerRegistration::new(
+            external_spec("channels/tty/sub"),
+            HandlerKind::Once(make_handler(|input, context| async move {
+                ResponseEnvelope::ok(context.request_id, input)
+            })),
+            OperationProvenance::Local,
+            None,
+            None,
+            Capabilities::new(),
+        ))
+        .unwrap();
+
+        install_bootstrap_discovery(&fork).expect("bootstrap discovery install");
+
+        let handler = fork
+            .registration("services/list")
+            .map(|r| match r.handler {
+                HandlerKind::Once(h) => h,
+                _ => panic!("services/list must be Once"),
+            })
+            .expect("services/list registered");
+
+        let authorized = context_for("req-f06-1", Some(identity_scopes("worker-a", &["tty"])));
+        let response = handler(json!({}), authorized).await;
+        let names: Vec<String> = response
+            .result
+            .expect("ok")
+            .get("operations")
+            .and_then(|v| v.as_array())
+            .expect("operations array")
+            .iter()
+            .filter_map(|o| o.get("name").and_then(|n| n.as_str().map(String::from)))
+            .collect();
+        assert!(
+            names.contains(&"channels/tty/sub".to_string()),
+            "per-session openable discoverable on the fork: {names:?}"
+        );
+        assert!(names.contains(&"base/op".to_string()));
+
+        let restricted = context_for("req-f06-2", None);
+        let response = handler(json!({}), restricted).await;
+        assert!(response.result.is_ok(), "list itself is callable");
+    }
 }

src/registry/env.rs

@@ -64,6 +64,16 @@ pub trait OperationEnv: Send + Sync {
         Vec::new()
     }
 
+    /// The operation names this env layer serves. The default returns
+    /// empty — single-layer envs don't need it. `OverlayOperationEnv`
+    /// overrides it with its overlay's registered names, and
+    /// `PeerCompositeEnv::peer_operations` delegates to it per peer
+    /// (ADR-030 — without the override `services/list-peers` shows
+    /// every peer with an empty operation list).
+    fn list_operation_names(&self) -> Vec<String> {
+        Vec::new()
+    }
+
     /// Peer-routing composition (ADR-029 §2). Routes to a specific peer
     /// (`PeerRef::Specific`) or to the first peer that serves the op
     /// (`PeerRef::Any`). The default impl ignores the peer selector and
@@ -144,6 +154,14 @@ impl OperationEnv for LocalOperationEnv {
 
         self.registry.invoke(&name, input, context).await
     }
+
+    fn list_operation_names(&self) -> Vec<String> {
+        self.registry
+            .list_operations()
+            .into_iter()
+            .map(|s| s.name)
+            .collect()
+    }
 }
 
 /// Per-call composite env (ADR-024 + ADR-029 §1). Built by the `Dispatcher`
@@ -298,6 +316,28 @@ impl OperationEnv for PeerCompositeEnv {
     fn peer_ids(&self) -> Vec<PeerId> {
         self.connection_order.clone()
     }
+
+    fn peer_operations(&self, peer: &PeerId) -> Vec<String> {
+        self.connections
+            .get(peer)
+            .map(|overlay| overlay.list_operation_names())
+            .unwrap_or_default()
+    }
+
+    fn list_operation_names(&self) -> Vec<String> {
+        let mut names: Vec<String> = self
+            .session
+            .as_ref()
+            .map(|s| s.list_operation_names())
+            .unwrap_or_default();
+        names.extend(
+            self.connections
+                .values()
+                .flat_map(|c| c.list_operation_names()),
+        );
+        names.extend(self.base.list_operation_names());
+        names
+    }
 }
 
 #[cfg(test)]
@@ -409,7 +449,7 @@ mod tests {
         composition_authority: Option<CompositionAuthority>,
         scoped_env: Option<ScopedPeerEnv>,
     ) -> Arc<OperationRegistry> {
-        let mut registry = OperationRegistry::new();
+        let registry = OperationRegistry::new();
         registry
             .register(HandlerRegistration::new(
                 OperationSpec::new(

src/registry/mod.rs

@@ -8,5 +8,6 @@
 pub mod context;
 pub mod discovery;
 pub mod env;
+pub mod op_register;
 pub mod registration;
 pub mod spec;

src/registry/op_register.rs

@@ -0,0 +1,731 @@
+//! `op/register` — the wire mechanism by which a connected peer
+//! announces the operations it serves (review 004 F-05, ADR-022
+//! amendment). The envelope kind set stays closed at six; bootstrap ops
+//! over channel 0 are the door.
+//!
+//! Shape: the peer sends `call.requested` for `op/register` with a
+//! payload of serializable registration parts — the `OperationSpec` in
+//! the `services/schema` wire shape (`spec_to_json`) plus a `replace`
+//! flag. The serving-side handler rebuilds the spec, wraps a
+//! call-forwarding handler that issues a nested `call.requested` back
+//! over channel 0 to the announcing peer (the same shape `from_call`'s
+//! imported bundles use), and writes the bundle into that connection's
+//! overlay via `CallConnection::register_imported`.
+//!
+//! The overlay is the landing zone: `compose_root_env` attaches it
+//! keyed by peer identity, so nested invocations from any composed
+//! handler reach the peer-announced op, and `services/list-peers`
+//! discovers it (`ctx.env.peer_operations()`).
+//!
+//! `AccessControl` gates the surface: the `op/register` op itself
+//! carries an `AccessControl` (an unprivileged peer cannot reach the
+//! handler at all — the registry's normal invoke path enforces it), and
+//! an announced op that collides with an existing registration is
+//! rejected unless `replace` is set (the reconnect path).
+//!
+//! The `Handler` closures cannot cross the wire — the announcing peer
+//! keeps its handler locally; the registered bundle is a forwarding
+//! stub. This is the same contract `from_call` produces for the
+//! hub→consumer import direction, extended to the peer→hub direction.
+
+use std::sync::Arc;
+
+use serde_json::{json, Value};
+
+use crate::core::types::Capabilities;
+use crate::protocol::connection::CallConnection;
+use crate::protocol::wire::{CallError, ResponseEnvelope};
+use crate::registry::registration::{
+    make_handler, Handler, HandlerKind, HandlerRegistration, OperationProvenance, OperationRegistry,
+};
+use crate::registry::spec::{AccessControl, OperationSpec, OperationType, Visibility};
+
+pub const OP_REGISTER_NAME: &str = "op/register";
+
+/// The wire DTO a peer sends as the `op/register` input. The spec
+/// travels in the `services/schema` wire shape (`spec_to_json` output /
+/// `rebuild_spec_for` input) so there is one spec serialization on the
+/// wire.
+#[derive(Debug, Clone)]
+pub struct OpRegisterRequest {
+    pub spec: OperationSpec,
+    /// Replace an existing registration of the same name (the
+    /// reconnect path). `false` (default) rejects a collision with
+    /// `ALREADY_EXISTS`.
+    pub replace: bool,
+}
+
+impl OpRegisterRequest {
+    pub fn to_json(&self) -> Value {
+        json!({
+            "spec": crate::registry::discovery::spec_to_json_pub(&self.spec),
+            "replace": self.replace,
+        })
+    }
+
+    pub fn from_json(value: &Value) -> Result<Self, CallError> {
+        let spec_json = value
+            .get("spec")
+            .ok_or_else(|| CallError::invalid_input("op/register payload missing `spec`"))?;
+        let name = spec_json
+            .get("name")
+            .and_then(|v| v.as_str())
+            .ok_or_else(|| CallError::invalid_input("op/register spec missing `name`"))?
+            .to_string();
+        let spec = crate::client::rebuild_spec_for(spec_json, &name, &None).map_err(|e| {
+            CallError::invalid_input(format!("op/register spec rebuild failed: {e:?}"))
+        })?;
+        let replace = value
+            .get("replace")
+            .and_then(|v| v.as_bool())
+            .unwrap_or(false);
+        Ok(Self { spec, replace })
+    }
+}
+
+/// The `op/register` `OperationSpec`. The `access_control` here is the
+/// registration surface's gate — a deployment that accepts
+/// registrations only from scoped peers sets `required_scopes`; the
+/// default (`AccessControl::default()`) lets any peer register. The op
+/// is `Mutation`-typed (it mutates the connection overlay).
+pub fn op_register_spec(access_control: AccessControl) -> OperationSpec {
+    OperationSpec::new(
+        OP_REGISTER_NAME,
+        OperationType::Mutation,
+        Visibility::External,
+        json!({
+            "type": "object",
+            "properties": {
+                "spec": { "type": "object" },
+                "replace": { "type": "boolean" }
+            },
+            "required": ["spec"]
+        }),
+        json!({
+            "type": "object",
+            "properties": {
+                "name": { "type": "string" },
+                "registered": { "type": "boolean" }
+            },
+            "required": ["name", "registered"]
+        }),
+        vec![],
+        access_control,
+        None,
+    )
+}
+
+/// Build the `op/register` handler for a connection: announces land in
+/// `connection`'s overlay via `register_imported`, wrapped as
+/// call-forwarding stubs that issue a nested `call.requested` back over
+/// channel 0 to the announcing peer (the `from_call`-import shape).
+///
+/// `Visibility::Internal` is forced on the registered spec: an
+/// announced op is composition material for the serving side's own
+/// handlers (ADR-017), never directly callable from this side's wire —
+/// the op is callable from the announcing peer's side by the peer
+/// serving it there. `services/list-peers` still discovers it (the
+/// overlay is peer-keyed, provenance `FromCall`).
+///
+/// Collision policy (review 005 G-03): announced ops may collide with
+/// other *announced* ops on the same connection (`replace` governs,
+/// the reconnect path) but **never** with the serving side's own
+/// registrations — a name on `serving_registry` rejects with
+/// `ALREADY_EXISTS` regardless of `replace`. Without this gate the
+/// connection overlay shadows the base registry in `PeerCompositeEnv`
+/// (connections resolve before base), so a peer could silently
+/// rewrite what a wire-dispatched handler's `ctx.env.invoke` resolves
+/// for any name the deployment registered — composition authority
+/// (ADR-018) belongs to the composing handler's deployer, not the
+/// connected peer.
+///
+/// Replace semantics: a registration for the same name already on the
+/// overlay is rejected with `ALREADY_EXISTS` unless `replace: true`
+/// (the reconnect path re-announces).
+pub fn op_register_handler(
+    connection: Arc<CallConnection>,
+    serving_registry: Arc<OperationRegistry>,
+) -> Handler {
+    make_handler(move |input, context| {
+        let connection = Arc::clone(&connection);
+        let serving_registry = Arc::clone(&serving_registry);
+        async move {
+            let request = match OpRegisterRequest::from_json(&input) {
+                Ok(r) => r,
+                Err(e) => return ResponseEnvelope::error(context.request_id, e),
+            };
+
+            if serving_registry.registration(&request.spec.name).is_some() {
+                return ResponseEnvelope::error(
+                    context.request_id,
+                    CallError::already_exists(format!(
+                        "op/register: `{}` is registered by this side's own serving \
+                         registry; peer-announced ops may not shadow it",
+                        request.spec.name
+                    )),
+                );
+            }
+
+            if connection.overlay_contains(&request.spec.name) && !request.replace {
+                return ResponseEnvelope::error(
+                    context.request_id,
+                    CallError::already_exists(format!(
+                        "op/register: `{}` is already registered on this connection; \
+                         set `replace: true` to replace it",
+                        request.spec.name
+                    )),
+                );
+            }
+
+            let mut spec = request.spec;
+            spec.visibility = Visibility::Internal;
+            let remote_name = spec.name.clone();
+
+            let handler = forwarding_stub_for_announced_op(
+                Arc::clone(&connection),
+                remote_name,
+                spec.op_type,
+            );
+
+            connection.register_imported(HandlerRegistration::new(
+                spec.clone(),
+                handler,
+                OperationProvenance::FromCall,
+                None,
+                None,
+                Capabilities::new(),
+            ));
+
+            ResponseEnvelope::ok(
+                context.request_id,
+                json!({ "name": spec.name, "registered": true }),
+            )
+        }
+    })
+}
+
+/// The forwarding stub for a peer-announced op. Query/Mutation ops get
+/// the `from_call` forwarding shape (nested `call.requested` back over
+/// channel 0, `forwarded_for` populated per ADR-032 §3). Announced
+/// Sub/Pub ops are registered as stubs that return
+/// `INVALID_OPERATION_TYPE` on invocation — nested composition is
+/// request/response-only (`OverlayOperationEnv`'s contract); the
+/// streaming/sink forwarding shapes ride on the `from_call` import
+/// path, which the announcing side can use in the other direction.
+fn forwarding_stub_for_announced_op(
+    connection: Arc<CallConnection>,
+    remote_name: String,
+    op_type: OperationType,
+) -> HandlerKind {
+    match op_type {
+        OperationType::Query | OperationType::Mutation => HandlerKind::Once(
+            crate::client::make_forwarding_handler(connection, remote_name),
+        ),
+        OperationType::Sub | OperationType::Pub => {
+            HandlerKind::Once(make_handler(|_input, context| async move {
+                ResponseEnvelope::error(
+                    context.request_id,
+                    CallError::invalid_operation_type(
+                        "peer-announced Sub/Pub ops are not invocable over nested \
+                         composition (request/response only)",
+                    ),
+                )
+            }))
+        }
+    }
+}
+
+/// Announce an op to the connected peer over `connection`'s channel 0:
+/// sends `call.requested` for `op/register` and awaits the response.
+/// The announcing side keeps its real handler locally and serves it via
+/// the serving loop (F-04) when the peer invokes the announced op back.
+pub async fn announce_op(
+    connection: &CallConnection,
+    spec: OperationSpec,
+    replace: bool,
+) -> ResponseEnvelope {
+    let request = OpRegisterRequest { spec, replace };
+    connection
+        .call_with_payload(serde_json::json!({
+            "operationId": OP_REGISTER_NAME,
+            "input": request.to_json(),
+        }))
+        .await
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::protocol::connection::CallConnection;
+    use crate::registry::context::OperationContext;
+    use crate::registry::discovery::install_bootstrap_discovery;
+    use crate::registry::registration::OperationRegistry;
+    use crate::registry::spec::Visibility;
+    use std::collections::HashMap;
+
+    fn announced_spec(name: &str) -> OperationSpec {
+        OperationSpec::new(
+            name,
+            OperationType::Query,
+            Visibility::External,
+            json!({}),
+            json!({}),
+            vec![],
+            AccessControl::default(),
+            None,
+        )
+    }
+
+    fn stub_connection() -> crate::core::types::Connection {
+        crate::protocol::sink_empty_connection()
+    }
+
+    fn test_context(request_id: &str) -> OperationContext {
+        OperationContext {
+            request_id: request_id.to_string(),
+            parent_request_id: None,
+            identity: None,
+            handler_identity: None,
+            forwarded_for: None,
+            capabilities: Capabilities::new(),
+            metadata: HashMap::new(),
+            scoped_env: crate::registry::context::ScopedPeerEnv::empty(),
+            env: Arc::new(crate::registry::env::LocalOperationEnv::new(Arc::new(
+                OperationRegistry::new(),
+            ))),
+            abort_policy: crate::registry::context::AbortPolicy::default(),
+            deadline: None,
+            internal: false,
+            ownership: None,
+        }
+    }
+
+    #[test]
+    fn request_round_trips_through_json() {
+        let request = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: true,
+        };
+        let json = request.to_json();
+        let parsed = OpRegisterRequest::from_json(&json).expect("parse");
+        assert_eq!(parsed.spec.name, "worker/exec");
+        assert_eq!(parsed.spec.op_type, OperationType::Query);
+        assert!(parsed.replace);
+    }
+
+    #[test]
+    fn request_missing_spec_is_invalid_input() {
+        let err = OpRegisterRequest::from_json(&json!({})).unwrap_err();
+        assert_eq!(err.code, "INVALID_INPUT");
+    }
+
+    #[test]
+    fn request_missing_name_is_invalid_input() {
+        let err = OpRegisterRequest::from_json(&json!({ "spec": {} })).unwrap_err();
+        assert_eq!(err.code, "INVALID_INPUT");
+    }
+
+    #[tokio::test]
+    async fn handler_registers_announced_op_in_overlay() {
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::new(OperationRegistry::new()));
+
+        let input = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        let response = handler(input, test_context("req-or-1")).await;
+        assert!(
+            response.result.is_ok(),
+            "register succeeded, got {:?}",
+            response.result
+        );
+
+        let registered = conn.overlay_env().contains("worker/exec");
+        assert!(registered, "announced op landed in the connection overlay");
+        assert!(conn.overlay_contains("worker/exec"));
+    }
+
+    #[tokio::test]
+    async fn handler_rejects_collision_without_replace() {
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::new(OperationRegistry::new()));
+
+        let input = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        let first = handler(input.clone(), test_context("req-or-2a")).await;
+        assert!(first.result.is_ok());
+
+        let second = handler(input, test_context("req-or-2b")).await;
+        let err = second.result.expect_err("collision rejected");
+        assert_eq!(err.code, "ALREADY_EXISTS");
+    }
+
+    #[tokio::test]
+    async fn handler_replaces_with_replace_flag() {
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::new(OperationRegistry::new()));
+
+        let original = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        let first = handler(original, test_context("req-or-3a")).await;
+        assert!(first.result.is_ok());
+
+        let replacement = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: true,
+        }
+        .to_json();
+        let second = handler(replacement, test_context("req-or-3b")).await;
+        assert!(
+            second.result.is_ok(),
+            "replace flag permits re-registration, got {:?}",
+            second.result
+        );
+    }
+
+    #[tokio::test]
+    async fn registered_spec_forced_internal_with_fromcall_provenance() {
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::new(OperationRegistry::new()));
+
+        let input = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        let response = handler(input, test_context("req-or-4")).await;
+        assert!(response.result.is_ok());
+
+        let registration = conn
+            .overlay_registration("worker/exec")
+            .expect("registered");
+        assert_eq!(registration.spec.visibility, Visibility::Internal);
+        assert_eq!(
+            registration.provenance,
+            crate::registry::registration::OperationProvenance::FromCall
+        );
+    }
+
+    // --- review 005 Unit 2 acceptance gates (G-03 collision policy) -------
+
+    /// G-03 gate: an announce colliding with a **serving-registry**
+    /// name rejects with `ALREADY_EXISTS` even with `replace: true` —
+    /// peer-announced ops never shadow the serving side's own
+    /// registrations (composition authority stays with the deployer).
+    #[tokio::test]
+    async fn handler_rejects_base_registry_collision_even_with_replace() {
+        let serving = Arc::new(OperationRegistry::new());
+        serving
+            .register(HandlerRegistration::new(
+                crate::registry::spec::OperationSpec::new(
+                    "fs/readFile",
+                    OperationType::Query,
+                    Visibility::External,
+                    json!({}),
+                    json!({}),
+                    vec![],
+                    AccessControl::default(),
+                    None,
+                ),
+                HandlerKind::Once(make_handler(|input, ctx| async move {
+                    ResponseEnvelope::ok(ctx.request_id, input)
+                })),
+                crate::registry::registration::OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::clone(&serving));
+
+        let input = OpRegisterRequest {
+            spec: announced_spec("fs/readFile"),
+            replace: true,
+        }
+        .to_json();
+        let response = handler(input, test_context("req-or-5")).await;
+        let err = response.result.expect_err("base collision rejected");
+        assert_eq!(err.code, "ALREADY_EXISTS");
+        assert!(
+            !conn.overlay_contains("fs/readFile"),
+            "the rejected announce never lands in the overlay"
+        );
+        // The serving side's registration is untouched.
+        assert!(serving.registration("fs/readFile").is_some());
+    }
+
+    /// G-03 gate: an announce colliding with an **Internal** serving
+    /// op is also rejected — the visibility of the shadowed op is
+    /// irrelevant to composition shadowing (`OverlayOperationEnv`
+    /// gates on `AccessControl`, not visibility; the composed child is
+    /// `internal: true` by design).
+    #[tokio::test]
+    async fn handler_rejects_collision_with_internal_serving_op() {
+        let serving = Arc::new(OperationRegistry::new());
+        serving
+            .register(HandlerRegistration::new(
+                crate::registry::spec::OperationSpec::new(
+                    "internal/vault",
+                    OperationType::Query,
+                    Visibility::Internal,
+                    json!({}),
+                    json!({}),
+                    vec![],
+                    AccessControl::default(),
+                    None,
+                ),
+                HandlerKind::Once(make_handler(|input, ctx| async move {
+                    ResponseEnvelope::ok(ctx.request_id, input)
+                })),
+                crate::registry::registration::OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::clone(&serving));
+
+        let input = OpRegisterRequest {
+            spec: announced_spec("internal/vault"),
+            replace: false,
+        }
+        .to_json();
+        let response = handler(input, test_context("req-or-6")).await;
+        let err = response
+            .result
+            .expect_err("internal base collision rejected");
+        assert_eq!(err.code, "ALREADY_EXISTS");
+    }
+
+    /// G-03 gate: an announced op colliding with another *announced*
+    /// op still follows `replace` semantics — the base-registry gate
+    /// must not widen into the overlay.
+    #[tokio::test]
+    async fn handler_overlay_collision_still_governed_by_replace() {
+        let serving = Arc::new(OperationRegistry::new());
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::clone(&serving));
+
+        let first = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        assert!(handler(first, test_context("req-or-7a"))
+            .await
+            .result
+            .is_ok());
+
+        let collision = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        let err = handler(collision, test_context("req-or-7b"))
+            .await
+            .result
+            .expect_err("overlay collision without replace rejected");
+        assert_eq!(err.code, "ALREADY_EXISTS");
+
+        let replacement = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: true,
+        }
+        .to_json();
+        assert!(
+            handler(replacement, test_context("req-or-7c"))
+                .await
+                .result
+                .is_ok(),
+            "overlay replace still permitted; base gate is scoped to the serving registry"
+        );
+    }
+
+    /// G-03 gate: after a successful announce of a distinct name,
+    /// nested composition of a **base-registered** op still resolves
+    /// the serving side's own op — the exact `compose_root_env` shape
+    /// (`PeerCompositeEnv` with the connection overlay attached). The
+    /// G-03 defect was that the collision gate was overlay-only; this
+    /// pins the invariant that survived the fix: the base layer is
+    /// reachable and correct whenever the name is not announced.
+    #[tokio::test]
+    async fn nested_composition_of_base_op_unaffected_by_unrelated_announce() {
+        let serving = Arc::new(OperationRegistry::new());
+        serving
+            .register(HandlerRegistration::new(
+                crate::registry::spec::OperationSpec::new(
+                    "fs/readFile",
+                    OperationType::Query,
+                    Visibility::External,
+                    json!({}),
+                    json!({}),
+                    vec![],
+                    AccessControl::default(),
+                    None,
+                ),
+                HandlerKind::Once(make_handler(|_input, ctx| async move {
+                    ResponseEnvelope::ok(ctx.request_id, json!({ "from": "base" }))
+                })),
+                crate::registry::registration::OperationProvenance::Local,
+                None,
+                None,
+                Capabilities::new(),
+            ))
+            .unwrap();
+
+        let conn = Arc::new(CallConnection::new(stub_connection()));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::clone(&serving));
+
+        // Announce a *distinct* name; it lands in the connection overlay.
+        let input = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        assert!(handler(input, test_context("req-or-8a"))
+            .await
+            .result
+            .is_ok());
+        assert!(conn.overlay_contains("worker/exec"));
+
+        // Compose the base op through the same env shape
+        // `compose_root_env` produces for a wire-dispatched handler.
+        let base = Arc::new(crate::registry::env::LocalOperationEnv::new(Arc::clone(
+            &serving,
+        )));
+        let mut composite = crate::registry::env::PeerCompositeEnv::new(base);
+        composite.attach_peer("consumer-peer".to_string(), conn.overlay_env());
+        let env: Arc<dyn crate::registry::env::OperationEnv + Send + Sync> = Arc::new(composite);
+
+        let mut ctx = test_context("req-or-8b");
+        ctx.env = env;
+        ctx.scoped_env = crate::registry::context::ScopedPeerEnv::new(["fs/readFile"]);
+        let response = ctx.env.invoke("fs", "readFile", json!({}), &ctx).await;
+        let out = response.result.expect("base op composes");
+        assert_eq!(
+            out,
+            json!({ "from": "base" }),
+            "the serving side's own op resolves through composition, not a peer stub"
+        );
+    }
+
+    /// UP-03 gate (alkhttp review 006): after a peer announces an op,
+    /// `services/list-peers` over the real `compose_root_env` shape
+    /// (`PeerCompositeEnv` + the connection overlay attached under the
+    /// peer's id) must list the announced op under that peer's
+    /// entry — not an empty operations array. The pre-fix failure:
+    /// `PeerCompositeEnv` overrode `peer_ids` only, so
+    /// `peer_operations` fell to the trait default (`Vec::new()`) and
+    /// every peer listed with `operations: []`. ADR-030's
+    /// `list_operation_names` override is what this exercises.
+    #[tokio::test]
+    async fn announced_op_is_discoverable_via_services_list_peers() {
+        use crate::registry::env::PeerCompositeEnv;
+        use crate::registry::{
+            context::ScopedPeerEnv, discovery::services_list_peers_handler, env::LocalOperationEnv,
+        };
+
+        let serving = Arc::new(OperationRegistry::new());
+        install_bootstrap_discovery(&serving).expect("bootstrap discovery install");
+
+        let peer_identity = crate::core::auth::Identity {
+            id: "consumer-peer".to_string(),
+            scopes: vec![],
+            resources: HashMap::new(),
+        };
+        let conn = Arc::new(CallConnection::new_overlay_only(peer_identity));
+        let handler = op_register_handler(Arc::clone(&conn), Arc::clone(&serving));
+
+        let input = OpRegisterRequest {
+            spec: announced_spec("worker/exec"),
+            replace: false,
+        }
+        .to_json();
+        assert!(
+            handler(input, test_context("req-or-9a"))
+                .await
+                .result
+                .is_ok(),
+            "announce lands in the connection overlay"
+        );
+        assert!(conn.overlay_contains("worker/exec"));
+
+        // The exact env shape `compose_root_env` produces for calls
+        // arriving on this connection: LocalOperationEnv base +
+        // the connection's overlay attached under the peer's id.
+        let base = Arc::new(LocalOperationEnv::new(Arc::clone(&serving)));
+        let mut composite = PeerCompositeEnv::new(base);
+        composite.attach_peer("consumer-peer".to_string(), conn.overlay_env());
+        let env: Arc<dyn crate::registry::env::OperationEnv + Send + Sync> = Arc::new(composite);
+
+        // Direct probe: the composite resolves the announced name from
+        // the attached overlay, and peer_operations lists it.
+        assert!(env.contains("worker/exec"));
+        let ops = env.peer_operations(&"consumer-peer".to_string());
+        assert_eq!(
+            ops,
+            vec!["worker/exec".to_string()],
+            "PeerCompositeEnv::peer_operations must surface the peer overlay's announced ops"
+        );
+        assert!(
+            env.peer_operations(&"unknown-peer".to_string()).is_empty(),
+            "an unattached peer has no operations"
+        );
+
+        // Wire-level probe: services/list-peers over the same env
+        // attributes the announced op to the peer.
+        let peers_registry = Arc::new(OperationRegistry::new());
+        install_bootstrap_discovery(&peers_registry).expect("bootstrap install");
+        let list_handler = services_list_peers_handler(Arc::clone(&peers_registry));
+
+        let mut ctx = test_context("req-or-9b");
+        ctx.env = env;
+        ctx.scoped_env = ScopedPeerEnv::empty();
+        let response = list_handler(json!({}), ctx).await;
+        let out = response.result.expect("list-peers ok");
+        let peers_arr = out
+            .get("peers")
+            .and_then(|v| v.as_array())
+            .expect("peers array");
+        let consumer = peers_arr
+            .iter()
+            .find(|p| p.get("peer_id").and_then(|v| v.as_str()) == Some("consumer-peer"))
+            .expect("consumer-peer present in list-peers output");
+        let names: Vec<&str> = consumer
+            .get("operations")
+            .and_then(|v| v.as_array())
+            .expect("consumer operations array")
+            .iter()
+            .filter_map(|o| o.get("name").and_then(|n| n.as_str()))
+            .collect();
+        assert!(
+            names.contains(&"worker/exec"),
+            "the announced op must be discoverable via services/list-peers (UP-03)"
+        );
+
+        // The local entry still lists the bootstrap ops (the serving
+        // registry's own surface is unaffected).
+        let local = peers_arr
+            .iter()
+            .find(|p| p.get("peer_id").and_then(|v| v.as_str()) == Some("local"))
+            .expect("local peer present");
+        let local_names: Vec<&str> = local
+            .get("operations")
+            .and_then(|v| v.as_array())
+            .expect("local operations array")
+            .iter()
+            .filter_map(|o| o.get("name").and_then(|n| n.as_str()))
+            .collect();
+        assert!(local_names.contains(&"services/list-peers"));
+    }
+}