diff --git a/tasks/implementation/core-consumer-associate.md b/tasks/implementation/core-consumer-associate.md
new file mode 100644
index 0000000..729f8eb
--- /dev/null
+++ b/tasks/implementation/core-consumer-associate.md
@@ -0,0 +1,79 @@
+---
+id: implementation/core-consumer-associate
+name: "Core — consumer ASSOCIATE: AssociateSession + e2e tests"
+status: pending
+depends_on: [implementation/core-consumer-connect, implementation/core-producer-associate]
+scope: moderate
+risk: medium
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement the UDP ASSOCIATE consumer session in `src/consumer.rs`
+per consumer.md §The associate session (POC #2's shape): the typed
+client that opens the channel, receives the sentinel, and speaks the
+framed datagram stage.
+
+Deliverables (port from `/workspace/alksocks-udp-poc/src/consumer.rs`,
+adapted to the crate's wire codec):
+
+1. **`AssociateSession::associate(client)`**: opens the channel,
+   negotiates (NoAuth), sends ASSOCIATE, reads the reply —
+   **receives the sentinel BND.ADDR and never uses it for
+   addressing** (the handle is the channel; ADR-003).
+2. **`send_datagram(dst, bytes)`**: frames once
+   (`[len: u16 BE][addr + data]` via the crate's wire codec — the
+   framing-once mirror) and writes the frame.
+3. **`recv_datagram()`** returns `(source_addr, bytes)` per
+   datagram — **the incremental reader state (mid-frame position +
+   batched-datagram pending queue) persists across calls**: a
+   datagram split across channel chunks survives a caller-driven
+   receive loop (POC #2's finding — the first cut dropped the reader
+   state per call; the 2-byte-prefix chunk test exposed it). Write
+   the persistence test (the POC's chunk-splitting echo shape).
+4. **`shutdown()`** half-closes the stream — the mux's implicit-EOF
+   sentinel reaches the producer's stage as EOF and the association
+   ends (association lifetime = channel lifetime).
+5. **Teardown ownership**: the session owns the channel; on Drop it
+   closes/joins (alktunnels ADR-005 posture). No `Clone`
+   (compile-asserted).
+6. **Tests** (end-to-end through the harness): associate →
+   send/recv round-trips (multiple in-order datagrams, one
+   association); the split-datagram persistence test; empty-payload
+   datagram round-trip; shutdown → the producer's stage ends and a
+   fresh associate succeeds immediately (the POC's test 4, from the
+   consumer side); Drop teardown (drop mid-stream → the producer's
+   stage sees EOF, the connection stays healthy); CONNECT +
+   ASSOCIATE on the same registration (the regression guard
+   extends).
+
+## Acceptance Criteria
+
+- [ ] `AssociateSession` with `associate`/`send_datagram`/
+      `recv_datagram`/`shutdown` per the POC shape
+- [ ] Reader state persists across `recv_datagram` calls (the
+      split-datagram test)
+- [ ] Framing-once mirror (send frames exactly once; recv decodes
+      incrementally)
+- [ ] Teardown ownership on Drop; no `Clone` (compile-asserted)
+- [ ] E2E tests through the harness (round-trips, shutdown, Drop
+      teardown, regression guard)
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/consumer.md §The associate session
+- docs/architecture/decisions/003-udp-handle-swap.md (the consumer
+  mirror invariants)
+- POC source: `/workspace/alksocks-udp-poc/src/consumer.rs`
+  (`UdpAssociation` — the reader-state finding's origin)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-consumer-bind.md b/tasks/implementation/core-consumer-bind.md
new file mode 100644
index 0000000..543835c
--- /dev/null
+++ b/tasks/implementation/core-consumer-bind.md
@@ -0,0 +1,72 @@
+---
+id: implementation/core-consumer-bind
+name: "Core — consumer BIND: BindSession (reply#2 reader) + e2e tests"
+status: pending
+depends_on: [implementation/core-consumer-connect, implementation/core-producer-bind]
+scope: narrow
+risk: medium
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement the BIND consumer session in `src/consumer.rs` per
+consumer.md §The BIND client (POC #5's `read_reply2` sketch): the
+wrapper-aware client that opens the channel, sends BIND, reads
+reply#1, awaits the inbound accept, and reads reply#2 before riding
+the pass-through data plane.
+
+Deliverables (port from `/workspace/alksocks-bind-poc/src/consumer.rs`,
+adapted to the crate's error surface):
+
+1. **`BindSession::bind(client, target)`**: opens the channel,
+   negotiates, sends the BIND command — the forked client
+   `request()` reads reply#1 (carrying BND.ADDR).
+2. **The reply#2 reader (`await_inbound`)**: reply#2 is hand-read
+   with the forked public `read_address` codec (reply#1's codec
+   applied again — **no second codec**, ADR-011's rationale). The
+   session then rides the pass-through data plane
+   (`BindDataPlane`-shaped: `into_inner()`/`get_socket_mut()`).
+3. **Error surface**: a refused BIND (the command gate's 0x02, or a
+   factory failure's reply#1 code) surfaces as a typed session
+   error carrying the RFC reply code; accept-stage failures surface
+   at `await_inbound` as reply#2's mapped code.
+4. **Teardown ownership**: the session owns the channel; on Drop it
+   closes/joins. No `Clone` (compile-asserted).
+5. **Tests** (end-to-end through the harness): bind → reply#1
+   (BND.ADDR as the harness accept advertises) → reply#2 (the peer
+   address) → pass-through echo both directions; the command-gate
+   refusal surfaced as a typed error (a `socks5:open`-only identity
+   cannot bind); factory-failure and accept-stage-failure mappings
+   surfaced; quiet-wait garbage from the client side ends the
+   session; the three-command regression guard from the consumer
+   side.
+
+## Acceptance Criteria
+
+- [ ] `BindSession` with `bind`/`await_inbound`/data-plane accessors
+      per the POC shape
+- [ ] Reply#2 read via the forked `read_address` codec (no second
+      codec)
+- [ ] Typed error surface (reply-code-carrying session errors)
+- [ ] Teardown ownership on Drop; no `Clone` (compile-asserted)
+- [ ] E2E tests through the harness (round-trip, gate refusal,
+      failure mappings, regression guard)
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/consumer.md §The BIND client
+- docs/architecture/decisions/007-bind-in-base.md (the client-side
+  phase reading)
+- POC source: `/workspace/alksocks-bind-poc/src/consumer.rs`
+  (`BindReply1`/`BindDataPlane`/`read_reply2`)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-consumer-connect.md b/tasks/implementation/core-consumer-connect.md
new file mode 100644
index 0000000..74af9a4
--- /dev/null
+++ b/tasks/implementation/core-consumer-connect.md
@@ -0,0 +1,77 @@
+---
+id: implementation/core-consumer-connect
+name: "Core — consumer half: Socks5ChannelSession (CONNECT) + e2e tests"
+status: pending
+depends_on: [implementation/core-producer-connect]
+scope: narrow
+risk: low
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement the CONNECT consumer session in `src/consumer.rs` per
+consumer.md §The Channels-Native Sessions: `Socks5ChannelSession` —
+the typed client that opens `channels/socks5/sub` and drives the
+client side of the RFC conversation in-stream (POC #1's
+`use_stream` shape).
+
+Deliverables (port from `/workspace/alksocks-connect-poc/src/
+consumer.rs`, adapted to the crate's params/error surface):
+
+1. **`Socks5ChannelSession::connect(client, target)`**: opens the
+   channel via `ChannelClient::open_channel(OP_SOCKS5_OPEN, params,
+   SOCKS5_ALPN)` (trivial params — ADR-001), then drives the forked
+   client typestate's `use_stream` path over the channel's
+   `BiStream` (generic over the backing stream — the POC #1 fit:
+   `Socks5Stream` impls `AsyncRead + AsyncWrite` itself, so the
+   session over a `BiStream` is the intended use).
+2. **Target construction** via `ToTargetAddr` (IP-string forms parse
+   to IP targets; domain form passes through unresolved — ADR-012).
+3. **Post-reply payload pipe**: after the reply, the session is a
+   raw byte pipe — expose `split()` (read/write halves) and
+   `into_inner()` (the underlying `Socks5Stream<BiStream>` for
+   advanced use). 0 B framing (CONNECT).
+4. **Teardown ownership** (alktunnels ADR-005 posture, inherited):
+   the session owns the channel; on Drop it closes/joins. No
+   `Clone` (compile-asserted — aliasing a session would alias its
+   teardown; the alktunnels `compile_fail` doctest shape).
+5. **Error surface**: open failures are the typed
+   `SocksOpenError` reasons (ADR-049 §4 — a refused session never
+   yields a phantom channel); a refused target is an RFC reply code
+   surfaced through the session type (consumer.md §Error surface).
+6. **Tests** (end-to-end through the harness): connect → echo
+   round-trip (payload bytes both directions); refused dial → the
+   RFC reply code surfaced as a typed session error; FORBIDDEN open
+   → typed open error, no session; half-close semantics (dropping
+   the write half EOFs the producer's pump cleanly); domain-form
+   target passes through unresolved (assert the wire bytes, the
+   echo dial ignores the target).
+
+## Acceptance Criteria
+
+- [ ] `Socks5ChannelSession` with `connect`/`split`/`into_inner`
+      per the POC shape
+- [ ] Teardown ownership on Drop; no `Clone` (compile-asserted)
+- [ ] Typed error surface (open errors vs reply-code refusals)
+- [ ] E2E tests through the harness (round-trip, refusals,
+      half-close, domain pass-through)
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/consumer.md §The Channels-Native Sessions
+- docs/architecture/decisions/003-udp-handle-swap.md (the framing
+  mirror — CONNECT is the 0 B case)
+- POC source: `/workspace/alksocks-connect-poc/src/consumer.rs`
+- alktunnels template: `/workspace/@alkdev/alktunnels/src/consumer.rs`
+  (the teardown-ownership + no-Clone posture)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-crate-deps.md b/tasks/implementation/core-crate-deps.md
new file mode 100644
index 0000000..5e722c1
--- /dev/null
+++ b/tasks/implementation/core-crate-deps.md
@@ -0,0 +1,81 @@
+---
+id: implementation/core-crate-deps
+name: Core — crate dependencies + module skeleton (alkcall wiring)
+status: pending
+depends_on: []
+scope: narrow
+risk: low
+impact: project
+level: implementation
+---
+
+## Description
+
+Add the dependency surface the wrapper layers need and the module
+skeleton so the batch-1 core tasks have their targets. This is the
+batch-1 analogue of the vendor scaffolding task: the crate gains
+`alkcall` (its substrate — first consumption of the 0.8.0 surface) and
+the module files named in overview.md §Module Map, born wasm-clean.
+
+Deliverables:
+
+1. **`Cargo.toml` `[dependencies]`** — add to the existing trio
+   (tokio wasm-clean subset, tracing, thiserror):
+   - `alkcall = "0.8.0"` (pinned per AGENTS.md convention 11 — bump
+     deliberately, never transitively)
+   - `serde = { version = "1", features = ["derive"] }`,
+     `serde_json = "1"` (the params schema + spec builder)
+   - `bytes = "1"` (the wire codec's buffer types)
+   - `futures = { version = "0.3", default-features = false,
+     features = ["std"] }` (BoxFuture for the seam callbacks — the
+     alktunnels precedent; verify the wasm-clean subset compiles)
+   - tokio stays `default-features = false` with the wasm-clean
+     subset (`rt`, `sync`, `io-util`, `macros`, `time`) — **never
+     `features = ["full"]`** (AGENTS.md convention 4). No
+     `tokio/net`, no `socket2` — those ride `local` in batch 2.
+   - `[features]` stays `default = []`, `local = []`,
+     `composed = []` — declared empty by design (ADR-009); populated
+     by their modules' batch-2 tasks.
+2. **`src/lib.rs`** — keep `#![forbid(unsafe_code)]` and the crate
+   docs; declare the module skeleton per overview.md §Module Map:
+   `params`, `wire`, `producer`, `consumer`, `egress`, `error` —
+   one-line stub files, each with a module doc naming its filling
+   task (the vendor-scaffolding precedent). `rfc1928` stays as is.
+3. **Re-export posture**: `lib.rs` re-exports public API per
+   convention 15 — for now the module declarations only; each
+   implementation task adds its own re-exports.
+
+## Acceptance Criteria
+
+- [ ] `cargo test` passes
+- [ ] `cargo clippy --all-targets -- -D warnings` clean
+- [ ] `cargo fmt --check` clean
+- [ ] `cargo check --target wasm32-unknown-unknown` clean — the wasm
+      gate stays green with alkcall + serde + bytes + futures in the
+      graph (alkcall 0.8.0's default crate is wasm-clean; verify, do
+      not assume)
+- [ ] `cargo clippy --target wasm32-unknown-unknown -- -D warnings`
+      clean
+- [ ] No `tokio/net`, `socket2`, `mio` in the default dependency
+      graph (`cargo tree` spot-check)
+- [ ] `alkcall` pinned at `"0.8.0"` (not a range)
+- [ ] Module skeleton matches overview.md §Module Map (module names
+      and file layout)
+
+## References
+
+- docs/architecture/overview.md §Dependencies, §Module Map
+- docs/architecture/decisions/009-feature-gates.md (the feature
+  layout; no `net` feature)
+- AGENTS.md conventions 4 (tokio subset), 11 (alkcall pin), 15
+  (module structure)
+- Sibling posture: `/workspace/@alkdev/alktunnels/Cargo.toml`
+  (alkcall 0.8.0 consumer; the futures/bytes/serde dep set)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-integration-suite.md b/tasks/implementation/core-integration-suite.md
new file mode 100644
index 0000000..0ac1dc7
--- /dev/null
+++ b/tasks/implementation/core-integration-suite.md
@@ -0,0 +1,86 @@
+---
+id: implementation/core-integration-suite
+name: Core — cross-cutting integration suite + docs sync
+status: pending
+depends_on: [implementation/core-consumer-associate, implementation/core-consumer-bind]
+scope: moderate
+risk: medium
+impact: phase
+level: implementation
+---
+
+## Description
+
+The batch-1 closing task: the cross-cutting integration suite that
+pins the whole protocol core's behavior in one place, plus the docs
+sync that records what landed. The per-command tasks own their
+verticals; this task owns the cross-cutting invariants and the
+spec-vs-implementation reconciliation.
+
+Deliverables:
+
+1. **Cross-cutting integration tests** (`tests/`):
+   - **One registration, all three commands** (the POC #2/#5
+     regression guard, now complete): CONNECT + ASSOCIATE + BIND all
+     served on the same `channels/socks5/sub` op, interleaved
+     sessions on one connection.
+   - **Concurrency**: multiple concurrent sessions of mixed command
+     types over one channels connection (the mux/demux under load —
+     the POCs tested sequentially; this pins concurrency).
+   - **Teardown semantics matrix**: client half-close ends a CONNECT
+     data plane cleanly (the producer's pump exits, the connection
+     stays healthy); ASSOCIATE EOF ends the association; BIND
+     client-gone teardown releases the accept; a fresh session
+     succeeds immediately after each teardown (no connection
+     poisoning).
+   - **Identity/scope matrix**: `socks5:open`-only (CONNECT +
+     ASSOCIATE yes, BIND 0x02), `socks5:bind`-only (BIND yes,
+     CONNECT/ASSOCIATE FORBIDDEN at the open gate), both-scopes
+     (everything) — the two-grants-never-implied posture (ADR-006)
+     pinned end to end.
+   - **The readiness contract** re-pinned at the integration level
+     (the producer-associate task owns the unit-level test; this
+     pins it under concurrent load).
+2. **Docs sync** (the architecture-sync check):
+   - `docs/architecture/README.md` Current State + the doc statuses
+     reflect the landed core (the rfc1928 doc's "remaining layers
+     follow the decomposition phase" line updated).
+   - Any spec drift the implementation surfaced (naming, shapes,
+     the wire.md/bast.md residue already fixed in the wire-codec
+     task) reconciled in the same change — never silently diverged.
+   - `lib.rs` crate docs updated to describe the landed surface
+     (producer + consumer + wire + seams), pointing at the specs.
+3. **Full verification sweep** (the batch gate's input):
+   `cargo test`, `cargo test --all-features` (features are still
+   empty — verify the flag combination compiles), clippy
+   `--all-targets -- -D warnings`, `cargo fmt --check`, both wasm
+   checks, and `cargo doc --no-deps` (docs changed).
+
+## Acceptance Criteria
+
+- [ ] The three-command regression + concurrency + teardown-matrix +
+      identity-matrix tests all pass
+- [ ] Docs reconciled (README Current State; any drift fixed)
+- [ ] The full verification sweep is green (test, all-features,
+      clippy, fmt, both wasm checks, doc)
+- [ ] No `local`/`composed` code exists yet (the batch boundary
+      holds — the default crate is protocol-only)
+
+## References
+
+- docs/architecture/wire.md §Sentinels and lifecycle (the teardown
+  matrix's normative source)
+- docs/architecture/decisions/006-access-control.md (the identity
+  matrix)
+- POC regression guards: `udp_composed_egress_round_trip`-adjacent
+  tests in `/workspace/alksocks-udp-poc`, test 7 in
+  `/workspace/alksocks-bind-poc`
+- docs/sdd_process.md (the architecture-sync-check posture)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-params-errors.md b/tasks/implementation/core-params-errors.md
new file mode 100644
index 0000000..f3d8948
--- /dev/null
+++ b/tasks/implementation/core-params-errors.md
@@ -0,0 +1,88 @@
+---
+id: implementation/core-params-errors
+name: Core — params schema + scope constants + typed error surface
+status: pending
+depends_on: [implementation/core-crate-deps]
+scope: narrow
+risk: low
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement `src/params.rs` and `src/error.rs` per wire.md §The Open
+Op, ADR-001, and ADR-006 — the wire-stable surface (the one-way door)
+and the typed open-error vocabulary. Template:
+`/workspace/@alkdev/alktunnels/src/params.rs` (the sibling's spec
+builder + scope-constant posture).
+
+Deliverables:
+
+1. **`src/params.rs`**:
+   - Constants: `OP_SOCKS5_OPEN = "channels/socks5/sub"`,
+     `SOCKS5_ALPN = "alk/socks5"`, `SOCKS5_OPEN_SCOPE =
+     "socks5:open"`, `SOCKS5_BIND_SCOPE = "socks5:bind"` (ADR-001/
+     ADR-006; the scope-string convention `<alpn-family>:
+     <capability>`; stable once published).
+   - `socks5_open_spec() -> OperationSpec` — `Sub`-typed, external,
+     channel-open marker on `SOCKS5_ALPN`, **trivial params**:
+     input schema `{type: object}`, `additionalProperties: false`
+     (empty params stay valid forever — ADR-001's additive-only
+     extension rule); reply schema `{"channel_id": integer,
+     minimum 1}`; `required_scopes = ["socks5:open"]`. Follow the
+     alktunnels spec-builder shape exactly (the registry validates
+     against these schemas).
+   - The trivial params struct (`Socks5Params {}` — serde
+     `deny_unknown_fields`, empty) for the consumer's open call.
+2. **`src/error.rs`** — the typed open-error surface (alkcall
+   ADR-049 §4):
+   - `SocksOpenError` — the consumer-side open-failure vocabulary
+     mapping the `channel:open_failed` reasons (`dial_failed` /
+     `unknown_resource` / `resource_shortage` / `handler_error` /
+     `timeout`) plus the registry's `FORBIDDEN`/`INVALID_INPUT`
+     classes, shaped on alktunnels' `TunnelOpenError` +
+     `establishment_reason` helper.
+   - The session-level error type for the consumer sessions
+     (`SocksSessionError`-shaped: open errors vs in-stream
+     `SocksError` vs I/O) — the exact variant set is the task's
+     judgment; keep it faithful to what the sessions actually
+     surface (a refused session never yields a phantom channel; a
+     refused target is an RFC reply code surfaced through the
+     session type — consumer.md §Error surface).
+3. **lib.rs re-exports** for the new public surface (convention 15).
+
+Tests: spec-builder shape assertions (op id, ALPN marker, scopes,
+schemas — the trivial-params schema rejects unknown fields), params
+serde round-trip (empty object valid; unknown field rejected), the
+error-reason mapping table.
+
+## Acceptance Criteria
+
+- [ ] Constants + spec builder per ADR-001/ADR-006 exactly (op id,
+      ALPN, both scopes, trivial schemas)
+- [ ] Input schema rejects unknown fields (the loud-rejection
+      posture)
+- [ ] Typed open-error surface maps the ADR-049 §4 reasons
+- [ ] lib.rs re-exports the public surface
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/wire.md §The Open Op (the normative op shape)
+- docs/architecture/decisions/001-alpn-and-open-op.md (trivial
+  params; the one-way door)
+- docs/architecture/decisions/006-access-control.md (the two scopes)
+- docs/architecture/decisions/011-inline-vendor.md §4 (the typed
+  open-error surface)
+- Template: `/workspace/@alkdev/alktunnels/src/params.rs`
+  (`tunnel_open_spec`, the scope constants, `establishment_reason`)
+- POC sketch: `/workspace/alksocks-connect-poc/src/params.rs`
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-producer-associate.md b/tasks/implementation/core-producer-associate.md
new file mode 100644
index 0000000..7ee7c69
--- /dev/null
+++ b/tasks/implementation/core-producer-associate.md
@@ -0,0 +1,93 @@
+---
+id: implementation/core-producer-associate
+name: "Core — producer ASSOCIATE: the select!-driven datagram stage"
+status: pending
+depends_on: [implementation/core-wire-codec, implementation/core-producer-connect]
+scope: moderate
+risk: medium
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement the UDP ASSOCIATE producer path in `src/producer.rs` per
+wire.md §UDP ASSOCIATE and ADR-003: replace the dispatch arm's
+refusal with the datagram stage (POC #2's validated scope).
+
+Deliverables (port from `/workspace/alksocks-udp-poc/src/producer.rs`
++ `egress.rs`, adapted to the crate's wire codec and seam types):
+
+1. **The ASSOCIATE arm**: `reply_success(sentinel)` — the sentinel
+   `127.0.0.1:0` BND.ADDR (pure RFC-compatibility surface; the
+   channel is the relay handle, ADR-003) — then the datagram stage.
+2. **The datagram stage** — `select!`-driven (channel reads ∥ egress
+   replies), with the POC's three load-bearing findings as
+   invariants:
+   - **The concurrent select shape** (not poll-then-block — the
+     poll-then-block window loses replies; the concurrent shape is
+     also the honest full-duplex shape).
+   - **The egress readiness contract**: `next_reply` is `Pending`
+     when nothing is ready — an instantly-ready `Ok(None)` starves a
+     biased select into livelock. **Write the contract test**: an
+     egress returning instant `Ok(None)` must not starve the stage's
+     channel branch (the POC's load-bearing API finding, re-pinned
+     by this suite per ADR-004's "re-pinned by the crate's test
+     suite").
+   - **Framing applied exactly once, by the wire layer**: the stage
+     reads frames via `DatagramReader` (channel → egress `send`
+     gets the payload) and writes frames via `frame()` (egress
+     `next_reply`'s inner bytes → the channel). The egress never
+     sees a length prefix.
+   - **Association lifetime = channel lifetime**: channel EOF ends
+     the stage; no extra control vocabulary. A undecodable frame
+     (garbage ATYP) fails the association loud
+     (`udp_undecodable_frame_ends_association`).
+3. **The `EgressFactory` lifecycle**: called once per ASSOCIATE; the
+   stage owns the egress; stage exit drops it. Ownership rides the
+   handle structurally — no `_` bindings of owning values (the POC's
+   teardown-at-birth witness).
+4. **Missing-egress refusal**: a registration without an
+   `EgressFactory` refuses ASSOCIATE RFC-faithfully
+   (`CommandNotSupported`-class) — never a panic (ADR-008).
+5. **Tests** (via the harness + raw RFC client): associate →
+   sentinel reply → datagram round-trip through the echo egress
+   (in-order, multiple datagrams, one association); empty-payload
+   datagram legal (not EOF); undecodable frame ends the association
+   loud; channel EOF ends the association and the connection stays
+   healthy (a fresh associate succeeds immediately after — the POC's
+   test 4); the readiness-contract test; missing-egress refusal;
+   CONNECT still served on the same registration (the regression
+   guard).
+
+## Acceptance Criteria
+
+- [ ] The datagram stage per the three POC invariants (concurrent
+      select, readiness contract, framing-once)
+- [ ] Sentinel BND.ADDR in the reply; association lifetime = channel
+      lifetime (EOF teardown tested)
+- [ ] Readiness-contract test present (instant-Ok(None) egress does
+      not starve the stage)
+- [ ] Missing-egress refusal RFC-faithful
+- [ ] CONNECT regression guard passes (one registration, both
+      commands)
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/wire.md §UDP ASSOCIATE (the normative stage)
+- docs/architecture/decisions/003-udp-handle-swap.md (the handle
+  swap; the readiness + framing invariants)
+- POC sources: `/workspace/alksocks-udp-poc/src/producer.rs`
+  (`serve_udp_associate`), `/workspace/alksocks-udp-poc/src/
+  egress.rs` (the readiness findings' origin)
+- docs/architecture/decisions/008-provider-defaults.md (the
+  missing-callback posture)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-producer-bind.md b/tasks/implementation/core-producer-bind.md
new file mode 100644
index 0000000..6af8a9f
--- /dev/null
+++ b/tasks/implementation/core-producer-bind.md
@@ -0,0 +1,103 @@
+---
+id: implementation/core-producer-bind
+name: "Core — producer BIND: the phase model + the socks5:bind command gate"
+status: pending
+depends_on: [implementation/core-producer-connect]
+scope: moderate
+risk: medium
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement the TCP BIND producer path in `src/producer.rs` per
+wire.md §BIND and ADR-007: replace the dispatch arm's refusal with
+the BIND phase model (POC #5's validated scope), plus the
+`socks5:bind` command gate (ADR-006).
+
+Deliverables (port from `/workspace/alksocks-bind-poc/src/producer.rs`
++ `accept.rs`, adapted to the crate's seam types):
+
+1. **The command gate**: at BIND dispatch, require `socks5:bind`
+   from the per-call opener identity (the open-op `AuthContext`'s
+   identity — CF-005/CF-006). An identity without it gets the
+   RFC-faithful `ConnectionNotAllowed` (0x02) refusal while CONNECT
+   still serves the same identity (POC #5 test 4). The gate is the
+   open-op identity seam consumed at command-dispatch time — no new
+   ops, no params change.
+2. **The BIND phase model**:
+   - **reply#1 carries BND.ADDR** — the `BindAcceptFactory` is
+     called first; its `BindAccept.bind_addr` is reply#1's carrier
+     (the local variant's real listener; the sentinel on the
+     wrapper-aware in-band path when no factory is wired — ADR-007
+     variant 3).
+   - **The quiet wait**: `select!`-single-shot with exactly two
+     exits — the one inbound accept (the `BindAccept` future), or
+     the control stream's EOF/garbage/error (the client-gone
+     teardown riding the accept future's drop). **Not a loop** (the
+     POC's `never_loop` finding — exactly two exits). Any byte in
+     the quiet wait is protocol garbage → the session ends with no
+     data plane.
+   - **reply#2 carries the peer address** (the accepted handle's
+     peer; sentinel `0.0.0.0:0` wrapper-aware).
+   - **Phase 2 is raw pass-through** (`pump_bidi` inline — the
+     CONNECT shape) between the channel and the accepted halves.
+3. **Failure mapping is RFC-faithful end to end**: factory failure
+   (pre-reply#1) → reply#1 with the mapped code; accept-stage
+   failure (post-reply#1) → reply#2 with the mapped code, no data
+   plane. No phantom listeners, no phantom data planes.
+4. **The establishment bound covers only the open-op round trip**
+   (ADR-007): the accept wait lives in the teardown-governed pump
+   phase — the quiet wait must not run under the establisher's
+   10s bound (FTP-length BIND waits stop resolving as spurious
+   establishment refusals).
+5. **Missing-accept refusal**: a registration without a
+   `BindAcceptFactory` serves the wrapper-aware in-band default
+   (sentinel reply#1 + sentinel reply#2 — ADR-007 variant 3) or
+   refuses — the ADR names variant 3 as the in-band default when
+   neither provider is wired; implement variant 3 as the
+   no-factory default and document the choice.
+6. **Tests** (via the harness + raw RFC client): BIND → reply#1 →
+   (the harness accept resolves) → reply#2 with the peer →
+   pass-through echo both directions; the command gate (a
+   `socks5:open`-only identity gets 0x02 on BIND while CONNECT
+   still works — the POC's test 4); quiet-wait garbage byte ends
+   the session with no data plane; client-gone teardown (EOF
+   mid-wait drops the accept future — a fresh dial to the
+   advertised address fails); factory failure → reply#1 mapped
+   code; accept-stage failure → reply#2 mapped code, no data plane;
+   CONNECT + ASSOCIATE regression guard (one registration, all
+   three commands).
+
+## Acceptance Criteria
+
+- [ ] The command gate: `socks5:bind` required at BIND dispatch;
+      `ConnectionNotAllowed` refusal; CONNECT unaffected (tested)
+- [ ] The phase model: reply#1 → quiet wait (single-shot, two
+      exits) → accept → reply#2 → `pump_bidi` pass-through
+- [ ] Two-stage failure mapping (factory → reply#1; accept-stage →
+      reply#2), tested
+- [ ] The accept wait is NOT under the establishment bound
+- [ ] No-factory default = wrapper-aware variant 3 (documented)
+- [ ] Three-command regression guard passes
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/wire.md §BIND (the normative phases)
+- docs/architecture/decisions/007-bind-in-base.md (the phase model +
+  the three variants)
+- docs/architecture/decisions/006-access-control.md (the command
+  gate)
+- POC sources: `/workspace/alksocks-bind-poc/src/producer.rs`,
+  `/workspace/alksocks-bind-poc/src/accept.rs` (the quiet-wait +
+  failure-mapping origin)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-producer-connect.md b/tasks/implementation/core-producer-connect.md
new file mode 100644
index 0000000..6cb01eb
--- /dev/null
+++ b/tasks/implementation/core-producer-connect.md
@@ -0,0 +1,102 @@
+---
+id: implementation/core-producer-connect
+name: "Core — producer half: establisher + dispatch + registration + CONNECT"
+status: pending
+depends_on: [implementation/core-seam-types, implementation/core-params-errors, implementation/core-test-harness]
+scope: moderate
+risk: medium
+impact: phase
+level: implementation
+---
+
+## Description
+
+Implement the producer half of `src/producer.rs` per producer.md:
+the session-gate establisher, the command-dispatch pump handler, the
+registration surface, and the CONNECT data plane (POC #1's validated
+scope). This is the crate's core task — the open op all three
+commands ride.
+
+Deliverables (port from `/workspace/alksocks-connect-poc/src/
+producer.rs` + `/workspace/alksocks-udp-poc/src/producer.rs`, adapted
+to the crate's seam types and alkcall 0.8.0):
+
+1. **The session-gate establisher** (`socks5_session_establisher`):
+   validates/records the opener identity (the CF-005/CF-006 seam —
+   ADR-005) and applies session-level policy. Runs
+   `Establishment::default()`-shaped (no plan payload — the dial
+   happens post-establishment, so there is nothing to thread
+   establisher→handler; POC #1's wire note). Optional
+   `identity_witness` probe seam (the alktunnels posture) for tests.
+2. **The pump handler** (`make_socks5_pump_handler(dial, egress,
+   accept)`): `accept_bi()` → drive the forked typestate over the
+   `BiStream` (`accept_no_auth → read_command`) → dispatch:
+   - **CONNECT** → invoke `DialFn(target)` at the command-read
+     interception point; success: `reply_success` then `pump_bidi`
+     **awaited inline** (R-02 — the returned `JoinHandle` tracks the
+     data-plane lifetime; early return = teardown-at-birth). Refusal:
+     `reply_error` with the mapped RFC code (dial-failure →
+     `ConnectionRefused`; scope-class → `ConnectionNotAllowed`) — no
+     phantom data plane.
+   - **UDP ASSOCIATE / TCP BIND** → this task lands the dispatch
+     arms as RFC-faithful refusals (`CommandNotSupported`); the
+     associate/bind producer tasks replace them. (The dispatch
+     skeleton exists from day one so the regression guard — one
+     registration, all three commands — is testable early.)
+   - The handler is substrate-agnostic by construction — it sees
+     boxed halves, never a socket type.
+3. **`register_socks5_openable`**: spec + establisher + pump via
+   `register_openable_with_establisher` (`timeout: None` = the 10s
+   default), scope-gated `socks5:open` (the registry runs the ACL
+   before the establisher). Missing callbacks = those commands
+   refuse RFC-faithfully (`CommandNotSupported`-class), never a
+   panic (ADR-008).
+4. **`serve_socks5_over<T>`** — the substrate-free protocol entry
+   point (the POC shape): generic over `T: AsyncRead + AsyncWrite +
+   Unpin`, driving the typestate and dispatching. This is what makes
+   the producer testable without channels (the differential-suite
+   pattern) and reusable by the batch-2 local backend.
+5. **Tests** (via the harness + the raw RFC client helper): CONNECT
+   echo round-trip through `pump_bidi`; dial refusal → 0x05 in the
+   reply, no phantom data plane; FORBIDDEN for a scope-less identity
+   (fails closed, no phantom channel); per-call opener identity
+   witnessed (CF-006); ASSOCIATE/BIND refuse `CommandNotSupported`
+   (the pre-vertical posture); half-close teardown (client EOF ends
+   the session cleanly).
+
+## Acceptance Criteria
+
+- [ ] Establisher: session-gate shape (`Establishment::default()`,
+      identity recorded; witness seam optional)
+- [ ] Pump handler: typestate-driven dispatch; `pump_bidi` inline
+      for CONNECT; no substrate types leak
+- [ ] `register_socks5_openable` wires spec + establisher + pump +
+      injected seams; missing-callback commands refuse
+      RFC-faithfully
+- [ ] `serve_socks5_over<T>` is substrate-free and directly testable
+- [ ] Integration tests: round-trip, refusal mapping (0x05),
+      FORBIDDEN fails closed, identity witness, refusal arms for
+      ASSOCIATE/BIND, half-close teardown
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/producer.md (the normative shapes; the wiring
+  requirement)
+- docs/architecture/decisions/001/004/005/006 (op, seams, auth,
+  gates)
+- POC sources: `/workspace/alksocks-connect-poc/src/producer.rs`
+  (CONNECT + establisher),
+  `/workspace/alksocks-udp-poc/src/producer.rs` (the dispatch
+  skeleton + registration)
+- alkcall ADR-049 (establishment; R-02 inline-await contract),
+  `/workspace/@alkdev/alktunnels/src/producer.rs` (the 0.8.0
+  registration template)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-seam-types.md b/tasks/implementation/core-seam-types.md
new file mode 100644
index 0000000..858ae6d
--- /dev/null
+++ b/tasks/implementation/core-seam-types.md
@@ -0,0 +1,106 @@
+---
+id: implementation/core-seam-types
+name: Core — the three seam types (DialFn / Egress / BindAccept families)
+status: pending
+depends_on: [implementation/core-crate-deps]
+scope: moderate
+risk: low
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement `src/egress.rs` per producer.md §The Three Seams and
+ADR-004: the three I/O-plane seam types as function-shaped callbacks
+with their error vocabularies. The protocol layer never names a
+substrate type — these types are the inversion point; the batch-2
+providers (local/composed) and the test harness all construct them.
+
+Deliverables (port from the POCs — the shapes are POC-validated):
+
+1. **The dial seam (CONNECT)** — from
+   `/workspace/alksocks-connect-poc/src/producer.rs` +
+   `/workspace/alksocks-udp-poc/src/producer.rs`:
+   - `TargetHandle { read, write }` — boxed halves
+     (`Box<dyn AsyncRead + Unpin + Send>`,
+     `Box<dyn AsyncWrite + Unpin + Send>`; `+ Sync` per the
+     ChannelPlan payload bound). Ownership rides the handle
+     structurally (ADR-004).
+   - `DialFn` — `Arc<dyn Fn(TargetAddr) -> BoxFuture<'static,
+     Result<TargetHandle, SocksDialError>> + Send + Sync>`.
+   - `SocksDialError` with faithful `to_reply_error()` mapping
+     (refused → `ConnectionRefused`, not-allowed →
+     `ConnectionNotAllowed`, unreachable → `NetworkUnreachable`,
+     general → `GeneralFailure` — the POC's mapping, RFC-faithful
+     per AGENTS.md convention 2).
+2. **The egress seam (ASSOCIATE)** — from
+   `/workspace/alksocks-udp-poc/src/egress.rs`:
+   - `Egress` trait: `send(addr, payload)` (payload-only — the frame
+     codec lives on the wire side) + `next_reply()` returning *inner*
+     bytes (`addr + data`).
+   - **The readiness contract is load-bearing API surface** (POC #2's
+     structural finding): `next_reply` must be `Pending` when no
+     reply is ready — an instantly-ready `Ok(None)` starves a biased
+     select into livelock. Document it on the trait; the contract
+     test lands with the producer-associate task.
+   - `EgressFactory` — called once per ASSOCIATE; the stage owns the
+     egress; stage exit drops it.
+   - `EgressError` with `to_reply_error()` mapping.
+3. **The accept seam (BIND)** — from
+   `/workspace/alksocks-bind-poc/src/accept.rs`:
+   - `BindAccept { bind_addr: SocketAddr, accept: BoxFuture<'static,
+     Result<BindTargetHandle, BindAcceptError>> }` — reply#1's
+     BND.ADDR carrier + the one-inbound-accept future. **The accept
+     future owns its resources end-to-end**; the quiet wait's
+     `select!` drop IS the client-gone teardown (ADR-004/ADR-007).
+   - `BindTargetHandle` — accepted halves + peer address; ownership
+     rides the handle structurally (the third teardown-at-birth
+     witness — no `_` bindings of owning values).
+   - `BindAcceptFactory` — called once per BIND with the requested
+     target.
+   - `BindAcceptError` (accept-stage, post-reply#1 → reply#2) and
+     `BindAcceptFactoryError` (pre-reply#1 → reply#1), each with
+     faithful `to_reply_error()` mapping.
+4. **No trait for the seams themselves** (ADR-004): DialFn /
+   EgressFactory / BindAcceptFactory stay function-shaped; `Egress`
+   is the one trait (the per-association instance is a returned
+   object, the POC shape). Note the re-evaluation threshold in the
+   module docs (three real trait impls).
+
+Tests: unit tests for the error→reply-code mapping tables (all
+variants), the `egress_factory` sync-lift convenience constructor,
+and compile-shape assertions (the handles are `Send`; the factories
+are `Send + Sync`). The readiness-contract behavioral test lands
+with the producer-associate task (it needs the stage).
+
+## Acceptance Criteria
+
+- [ ] All three seam families implemented with the POC shapes
+- [ ] Every error type's `to_reply_error()` mapping table-tested
+      (all variants, RFC-faithful codes)
+- [ ] Readiness contract documented on `Egress::next_reply`
+- [ ] Ownership-rides-the-handle shapes (no `_` bindings; handles
+      own their payloads structurally)
+- [ ] No substrate types named anywhere in the module
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+
+## References
+
+- docs/architecture/producer.md §The Three Seams (the contracts
+  table)
+- docs/architecture/decisions/004-dial-callback-shape.md (no trait;
+  ownership rides the handle)
+- POC sources: `/workspace/alksocks-connect-poc/src/producer.rs`,
+  `/workspace/alksocks-udp-poc/src/egress.rs`,
+  `/workspace/alksocks-bind-poc/src/accept.rs`
+- docs/architecture/decisions/007-bind-in-base.md (the accept
+  variants' shapes)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-test-harness.md b/tasks/implementation/core-test-harness.md
new file mode 100644
index 0000000..8a6855e
--- /dev/null
+++ b/tasks/implementation/core-test-harness.md
@@ -0,0 +1,115 @@
+---
+id: implementation/core-test-harness
+name: Core — integration test harness (duplex transport + identities + in-process seams)
+status: pending
+depends_on: [implementation/core-seam-types, implementation/core-params-errors]
+scope: moderate
+risk: medium
+impact: phase
+level: implementation
+---
+
+## Description
+
+Build `tests/harness.rs` — the producer↔consumer wiring over a
+`tokio::io::duplex` carrying the channels 8-byte chunk format — so
+every subsequent core task tests end-to-end without the `local`
+feature. This is the alkcall-0.8.0-first-consumption task: the POCs
+ran 0.7.1; the harness proves the 0.8.0 establishment surface
+(`register_openable_with_establisher`, `Establishment`,
+`ChannelClient::open_channel`) works for this crate's shape before
+the producer task builds on it. Template:
+`/workspace/@alkdev/alktunnels/tests/harness.rs` (the sibling's
+harness — same topology, adapted to this crate's seams).
+
+Topology (the alktunnels harness shape):
+
+- **producer** (connect side of the channels connection — runs
+  `ChannelClient::from_connection_with_serving`): serves the socks5
+  open op + generic channel ops on its serving registry. Its manager
+  allocates the channel IDs (`ChannelSide::Connect` — the serving
+  side allocates; the initiator adopts).
+- **consumer** (accept side): the `ChannelsAdapter` + capturing
+  install hook. The hook captures the consumer's channel-0
+  `CallConnection` and `ChannelManager`.
+- **The POC #1 wiring requirement is load-bearing here** (producer.md
+  §Harness wiring requirement): the accept-side install hook must
+  call `channel0_conn.set_identity(resolved)` before building the
+  `CallConnection` — without it, scope-gated opens fail closed with
+  `FORBIDDEN "authentication required"` (correct behavior, implicit
+  wiring). The harness makes the requirement explicit and tested.
+
+Deliverables:
+
+1. **Identities**: `consumer_identity()` (`socks5:open` only),
+   `bind_identity()` (`socks5:bind` only — the command-gate probe),
+   `both_scopes_identity()` (both grants, explicitly — never implied,
+   ADR-006), plus a `TestIdProvider` resolving them (the alktunnels
+   harness shape).
+2. **In-process seam closures** (no sockets — the protocol layer is
+   substrate-agnostic, so the harness dials in-process):
+   - `echo_dial()` — a `DialFn` returning duplex-halves wired to an
+     in-process echo server (the POC #1 shape); plus
+     `refusing_dial()` (a failing dial for the refusal-mapping
+     tests).
+   - `echo_egress()` — an `EgressFactory` whose egress echoes
+     datagrams back (the POC #2 shape, readiness-contract-clean:
+     parks on a channel `recv()`, `Pending` when nothing is ready).
+   - `echo_accept()` — a `BindAcceptFactory` whose accept resolves
+     immediately with in-process halves (the POC #5 shape); plus a
+     failing variant.
+3. **The producer registration helper**: a `spawn_producer(...)`
+   function wiring `register_socks5_openable` (or its task-6
+   successor) with the harness seams — the single place the
+   producer-side wiring lives, so the per-command tasks only write
+   tests.
+4. **A raw RFC client helper**: a hand-rolled minimal RFC 1928
+   client over any `AsyncRead + AsyncWrite` (greeting → method
+   selection → command with target → reply read) — the differential
+   suite's established pattern — so producer tasks test without
+   depending on the consumer session types (which land later).
+5. **One smoke test** proving the full loop: harness up → producer
+   registered → consumer opens a channel → raw RFC CONNECT
+   conversation → echo round-trip. (This test needs a minimal
+   CONNECT producer path; if the producer task hasn't landed, stub
+   the handler with the POC #1 `serve_socks5_over` shape inline in
+   the harness — the producer task then replaces it. Alternatively
+   sequence this task to land immediately before the producer task
+   and let the smoke test be the producer task's first test — the
+   implementer's judgment; the acceptance bar is the harness
+   compiling with the identities + seams + helpers tested.)
+
+## Acceptance Criteria
+
+- [ ] Harness topology works over alkcall 0.8.0 (the establishment
+      surface consumed successfully — the first-consumption risk
+      retired)
+- [ ] Identity wiring explicit: `set_identity` in the install hook;
+      a scope-less identity gets `FORBIDDEN` (fails closed) — tested
+- [ ] All three identities + all in-process seam closures unit-tested
+- [ ] The raw RFC client helper round-trips a greeting + command +
+      reply against a hand-rolled server over a duplex
+- [ ] No `local` feature, no sockets — everything in-process
+- [ ] `cargo test` green; clippy/fmt clean; both wasm checks clean
+      (the harness is dev-only — verify it stays out of the wasm
+      build)
+
+## References
+
+- Template: `/workspace/@alkdev/alktunnels/tests/harness.rs` (the
+  topology + identity shapes)
+- POC harnesses: `/workspace/alksocks-connect-poc/src/harness.rs`,
+  `/workspace/alksocks-udp-poc/src/harness.rs`,
+  `/workspace/alksocks-bind-poc/src/harness.rs`
+- docs/architecture/producer.md §Harness wiring requirement (the
+  `set_identity` requirement)
+- docs/architecture/decisions/006-access-control.md (the scope
+  identities)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/core-wire-codec.md b/tasks/implementation/core-wire-codec.md
new file mode 100644
index 0000000..83983de
--- /dev/null
+++ b/tasks/implementation/core-wire-codec.md
@@ -0,0 +1,110 @@
+---
+id: implementation/core-wire-codec
+name: Core — the ASSOCIATE datagram frame codec (src/wire.rs)
+status: pending
+depends_on: [implementation/core-crate-deps]
+scope: narrow
+risk: low
+impact: component
+level: implementation
+---
+
+## Description
+
+Implement `src/wire.rs` per wire.md §The Data Plane and bast.md: the
+first-party `[len: u16 BE][addr + data]` frame codec for the UDP
+ASSOCIATE datagram stage (ADR-003), re-implemented in first-party
+code shaped on alktunnels' length codec (`/workspace/@alkdev/
+alktunnels/src/wire.rs` — `frame_datagram`/`DatagramReader` — same
+F-2 behavior, but the payload here is the RFC 1928 address encoding
+plus data, not a bare payload).
+
+The codec is the crate's one framed stage (CONNECT/BIND are raw
+pass-through — no framing). Framing is applied **exactly once, by
+this layer**: `Egress::next_reply` returns *inner* bytes
+(`addr + data`); the producer's frame writer adds the single length
+prefix (the double-framing bug class POC #2 caught is an
+API-design invariant this module's docs must state).
+
+Deliverables:
+
+1. **`frame(addr: &TargetAddr, payload: &[u8]) -> Result<Bytes,
+   WireError>`** — one frame per datagram: `[len: u16 BE][addr +
+   data]` where `addr + data` is `TargetAddr::to_be_bytes()` (ATYP +
+   addr + port) followed by the payload.
+2. **The joint bound (the load-bearing constraint)**: reject
+   `Oversize` at frame time whenever `len(addr) + len(payload) >
+   65535`. The UDP payload bound alone (65507) does NOT fit u16
+   unconditionally — a 65507-byte datagram to a 255-character domain
+   is 65766 frame bytes. Max data: 65276 for domain-form
+   destinations, 65516 for IPv6, 65528 for IPv4. No unchecked u16
+   arithmetic on the payload bound (wire.md's explicit requirement).
+   Expose `addr_encoding_len(addr)` (7 v4 / 19 v6 / 259 max domain)
+   so callers can compute the joint bound.
+3. **`DatagramReader`** — the incremental decoder: feed it whatever
+   `read()` yields; it reassembles frames across chunk boundaries
+   (split frames, batched frames, partial headers). Returns
+   `(TargetAddr, Bytes)` per completed frame — it decodes the addr
+   sub-layout (ATYP determines the address length) and splits
+   `addr + data`. Infallible decode direction (the oversize bound is
+   enforced at frame time); `is_mid_frame()` diagnostic for teardown.
+4. **The F-2 invariant, correctly stated**: a **zero-length payload**
+   is a legal empty datagram — never EOF. (`len == 0` is unreachable
+   on this wire since `addr` is always ≥ 7 bytes; the invariant that
+   matters is payload-level. See the doc-sync deliverable.)
+5. **Doc-sync (wire.md + bast.md residue)**: both docs say "`len = 0`
+   is a legal empty datagram" while also pinning that `len` covers
+   `addr + data` (joint bound) — `len = 0` is unreachable since
+   `addr` ≥ 7 bytes. Correct both docs to state the operative
+   invariant at payload level ("a zero-length payload is a legal
+   empty datagram, never EOF — the F-2 invariant") in the same
+   change; do not silently diverge.
+6. **EOF layering note in the module docs**: EOF is the
+   channels-level sentinel on the `BiStream`, outside this codec;
+   the codec never emits zero-length reads, so the layers never
+   collide (the alktunnels wire.rs module-doc posture).
+
+Tests: round-trip (v4/v6/domain), empty payload legal + not EOF,
+split across awkward chunks, batched frames in one chunk, partial
+header waits, joint-bound rejection (the 65507 + 255-char-domain
+case), oversize at frame time, mid-frame state observable, garbage
+ATYP decode error (the `udp_undecodable_frame_ends_association`
+shape — the decode error type must distinguish a bad ATYP so the
+producer stage can fail the association loud).
+
+## Acceptance Criteria
+
+- [ ] `frame`/`DatagramReader`/`addr_encoding_len` implemented per
+      the joint bound; no unchecked u16 arithmetic
+- [ ] Zero-length payload round-trips as a legal empty datagram
+- [ ] Split/batch/partial-header reassembly tests pass
+- [ ] Joint-bound rejection test (65507-byte payload + 255-char
+      domain → `Oversize`) passes
+- [ ] Garbage-ATYP decode surfaces a typed error (not a panic)
+- [ ] wire.md + bast.md residue corrected (payload-level F-2
+      statement) in the same change
+- [ ] `cargo test` green; clippy `--all-targets -- -D warnings`
+      clean; `cargo fmt --check` clean
+- [ ] `cargo check --target wasm32-unknown-unknown` +
+      `cargo clippy --target wasm32-unknown-unknown -- -D warnings`
+      clean
+
+## References
+
+- docs/architecture/wire.md §The Data Plane (the frame + the joint
+  bound arithmetic)
+- docs/architecture/bast.md (the BAST contract for the frame)
+- docs/architecture/decisions/003-udp-handle-swap.md (framing-once,
+  F-2)
+- Template: `/workspace/@alkdev/alktunnels/src/wire.rs` (the length
+  codec + its test matrix — shape on it, do not depend on it)
+- POC sketch: `/workspace/alksocks-udp-poc/src/datagram.rs`
+  (`frame`/`inner`/`unframe`/`addr_encoding_len`)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file
diff --git a/tasks/implementation/review-protocol-core.md b/tasks/implementation/review-protocol-core.md
new file mode 100644
index 0000000..6ded7d2
--- /dev/null
+++ b/tasks/implementation/review-protocol-core.md
@@ -0,0 +1,68 @@
+---
+id: review-protocol-core
+name: Review — the protocol core before any assembly-layer work
+status: pending
+depends_on: [implementation/core-integration-suite]
+scope: narrow
+risk: low
+impact: project
+level: review
+---
+
+## Description
+
+Review checkpoint before batch 2 (the assembly layer: `local`
+providers, front doors, `composed` providers). Everything in batch 2
+consumes the seams and the session types, so a core defect — a seam
+contract drift, a teardown hole, a wasm regression — would propagate
+into every downstream task. This review is the gate between the core
+group and the assembly layer, and the point where batch 2 gets
+decomposed (its task shapes may adjust to what the core actually
+landed).
+
+Review against the architecture, in this order:
+
+1. **Seam contracts** (ADR-004, producer.md §The Three Seams): the
+   readiness contract (`next_reply` `Pending` when nothing is ready —
+   the contract test exists and pins it), ownership-rides-the-handle
+   (no `_` bindings of owning values; handles own their payloads
+   structurally), the accept future owning its resources end-to-end
+   (the quiet-wait drop IS the client-gone teardown), framing-once
+   (the egress speaks inner bytes; the wire layer frames once).
+2. **Wire fidelity** (wire.md, ADR-003/ADR-007): the datagram frame
+   (joint bound, F-2 at payload level), the phase model (no mode
+   byte; post-reply garbage refused structurally in all three
+   commands), the sentinel surfaces (ASSOCIATE's `127.0.0.1:0`,
+   BIND's variant-3 zeros), association lifetime = channel lifetime.
+3. **Access control** (ADR-006): the open gate (`socks5:open` at the
+   registry), the command gate (`socks5:bind` at BIND dispatch),
+   two grants never implied; refusals RFC-faithful everywhere
+   (convention 2 — never collapsed into generic errors).
+4. **Teardown/lifecycle** (wire.md §Sentinels, alkcall ADR-050):
+   `pump_bidi` awaited inline (R-02 — no spawn-and-forget); the
+   teardown matrix tests actually pin the semantics; no
+   teardown-at-birth paths (the third-witness pattern held).
+5. **Convention compliance** (AGENTS.md): no `unwrap`/`expect`/
+   panics outside tests; `thiserror` surfaces; no comments in code
+   except doc comments + pinned non-obvious-correctness notes;
+   naming; module structure (one module per file, `lib.rs`
+   re-exports); producer/consumer vocabulary (no server/client
+   framing outside RFC vocabulary).
+6. **Wasm-clean invariant** (ADR-009): both wasm checks green; no
+   `net`-shaped feature; tokio on the wasm-clean subset; the harness
+   and tests stay out of the wasm build.
+7. **POC-fidelity spot-check**: for two or three of the trickiest
+   behaviors (the readiness contract, the quiet-wait two-exit
+   semantics, the split-datagram reader persistence), diff the
+   crate's behavior against the POC crates' findings — the POCs are
+   the empirical record; drift is a finding.
+8. **Docs sync**: the specs match what landed (the integration
+   task's reconciliation verified; any residue is a finding).
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion
\ No newline at end of file