docs(design): 054 zero-alloc connector boundary, 055 wildcard inbound links, 056 aimdb join CLI

aimdb-bench/Cargo.toml

@@ -62,6 +62,11 @@ harness = false
 name = "b1_b2_remote_json"
 harness = false
 
+# Connector-boundary allocations per message (design 054 baseline).
+[[bench]]
+name = "b0_alloc_connector"
+harness = false
+
 [features]
 default = ["std"]
 # Gates `profiles`/`reports`/`harness` and their criterion/serde_json/

aimdb-bench/README.md

@@ -26,6 +26,10 @@ Plus two informational benches that exercise the full runner-driven pipeline.
   `b1_b2_remote_json` (host). These compare issue #196's direct JSON bytes with
   the compatibility `serde_json::Value` tree through the real typed record,
   buffer, `Payload` and AimX envelope. Socket I/O and scheduling are excluded.
+- **connector boundary** — `b0_alloc_connector` (host). Baseline for design
+  054: `Router::route`, `pump_source` with a minimal `Source`, and the
+  per-message `recv_into` + `Connector::publish` calls of `pump_sink`, on a
+  no-op connector.
 - **Embassy** — `b0_alloc_embassy`, `b1_b2_embassy`
   (host). These drive the real [`EmbassyBuffer`] backend via
   `futures::executor::block_on` over embassy-sync's poll methods — no
@@ -65,6 +69,9 @@ cargo bench -p aimdb-bench --bench b0_alloc_linkable
 # B0 — direct-vs-tree JSON allocations through the AimX envelope boundary
 cargo bench -p aimdb-bench --bench b0_alloc_remote_json
 
+# B0 — allocations at the connector boundary (design 054 baseline)
+cargo bench -p aimdb-bench --bench b0_alloc_connector
+
 # B1 + B2 — latency (time/iter) and throughput (msgs/sec), one Criterion suite
 cargo bench -p aimdb-bench --bench b1_b2_tokio
 
@@ -124,6 +131,8 @@ The required result is **0 allocation calls and 0 allocated bytes**. It isolates
 
 `b0_alloc_remote_json` warms the production in-memory `record.get` and subscription-event paths, then compares 5000 tree/direct operations. Its gate is relative: direct JSON must reduce both allocation calls and allocated bytes. It does not require zero allocations because the owned JSON `Vec`, `Arc<[u8]>` payload and AimX envelope serialization still own storage.
 
+`b0_alloc_connector` measures what the connector interfaces cost per message, with no transport. It asserts today's values exactly (0 for `route`, 2 for `pump_source`, 2–3 outbound), so a regression *or* an improvement fails it until `EXPECTED` in the bench and `data/baselines/b0_alloc_connector.json` are updated together. The inbound `pump_source` row is the difference of two runs, so pump setup cancels out. See design 054 for where each allocation comes from.
+
 > **Embassy eager registration (design 039 F8/F9).** An Embassy `SpmcRing` reader registers its embassy `Subscriber` eagerly, at `subscribe()` time — matching Tokio's `broadcast` — so no separate priming step is needed before the first `push`.
 
 **Noise reduction:** a `new_current_thread()` Tokio executor is used so there are no work-stealing threads and Tokio's scheduler does not allocate per-poll in the hot path.