Applied, to the
BCS architecture.

One Valkey — 9.1.0, listening on :6390 in the committed record — carries two surfaces, and the two surfaces ground the two halves of the catalog.

The EchoMQ bus. EchoMQ owns its protocol: the braced emq:{q}: keyspace, jobs as entities, and a Lua state machine. The committed key map reads emq:{orders}:pending | emq:{orders}:job:ORD0NgWEfAEJfs | {emq}:version | {emq}:locks — 17 bytes of grammar before a fourteen-byte branded payload. A job’s row is a hash of three fields; four sorted sets per queue carry the lifecycle — pending score-zero, active scored by lease deadline, schedule scored by run-at, dead score-zero — and every transition is a single atomic script. The coordination, queue, time, streams, and flow families (R2–R6) and the operations chapter (R8) ground here.

The EchoStore near-cache. EchoStore is “branded keys, local speed, bus-driven coherence”: an L1 of declared ETS tables in front of the shared L2 Valkey. The committed figure: 1311621 hit reads per second (762 ns each) against 31 us per L2 GET on the same wire — the L1 hit is 40 times cheaper than the round trip it replaces. The caching family (R1) and the R7 read-models ground here.

The split is the catalog’s map. Pick a surface to read which chapters it carries and the committed figures that ground it.

Before a single pattern runs, the keyspace states which protocol it speaks. The version fence lives at {emq}:version — the one key that is about the deployment rather than about any queue — and the connector reads it on every connect, first boot and every reconnect, refusing to serve against any other value. The committed read is self-referential: it travels through a connection that could not exist had the fence not held.

Two orientation modules set up the grounding the pattern chapters reuse.

The eight pattern chapters, sequenced along the EchoMQ build. Each opens with its use cases and closes with a workshop that builds one slice of the BCS Redis tier.