The loop closes.

Module 12 left a deliberate tension: a stream is an append-only log, XACK does not delete, so retention must trim — but trimming must never outrun what has been saved (the fold-before-trim rule of fork F3.4-A). This dive is where “saved” gets its meaning. The emq3.5 archive is a dedicated fold consumer that reads the stream’s old tail — everything below the watermark W — and commits those slices into the engine with EchoStore.Graft.commit/3. The engine streams those pages from CubDB to Tigris exactly as any other commit would, so a slice of the bus’s history becomes a set of the store’s durable segments. Only then may the trim advance, which is why the fold watermark always sits at or ahead of the trim watermark. The payoff is the merge-read of Module 12.3: a deep read concatenates the segments below W (now in the store) with the live tail at or above W (still on the bus), with no gap and no overlap, because the same watermark divides them. The fold is not a backup job bolted on; it is the return leg that turns the bus’s retention into the store’s durability, using the one commit primitive the whole course has been building.

Stand back and the four chapters are four tiers of one cycle. A write is accepted by Champ (Chapter I) with a bounded loss window; it is held in the local CubDB store (Chapter II) with copy-on-write versioning; it is committed by Graft (Chapter III) as a single replicated LSN behind the fence; and it is published onto an EchoMQ stream (Chapter IV) for consumers and replicas. Then the archive fold carries the stream’s aged tail back into the engine, and the cycle is closed. Two things bind the ends, and both are earlier dives in this very module: the commit LSN is the cursor every tier names its position with (12.1), and the change feed off that LSN is how a commit becomes a bus event in the first place. The outbox (12.2) sits to one side of this cycle, keeping the enqueue hot path off durable storage while still surviving a crash. So “the bus and the store are one system” is not a slogan: it is a literal loop, joined at a shared integer and a shared feed, where the engine’s commit drives the bus and the bus’s retention drives the engine’s commit. That is where Echo Persistence has been heading all along — and the door past it, in the chapters to come, is where this durable substrate stops being plumbing and starts being the Branded Component System.