Admission as a typed answer at the door.

LMAX's published architecture centers a Business Logic Processor that runs entirely in memory by event sourcing and handles "6 million orders per second on a single thread" (Fowler), surrounded by Disruptors — pre-allocated ring buffers where producers claim slots, a single writer applies in batches, and consumers chase sequence numbers without locks. The team's reported conclusion is the part that generalizes: contended queues are at odds with how modern CPUs run, and the cure is the single-writer principle plus mechanical sympathy — one owner per mutable thing, batch when behind, make waiting policy explicit.

The 6-million figure is external, and it is quoted with its attribution. It is not a number the trading engine claims; the Exchange.* suite on top of the Ring has run no rung yet. The Ring itself is real and measured — its numbers are on the B4.3 record. The engine takes the principles, not the count.

The BEAM cannot and should not imitate the Java mechanics — and does not need to, because the principles transfer without them. Select an element to read why it transfers whole or why the BEAM deliberately drops it.

The seat is real source, and it matches the pattern part for part. The moduledoc names the correspondence in the house's own words (source: echo/apps/echo_cache/lib/echo_cache/ring.ex): "This is the Disruptor's shape translated to the BEAM — sequences, preallocated slots, batched consumption — standing beside the bus's park-don't-poll: both replace discovery with arrival," and "the single-producer requirement is structural, not advisory." The committed B4.3 record carries the shape in numbers — the bound, the gauge, the batch-shaped drain — every figure verbatim from the ring's own rung:

G4 full ok -- the bound held its shape: 64 accepted, 136 refused with :dropped and counted -- never blocked, never overwritten -- then the release drained all 64 and publish 201 landed and applied: a storm bends the lane's at-most-once contract no further than the contract already bends
G3 occupancy ok -- mid-storm the gauge read 600 of 4096 and drained to exactly 0; priced, the ring moved 100000 items in 99 ms -- 1005116 items per second end to end on one scheduler, inside the derived band, largest batch 200, nothing dropped
G1 surface ok -- the ring's surface is whole -- publish, occupancy, stats, stop, a generic one-batch apply function -- ... a fresh ring stands at occupancy 0
PASS 6/6

The bound is real: 64 accepted, 136 refused with :dropped and counted, then the release drains all 64 and the next publish lands. The gauge is observable: mid-storm the gauge read 600 of 4096 and drained to exactly 0. The drain is single-consumer and batch-shaped, with the largest batch a reported stat: largest batch 200, nothing dropped. The ring under storm refuses, recovers, and keeps serving — exactly the admission contract a match path needs. The Exchange.Book that drains it is PROPOSED: the single writer, woken to drain whole batches, shelling out to a pure Decider — the trading engine on top of a real, hardened substrate, taught as design because it has run no rung, and so it claims no number; the Ring beneath it is shipped and measured.

The one twist this house adds to the pattern: the chase sequence is not a slot counter beside the data but the branded Snowflake inside it. Mint order is the sequence, so a consumer's resume cursor is an id — EchoData.BrandedTree.page_after/4 reads strictly after it in creation order — the same order holds in every store by the byte-sort theorem, and there is no second numbering scheme to reconcile with the first. A counter beside the data is a second source of truth; the id is the first one, already paid for at mint.

The alternatives are named honestly. The unbounded GenServer mailbox — the BEAM default and the proper baseline: it preserves order and the single writer for free; what it lacks is an admission answer, so overload arrives as mailbox depth, which arrives as latency. The Ring's :dropped is the same architecture with the truth told at the door. GenStage / Broadway — demand-driven pull is the right shape where the consumer sets the pace (draining work), and the wrong shape at market ingress, where producers are external and the answer must be immediate accept-or-shed. A stream as the ingress (XADD first, decide later) — durable and replayable, but one wire round trip before every decision, and it logs commands, not facts, re-opening the command-sourcing coupling the next dive names.

Where the seat is the wrong tool, stated: low-rate aggregates — an admin book, a reference-data writer — gain nothing from a Ring over a plain call; fan-out is never the Ring's job (one drainer, by law); and durability is never the Ring's job either — the Ring may drop by design, the log may not lose by design, and the two truths are different and both stated. The pure core that reads what an accepted command means is B8.1.2 · The Decider.