# Chapter 3 · One finished session, and everyone who wants to know · brief · 12 September 2026

The third chapter of the _Shared systems_ track, rewritten in the Eve edition on
12 September 2026. It merges two inherited guides — the ten-slide
`communication-discovery` overview and the thirty-four-slide
`v1-events-jobs-discovery` field guide — into one forty-four-slide chapter under
a new `events-jobs-discovery` id. It follows the plan in
`EVE-V1-REDESIGN-TODOS.md`; the receipt is under `verification/`.

## The job

The previous chapter followed one write to the row it became. This one starts
where that row is written and asks who else needs to know. A member finishes a
practice session. The response confirms it, an event announces it, a job does
the expensive follow-up, and tomorrow the member searches for it and is offered
a next step. Five promises, five different acknowledgements.

The chapter's job is to keep those apart. The word _asynchronous_ covers an
event and a job without distinguishing them, and only one of the two promises
that the work will finish.

**Running example.** One member finishes one practice session. Executed on this
host against the pinned modules: an event envelope the contract will parse, a
topic registry of seventy-six topics, five job priorities that round-trip
through BullMQ, a lexical ranker that never drops a candidate, and a
recommendation scorer whose five weights sum to one.

## Three learning promises (the cover)

1. **How the fact is carried** — four transports, and the promise each makes.
2. **What delivery actually proves** — publishing, claiming, retrying, signing
   it out.
3. **How it is found again** — three candidate pools, one ranker, six adapters.

## Three things to remember (the close)

1. **Ask which transport.** Only one of the four promises the work finished.
2. **Ask which mark records it.** No field anywhere means everybody finished.
3. **Ask whether it is exercised.** Seventy-six topics; zero using the evolution
   machinery.

**Bridge.** The next chapter turns to Sophia, where answers are assembled from
sources, and the question becomes what a composed answer may claim about the
passages underneath it.

## Sections

Six sections of five, five, six, six, seven and seven, plus the cover and the
close. All six fall inside the plan's five-to-seven range.

| Section                  | Slides                                                                                   |
| ------------------------ | ---------------------------------------------------------------------------------------- |
| Choosing how to carry it | the case, three obligations, four transports, priority, an integration                   |
| What an event is made of | ten fields, two envelopes, the registry, a census, the Redis primitives                  |
| Getting it delivered     | publish, subscribe, five continuations, groups, recovery, outbound                       |
| Work that has to finish  | the job envelope, priorities, statuses, the lease, dead letters, queue health            |
| Finding it again         | the route, three pools, the ranker, order and cache, the envelope, curation, a diagnosis |
| Offering the next thing  | six adapters, nine reasons, five weights, feedback, the gate, the boundary, the review   |

## Treatment decisions

- **No capture.** The chapter's subject is a fan-out and a request path, not a
  surface. Every claim is executed or counted at the pin instead, and the
  receipt lists what was run.
- **Two guides merge into one chapter.** The ten `communication-discovery`
  slides were a shallower pass over the same material. Seven of them belong to
  the reviewed foundation and must keep their identifiers, so the rewrite
  carries the foundation id and the V1 slide that duplicated it is the one that
  merges away. The remaining three are absorbed by neighbours.
- **The old manuscript is retired.** `content/70-v1-events-jobs-discovery.json`
  held those thirty-four slides and nothing else, so it empties and is removed.
- **Ten slides are new**: a cover, six section dividers, a close, and two
  teaching slides — `ejd-obligations`, which separates the three receipts a
  fan-out writes, and `ejd-topic-census`, which counts the registry machinery
  nothing uses.
- **One citation object per path.** Authoring a path twice, once per heading,
  reads the same on the slide but hides the second set from anything that looks
  the path up — including the assignment review. The assembler now folds them.
- **No table.** Four slides the plan would have allowed a table for read better
  as card grids and compare panels; the chapter ships without one.
- **All four stat panels are declared `measured`** — the new `ejd-topic-census`
  and three slides that became stat panels in the rewrite — because every number
  on them was produced by executing a pinned module on this host or counting at
  the pin.

## What this chapter does not establish

Nothing here touches Redis, a deployed service, a real member account or a live
queue. Every module was executed in isolation against declared inputs, every
count was taken across the repository at the pin, and the event bus, the search
route and the worker were read rather than run. No claim is made that a
published event reaches a subscriber in a deployment.
