The eighth track of Phase D. Six chapters about what the rooms stand on: the contracts and the registry behind every stored row, tenancy and persistence, events and jobs, Sophia's answers, Iris's memory, and the live turn Psyche and Lilith share. The plan is the master TODOS document, Phase D, "Track · Shared systems". Each chapter has its own brief, listed under Sources, and this brief is assembled from those six. The facts below are the chapter briefs' statements. The choice of audience, and the paragraphs marked as this brief's reading, are this brief's own.
Audience#
Somebody who knows a room and now has to say what is underneath it: an engineer joining a shared library, or a reviewer who has to judge a sentence such as "the write is idempotent" or "the answer is grounded". Chapter one opens by saying that every chapter before it was about a room, and that this one is about what the rooms stand on. The track answers which layer a claim belongs to, and how strong the evidence for it is.
Running example#
Each chapter follows one small thing all the way through, and the titles say which:
- one stored row, a Veritas timeline row under the tenant
tenant-a; - one saved preference, from the button press to the row;
- one finished practice session, and everyone who has to hear about it;
- one question asked of Sophia: what supports this claim?
- one sentence a member says in passing: "I prefer ten-minute practices";
- one live avatar session that degrades and then meets a safety boundary.
The examples hand over to each other. Chapter two follows a write to the kind of row chapter one took apart, chapter three starts where that row is written, and chapter six ends where memory begins, because inside a crisis frame Iris records nothing.
Chapter order#
| # | Chapter | Promises |
|---|---|---|
| 1 | One row, and everyone allowed to touch it | What makes a row owned, what the registry produces for it, and what delete means |
| 2 | One saved preference, and seven things to prove | What shape a write must have, who was allowed to send it, where it landed and what remains |
| 3 | One finished session, and everyone who wants to know | How a fact is carried, what delivery proves, and how it is found again |
| 4 | One question, and what the answer may claim | How an answer is composed, how a source is judged, and what is actually reachable |
| 5 | One sentence, and what decides whether it returns | What is kept, what comes back, and what the member controls |
| 6 | What happens in the moment, and what it costs | Who decides a live turn, what degrades first, and what safety costs |
Chapter three merges two inherited guides, communication-discovery and
v1-events-jobs-discovery, under the new id events-jobs-discovery. The other
five are under the ids architecture, v1-contracts-tenancy, sophia, iris
and psyche-lilith.
Why this track exists#
Chapter one says how this track differs from the room chapters: "the answer here is mostly that the substrate is real and wired". What each chapter then finds is narrower than that sentence:
- Chapter one finds the substrate mostly real and wired, with its defects in one place: agreeing which record an identifier names.
- Chapter two finds three mechanisms that are complete, correct and barely reached.
- Chapter three separates an event from a job, because only one of the two promises the work will finish.
- Chapter four separates Sophia into what is live, what is built and imported by nothing, and what is only specified.
- Chapter five treats memory as governance: consent stores a sentence, and a second check decides whether this is the conversation to mention it in.
- Chapter six finds the contract layer of live interaction real, and the transport and the providers not built.
This brief's reading: the questions the chapters close on are the track's method. Ask what enforces it, ask who turns it on, ask whether it ran.
What this track does not do#
Each chapter brief states its own limit, and they differ:
- Chapters one and two touch no deployed service, real member account or live database. Chapter one's contract repository ran against a declared delegate double, not PostgreSQL.
- Chapter three touches no Redis, deployed service, real member account or live queue; the event bus, the search route and the worker were read rather than run.
- Chapter four called no model, reached no deployed service and used no member data; the routes, the adapter and the verification loops were read at the pin.
- Chapter five used no member data, reached no deployed service and wrote no durable store; the store, the consent ledger and the inspection machine were read at the pin.
- Chapter six used no live transport, called no speech or avatar provider and ran no session.
In every chapter the modules that were executed ran in isolation against declared inputs. Chapters two and three also took counts across the repository at the pin. Chapter two says no claim is made that a request from a real client behaves this way in a deployment, and chapter three says the same of a published event reaching a subscriber.
Sources#
The six chapter briefs in Eve, at the revisions read for this brief. All six are drafts: none has been through independent review, and this brief repeats their statements without re-running their probes.
| # | Chapter brief (Eve document) | Revision | State |
|---|---|---|---|
| 1 | 84fe678e-53ed-86e4-b389-d79b7386ddb4 |
1 | draft |
| 2 | a9b7393e-4c9f-88c8-ab5a-0aad0141bc56 |
1 | draft |
| 3 | 7586571c-a07c-8632-86d5-63a85242a99f |
1 | draft |
| 4 | 013778e3-ce30-830a-978c-38f574c3456e |
1 | draft |
| 5 | 7a483f63-094d-8512-a075-ba843f2ba875 |
1 | draft |
| 6 | ed86abb4-b638-853a-8941-603c46176c28 |
1 | draft |