Two corrections, both from the operator and both better than what was written.
An SDK is an implementation of the mesh's module protocol in one language, and
nothing more. The first draft defined it by the test it passes, which describes
how you check one rather than what one is — and leaves it sounding like a library
that helpers could accumulate in.
And the protocol is split per capability, which was missing entirely. A module
that only consumes events uses the event capability; one that serves tools uses
the tool capability; a provider uses provisioning. Nothing about consuming an
event requires knowing how a grant is answered, so an SDK need not implement all
of it to be real.
That has a precedent here: a host declares which resource kinds it can apply, and
a partial host is a real thing rather than a broken one (ADR 0005). An SDK
implementing the floor and events is exactly as legitimate, and a module written
against it is a module that does events.
Which changes what adding a language costs. A Rust SDK doing connection and
events is useful the day it exists, with tools and provisioning following when
something needs them — rather than a language being unsupported until it is
entirely supported, which is what makes adding one a project instead of a
contribution.
Conformance is therefore per capability too: a monolithic pass or fail would make
a partial implementation indistinguishable from a broken one, which is the
distinction the whole thing rests on.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
ADR 0039 settles what belongs in an SDK. It does not say what happens when there
is more than one, and there already is: the contracts are expressed as Go types
in the control plane and host and as TypeScript types in the SDK, and nobody has
felt it because both live in one repository.
They already disagree. The provision's field is "resource" in one and "Provision"
in the other; "consumer" means the module in one and the node in the other; the
envelope declares six headers on one side and emits four on both — the missing
two being x-causation-id and x-schema, the second of which is exactly what a body
needs in order to change shape without silent misreads.
That class of failure does not announce itself. Two implementations disagreeing
about an envelope do not fail to compile — they ignore each other's messages, and
a mesh where a module stops reacting looks like a mesh where nothing happened.
So the decision is to specify the wire rather than share the types, because the
shapes are the easy half. What two implementations actually disagree about is
behaviour: queue naming and durability, which headers are required and what an
unknown one means, taking identity from the sealed credential rather than the
environment, dedup on an id only the emitter can make, pinning a fingerprint
rather than trusting an authority.
And the suite is executable rather than prose, because a specification nobody can
run is a document two implementations drift from while both believe they conform.
The two existing implementations are the first made to pass it — a suite only new
SDKs must satisfy would certify every future language against a disagreement that
is already here.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
Written an hour ago claiming a produced image must be published before anything
can fetch it, so the registry had to precede the control plane. The premise is
false: the machine that builds the image is the machine that runs it, and the
temporary control plane names a built image exactly as it names a carried one —
by the digest of its own configuration, which requires nothing to have served
it. Building changes where the bytes came from, not where they are.
Rewritten rather than superseded because nothing has been built on it and
nobody has read it: a record that contradicts itself is a draft, not a decision.
The argument is kept, because it was asked for and a negative answer is the
result.
The two questions the design record named as the one gap stopping a fresh mesh
from producing anything. They cannot be answered apart: a builder with nowhere
to publish has made a file on a disk.
The registry's role did not change — the answer to 'must it precede the control
plane' did, because the control plane's image is now produced rather than
carried, and a produced image must be put somewhere before it can be fetched.
Corrects ADR 0070, written an hour earlier, which had the control plane consuming
the catalogue in order to compose a declaration. That was written before the two
graphs had been told apart and creates a dependency that need not exist: a
catalogue that is down would leave the control plane unable to compose the thing
that would repair it.
The catalogue links module-versions to each other and does not know nodes exist.
The control plane links module-versions to nodes, and holds capabilities and
claims. They meet only when something is installed, and everything between them
travels as events over the broker, on durable queues, so nothing is lost when a
receiver is away.
Build order is not computed anywhere. The builder never consults the graph and
builds what it is asked for; the catalogue asks for the next build after the
previous registration, so ordering holds by construction. Its rule is a condition
rather than a schedule — rebuild once everything a module was built against is
current — which covers a chain and a diamond alike.
Left open: whether an upgrade is applied or merely noticed, and whether a module
on several machines upgrades on all at once.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
ADR 0070 has the init builder clone the source and does not say from where, and
ADR 0067 had rejected building at genesis partly because the forge runs on the
mesh being rebuilt. That objection binds only when those are the same mesh, which
is true exactly once.
So genesis clones from a mesh by name, and if that mesh is lost the name moves to
another that holds a copy — recovery is a name pointing elsewhere rather than a
backup being restored, and every installation adds somewhere it could point.
It names a commit and checks what it got, because the forge a mesh installs from
is the trust anchor for everything that mesh will run. On 2026-09-11 that forge
was running a cryptominer and tampering with git operations in flight; nothing was
altered, but a mesh installing during those hours could not have known that.
What relationship a mesh keeps afterwards is left open and named, so that whoever
writes the init builder does not settle it by accident: a snapshot and then
independence, or a continuing upstream for core modules.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
The graph had no owner: what modules are, what they require, what they claim and
what they are built against all sat in the control plane because that is where it
was written first. The control plane's own test says otherwise — anything a single
machine could answer alone is not its work, and what a module needs requires no
knowledge of any node.
So the catalogue becomes a core module beside the control plane and the builder,
owning the graph and serving tools over it. The control plane consumes it, which
is the opposite of what the tiers suggest and is therefore stated rather than
inferred.
That makes the catalogue a fourth thing that cannot arrive through the ordinary
path, so the claim written this morning that the list was closed at three is
corrected. All four are answered by one mechanism instead: the installer carries
an init builder and the core modules are built on the machine, so what is carried
is a builder rather than a result and nothing is left without a route.
Two things are open and named rather than assumed: where the init builder clones
from, given the forge normally runs on the mesh it would be rebuilding, and what
it publishes into, given the registry is installed later in the order today.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
Raising a scenario occupies the machine and the person who started it, and
running in the background against a working copy is worse than waiting: a run
reads that copy as it goes, so editing while it runs yields a result describing a
state that never existed.
Proposed rather than accepted. The load-bearing part is the restriction — a
request names a bed and a commit and nothing else — because a request that could
say what to install and where would make the lab a second way of installing a
mesh, which is the arrangement that just cost a year of late-found faults.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
The design said a module's manifest sits at a repository's root, full stop, which
means one repository per module. Nothing that exists is shaped that way: the
catalogue holds sixty-seven modules one to a directory, no code repository has a
manifest at its root, and the system being replaced has always built a module
from a repository and a path.
So the builder could be asked to build nothing that exists — pointed at the
catalogue it finds no manifest, pointed at a module's source it finds none
either. Recorded as a decision because it moves the core modules' manifests
beside their source, and corrects the design that said otherwise.
Also corrects, in the same document, how the three things the build loop cannot
produce actually arrive. They were written as though all three were carried in.
Only the control plane is: the registry is pulled from the public internet, and
the builder has no route at all — which is now stated as the open one rather than
implied to be solved.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
Both carried 'model access', which is not one of the six the reading order
knows, so neither had a place in it. 0050 — the record they extend, on the same
subject — is 'what runs on it'.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
0066 was proven on a four-node bed before it was ratified: one node setting
moved an entire domain, a routed name resolved inside the mesh and was issued a
certificate by the internal authority, and TLS verified against that authority
with no override. 08-connectivity rests on it and could not while it was
proposed.
0067 records what deleting the lab's registry exposed — that pinning quietly
required a registry before the thing that lets a mesh have a registry could
start — and the pivot that resolves it.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
Both carried a topic outside the six the index knows, so neither had a place to
be read in — 0067 had none at all. Both are 'the tiers', beside 0036 (bootstrap
ends at a usable mesh) and 0007 (connectivity), which is what they extend.
0067 cited 0041 for tier 0's property; on this trunk 0041 is events, and the
record it meant is 0005. A citation that resolves to the wrong record reads as
corroboration, which is worse than a dead link.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
029 says a module providing the artifact store may not build artifacts. The
pivot shows that is the narrow case: it may not require anything the store is
needed to deliver. A route-label migration gave the registry a public name and
a route requirement, and at genesis nothing provides a route — nor can anything,
since the routing stack needs images and images need the store.
The same cycle through a door the existing wording did not cover, so the rule is
widened where the bootstrap decision states it, with a check that would catch the
next one where it is written.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
The control plane's image is built from source and pushed nowhere, so it has no
manifest digest; a registry assigns those. Pinning therefore required a registry
before the thing that lets a mesh have a registry could start — a dependency the
rule created by accident. The lab hid it by raising a disposable registry no
production has.
Proposed, not accepted.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
0056 is 'the authority is the control plane, not a database', drafted on the
in-progress record chain this branch was cut from before those three records
landed. Two files would have collided at merge, which is the kind of thing that
is cheap now and confusing later. The code written against it still says 0056
and is corrected separately.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
A route contribution carries a label; the node carries its public domain; the
mesh composes <label>.<public-domain> and holds no name map. A granted route is
published into internal resolution so anything in-mesh (notably an internal ACME
authority) can resolve and reach it. That authority certifies routed names by the
same path a public one would, differing only in issuer and trusted root.
Records the decision as proposed and amends connectivity SS2/SS3/SS5 plus its Open
list with the lab findings behind it.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
Extends ADR 0050: model-access, a provision, may be answered by a NODE that
hosts a model (Ollama/vLLM serving an OpenAI-compatible endpoint) as well as by
a licence record. A node-answer delivers an endpoint (base URL + model, and a
key only if the server wants one), rides the ordinary serves/binds path, and
uses no adapter; the resolver already prefers a local model over a licence. One
provision, two answers — the mesh's own model sits behind the same interface as
a vendor's. Names the mesh-scope-alongside-licence limitation and its fix.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
Closes the usage-tracking half ADR 0050 left open. One row shape (0050's), recorded at two
consumer grains: the holding module (licence-level, e.g. Anthropic utilization%) and the agent
session (per-session tokens/cost, since a session IS a consumer per ADR 0026). Produced by the
vendor adapter; read on a schedule (0053); recorded as events the audit-logger keeps (0041/0042)
plus a queryable usage context store (0008); usage is not a credential and is recorded in the
clear. Reuses the session, schedule, event, and store the mesh already has. Accepted per the
user's choice to build the full feature incl. per-session token/cost.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
The recurring twin of run-once (0052): a schedule modifier on the container shape, reusing
its security bound (no new host action/shape, strictly less than an action) and reversing its
gating rule — a scheduled step runs after convergence, does not gate the apply, and a failed
run is logged, not fatal. Unblocks kometa's sync and pollers. Accepted per direction to build
the primitive now.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
The run-once lifecycle primitive: run-once:true on the existing container shape,
gated by declaration order + exit 0, idempotent by digest. No new host shape, no
arbitrary command — strictly less powerful than an action. Resolves issue 037.
Verified sound and faithful (ADR 0005/0047); status proposed->accepted.
A module can declare state but not a step that runs. mosquitto must seed its
dynsec admin into dynamic-security.json before the broker starts, or the plugin
aborts; the database providers need the same for first-boot migrations and
health gates (04-ISSUES/037). The old event-hook engine that did this was
powerful and flaky; this is the narrowest sound mechanism instead.
A run-once step is an ordinary container marked `run-once: true`: the host runs
it to completion, requires exit 0, and gates the apply on it — so what the
declaration places after it (the broker) starts only once it has finished.
Gating is by declaration order, not a resolved dependency (ADR 0005); the
completion marker is the recorded declaration digest (ADR 0018), so a re-apply
does not re-run it unless the declaration changed. No new host shape and no
arbitrary host command: strictly less powerful than an `action`.
Points 04-ISSUES/037 fixed-by/amended-design at the record; index regenerated;
records.py and index.py pass.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
Your decision, ratified: the media library (and shared/pre-existing data) is
operator-owned and external; a module declares access, not ownership; the host
mounts but owns nothing (no create/chown/reconcile/remove); an absent accessed
path is refused clearly; several accessors co-resolve. Status proposed->accepted;
index regenerated (records + index checks pass). Implementation lives on the code
branches (mesh-control/catalog/host), held for merge after the convergence fix.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
Resolves 04-ISSUES/036: eight media modules each declared the shared
library and download directories as their own resources, and the
resolver's duplicate-owner refusal — right in general — would refuse the
stack's only sensible assignment the first time two landed on one node.
The decision, from the operator: shared, pre-existing data is
operator-owned and external. The mesh does not create, chown, reconcile
or remove it. A module declares it needs access to such a path (read or
read-write); the host mounts it and owns nothing. Several modules
accessing one path is normal — the duplicate-path refusal is about
ownership, not use. An accessed path absent at apply is refused clearly,
not created. Extends ADR 0030: the third case the host had no word for,
what it neither made nor configured and must not touch.
Point 036's fixed-by/amended-design at the record; mark it located in
mesh-control, mesh-catalog and mesh-host. Regenerate the decision index.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
Turn the completed vendor-agnostic analysis into HQ design. The model-access
provision stays one vendor-blind interface (extends 0024/0027); the
vendor-specific lifecycle moves into a per-vendor adapter keyed by the licence's
`vendor` field, mirroring registrar-scoped public-dns providers (0044), named at
the consumer's real coupling per 0040.
The crux is the sealing-vs-central-rotation carve-out: for refreshable-grant
vendors only, the manager node holds the refresh token encrypted at rest (a
bounded, declared exception), access tokens sealed per holder, refresh stripped
on delivery. Static-key vendors keep full sealing.
Amend 03-DESIGN/01-to-be/14-model-access.md with the adapter generalisation as a
proposed section (prose + diagram, no code); regenerate the decision index.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
Jochen's call, and the right one: a module cannot choose a port well,
because it is written once and assigned anywhere. Any number it picks is
a guess about a machine it has never seen, and two modules guessing the
same number is not a mistake either of them made.
Writing it up turned up something the issue had missed. The same number
appears three times in every module — the rule set, what a consumer is
told, and what the runtime publishes — and nothing checks that they
agree. They agree today because one person wrote all three. A module
whose `serves` said one thing and whose container published another
would resolve, compose, apply, and hand every consumer a port that
answers nothing.
So the decision is one source with the other two derived, and an
assignment made once and kept, as a credential is.
The part that needed thought is ports that cannot move — mail on 25,
submission on 587. Those become claims, which is what the mesh already
has for what is singular on a machine. Two modules wanting 25 is the
same shape as two wanting the seat, and gets refused by name at
assignment rather than by a container runtime at apply. That makes this
mostly a matter of pointing an existing mechanism at ports.
Left open: whether a module should publish to the machine at all.
Assignment makes publishing safe without making it necessary.
The module descriptions sit in `examples/` inside the control plane, and
that name has been doing harm: everything there reads as a sketch, and
one shipped naming a container image nothing builds. A directory called
the catalogue would have made "does this work" the obvious question.
The shape of the answer turns on one measurement. Of the 126 modules in
the system being replaced, 47 are software in their own right — the
largest is 182 source files, and a speech-capture module carries a whole
daemon. Another 44 ship helper scripts. Only 35 are a description and
nothing else.
So a catalogue cannot be a folder of manifests, because two thirds of
modules are programs. That splits them four ways, and only two of the
four belong in a catalogue: things the world made that we describe, and
packages with some files. What the mesh is made of stays in the
repositories that build it. What we wrote keeps its description beside
its code, in the same commit, because nothing else can stop the two
drifting.
The mesh's list of modules is a table, not a repository, and it already
records where each module came from and at which commit. Nothing needs
inventing for modules from anywhere; a repository of ours is just the
source we curate.
The check that a description is valid should move to a command on the
control plane's binary. Today a test reaches into the control plane's
internals to parse manifests, and another reads its build file to check
images exist — two jobs tangled. A command would also give the same
check to somebody describing their own application, which is the case
that matters most and has none.
Left open: how a provisioner's image gets published and pinned, and
whether thirty-five install-a-package modules deserve to be modules at
all.
Bootstrap stopped when the control plane started — a mesh that runs and
cannot be used by anybody not standing at the machine, since the
networked surfaces need an identity provider and no module has been
assigned yet. It now runs through the provider and the first login.
The obstacle was not incidental. The mesh has never held a readable
secret: Make generates and seals, keeping no readable copy. An initial
administrator's credential is the first value a person must read.
Generating it and printing it once was the convenient option and is
refused. It would give the control plane a plaintext secret for the
first time — briefly, and to one terminal, but the capability would then
exist, and an exception made for one case does not stay one. The next
awkward credential gets printed too, and "a copy of the database is a
copy of nothing" stops being checkable by reading the code.
So the operator supplies it, on standard input, not echoed — the path
that already exists for a model-access key. What is created is an
account in the identity provider, not a user of the mesh; there is still
no user model.
Unattended bootstrap remains possible and the value still comes from
outside: automation supplying it is the operator supplying it. What is
refused is the mesh inventing one, so an unattended bootstrap with
nothing provided yields a mesh with no administrator — correct rather
than broken.
The mesh is operated from a command line and must be operable from a
browser and from a model's tools, without becoming three systems. The
pattern is already in the code and was unnamed: `board` serves HTTP by
calling the same functions the CLI calls, holding nothing.
Takes the decision 0034 said had to be taken deliberately rather than
arrive with a feature: the HTTP surface is not read-only, so a browser
login now carries authority over the mesh.
Names the dependency by protocol — an OAuth2 identity provider — as the
mesh does for AMQP, S3 and OCI. Keycloak is what fills the role; what
the control plane knows is that it validates a token, and replacing the
provider is a migration rather than a redesign.
Says what this must not become, because it is the failure the project
was started over: a kernel every module imports, 155 files of code from
every context. Shared surfaces are not a shared library. Three adapters
calling the same functions is not the same as logic leaving the context
that owns it.
And records the loop it creates. The networked surfaces depend on a
module the control plane assigns, so when identity is down nobody can
authenticate — including whoever is trying to fix it. The way out is the
command line, which authenticates through nothing and is available to
the account that owns the machine. Hence the rule: no capability exists
only behind an authenticated surface, because that is a capability which
disappears exactly when identity does.
Supersedes 0032, which decided the right thing and described it wrongly.
The decision is unchanged: the account that installed the host owns the
mesh, and there is no user model.
What was wrong was inventing "a surface that delegates authentication"
for the board. It is a web application with a login, in the way every
web application has a login. That is a fact about an application, not a
property of the mesh.
The cost was not cosmetic. It made the identity module look like part of
the mesh's authority — something the mesh depends on to know who anybody
is — when the mesh knows nothing about people at all and one of the
applications running on it happens to have a login.
Keeps the line that is worth writing down, and states it more plainly:
signing in to an application must not become authority over the mesh.
Today it cannot, because the board reads and does not act. The moment it
can assign a module, whoever it lets in has mesh authority — and it
would arrive as a feature rather than as a decision. So a surface that
can change the mesh is a change to who owns the mesh, and is taken as
one. Not forbidden; just not something that turns up in a pull request
titled "add assign button".
Third correction to one table today, found the same way as the other
two: by asking whether both halves of the test were answered, or only
the easy one.
0006 admits the registry because "it cannot grant itself a repository" —
true, and the second half. Nothing established that the control plane
needs one in order to run. Counted rather than argued: the bundle raises
twelve resources and no registry is among them. The registry arrives
afterwards as an ordinary module, which is exactly what the lab asserts.
0006 half-said this already, calling it "substrate by role and ordinary
by delivery, provisioned once there is a control plane to do it". A
member provisioned by the thing it supposedly precedes is not a member;
that phrase was carrying a contradiction rather than resolving one.
The registry is a closer call than the object store and the difference
is worth keeping: the control plane never touches an object store at
all, but it genuinely uses the registry. So the registry is a real
dependency of the mesh operating and not of the control plane starting —
and it is the second that the word means.
The substrate is now exactly what the bundle raises, which is the
strongest form the list can take: checkable by counting rather than by
reading an argument, and the two cannot drift.
The finding is not about substrates. A test with two conditions is a
test only when both are asked.
Answers what 0031 left open, and a question it did not ask — who owns
the mesh at all. There was no answer, and the absence was invisible
because every operation so far has been run by the person sitting at the
machine, so nothing had to say whether that was the design or the
circumstance.
The account that installed the host owns the mesh on that node. No user
model, no roles, nothing to administer. It follows from 0004 rather than
adding to it: there is no authorisation between nodes because every node
is the operator's own, so a user model inside that boundary would guard
nothing — anyone it could stop could read the node's key off the disk.
The board is different, and the difference is the network. A surface
reachable by a browser has to know who is asking, because those people
are not by construction people with a shell on the machine. So it
delegates to an OAuth provider, which is a module.
That does not make identity substrate. A surface delegating
authentication is not the control plane delegating it: the control plane
runs, applies declarations and reaches nodes with no identity provider
in existence. Only the board needs one.
Records the cost plainly: anybody with a shell on a node has full
authority there, and there is no way to give somebody authority over one
node without giving them a login on it.
Closes the last open question about what the substrate contains. 0006
left an identity provider conditional — substrate only if the control
plane delegated authentication — and said the decision had not been
taken. It is now: it delegates to nothing.
The conditional was never about machines. A node proves itself with a
keypair it generated over a broker account issued at enrolment, and
declarations are verified by signature; none of that involves an
identity provider. It was only ever about whether a person signing in to
a mesh surface would be authenticated by something else.
So the substrate is three — a relational store, a message bus, an image
registry — and with 0028 having removed the object store, no member is
conditional and every one is there for the same reason.
It does not settle how a person signs in to a surface, deliberately.
What is settled is that whatever answers that is not something which
must exist before the mesh does, so it can be decided late or replaced —
which being substrate would have prevented.
0030, found by asking what the conversion actually needs rather than by
reviewing anything. The host deleted a directory and everything under it
when it stopped being declared — which happens when a module is
unassigned, or when a manifest is edited to move a data folder, which is
the exact operation this plan needs. A database's files, a mail spool.
The report said "removed".
A directory still holding something is now kept and said so. No flag and
nothing to remember: emptiness is the test, and it works because the
removal order was already right — the mesh's own contents are gone by
the time the directory is reached, so what remains is by definition
something nobody declared.
The plan now says data outranks its own ordering: copy, read back
through the service that owns it, and only then point anything at the
new location. Never move and then check.
And it records where this starts — the node holding all the production
data — with what that costs stated rather than argued with. Everything
proven so far was proven on machines that could be destroyed and raised
again. A scenario proves the mechanism, not the state on that machine.
Ordering needed no change for the third time running — resources apply
in the order declared and nothing sorts them — and is now asserted,
because sorting them for any sensible reason would have passed every
other test.
Separates ordering from readiness, which the task had run together: a
container started is not a container ready. Nothing waits, and what
needs something usable retries. That is deliberate and more robust than
start ordering, since a dependency can restart long after apply.
The network was the first thing in Phase 1 that genuinely needed
building, and the first that needed a decision: 0029 records why a shape
rather than an action, and the vocabulary is nine.
**0024 accepted.** Model access was decided, built, and proven in the
lab, and two design documents rest on it; only the status had never
moved. The gate is green again.
**The work breakdown rewritten.** It planned a decomposition of the
existing system in place — extract contexts, declared features, shrink
the shared library. That is not the work. A replacement is being built
beside it, and only the old Phase 0 survived contact with reality, so
the one document meant to say what happens next was describing a system
being retired.
Now ordered by what "modules move across one at a time until the old
registry is off" actually requires:
- Phase 0 is marked done against the twenty-two lab assertions, **and
carries its own limitation**: every module exercised was written to
test the mechanism, so the vocabulary was shaped by its own fixtures.
- Phase 1 is the vocabulary gaps found by asking what real modules
need — an object-store provision, a session as a licence consumer, a
network shape with ordering, public certificate issuance.
- Phase 2 is one module, then a week of running it, because the point of
going first is to find what Phase 1 missed.
- Phase 3 picks modules that each prove something the first did not; the
mail system is last because it is the one that may send work back into
the declaration language.
- Phase 4 is switching the registry off, named as a phase so it is not
mistaken for the goal.
Keeps the rules of engagement unchanged — they were about how work is
done, not what it is — with one addition: stop and ask before anything
that touches a machine outside the lab.
Adds a section on keeping the list true, since the document it replaces
was wrong for weeks and nothing said so. A claim here is counted, not
reasoned, and a phase is done when the lab says so.
**0027 — provisions.** A module written against PostgreSQL could be
matched to a provider of SQL Server, resolve as satisfied, and fail on
its first query. The name said the role, so nothing distinguished
engines. Refusing on ambiguity could not help: with one provider of
each name nothing is ambiguous. Enforced at parse rather than
documented, because the old naming was the documentation.
**0028 — the substrate.** 0006 admits an object store on the grounds
that it cannot grant itself a bucket. That answers the second half of
the test and assumes the first: the control plane does not need one.
Verified — no S3 client in mesh-control, and internal/builder/registry.go
records the deliberate choice to put artifacts in the OCI registry as
content-addressed blobs. The row was inherited from the system being
replaced, where an object store distributed module tarballs, and was
never re-tested against the definition above it.
So an object store is an ordinary module, and a mesh with nothing
needing one runs none. Migrating it is module work, not substrate work.
0028 also states what 0006 left unsaid: a substrate service and a
module of the same product are different instances. The substrate is
raised from the bundle before any mesh exists, so it is not in the
module graph — a workload depending on it would depend on something the
graph cannot see, cannot rotate a credential for, and cannot move, and
would put workload data in the store the control plane keeps its own
state in.
Both records were found by reading code against design rather than
design against itself, which is the review that should have happened
sooner.
A session for the mesh itself, addressed as the mesh, differing from a
node's in exactly three things: the context it starts in, its engram,
and its licence binding. Not a new kind of agent — the same mechanism
pointed at a different root. Two implementations of one mechanism drift,
and the vocabulary collision 0001 exists to undo began exactly that way.
It runs on the control-plane node, and the reasoning is easy to get
backwards: not "the important agent on the important machine", but that
this node is already the one place excepted from "compromise of a node
is compromise of that node". Placed anywhere else it would create a
second such place.
It is an addition to per-node messaging and never a replacement. 0001
holds that losing the control plane costs change, not operation — and a
mesh whose only conversational surface lived there would lose the
ability to ask anything while every machine kept running perfectly.
Writing it up exposed that the node session's setup was never designed
at all. 0004 gives behaviour and stops: nothing said how a session
starts, where its context lives, or how a broker message becomes a
prompt. That gap was invisible until something had to be built *like* a
node session. 15-the-agent-session.md covers both as one mechanism.
It also makes "a consumer that is not a machine" undeferrable. The
control-plane node now hosts two sessions that must hold different
licences, and a per-machine binding cannot express that at all. Noted in
14-model-access.md against the gap it was already recorded as.
Also completes the to-be index, which stopped at 10 and omitted four
documents. Pre-existing broken ADR references in the older rows are left
alone rather than guessed at.
Decides the question 006 narrowed to. An agent reads this repository
directly and the search consults it, so these documents surface beside
ordinary results instead of only when somebody already suspects they
exist.
A scheduled sync into the mesh's memory was the option that works with
what exists today, and lost on the ground this repository can least
afford: it makes a second copy, and the copy that is searched quietly
stops matching the copy that is edited. A design record that has
silently diverged from the reasoning it claims to carry is worse than
one that cannot be found — the first misleads, the second merely fails.
Amends what 0019 promised rather than satisfying it: these documents
will not be indexed, they will be read. The commitment that survives is
the one that mattered — that a searcher finds them without already
suspecting they exist.
Gated on an agent that does not exist yet, so 006 stays open on the
build with a decided shape. What closes it is a check that fails today
by design: search the mesh's memory for a phrase that appears only in a
design document here, and require it back.
A new requirement, and it is mostly a shape the mesh already has. A
module that needs to think requires model-access; several vendors and a
locally-run model are several modules providing it; choosing is assigning
the one you want. A model the mesh runs itself needs nothing new at all —
it is a mesh-scoped provision on the node with the hardware, credential
included.
A licence is a named thing because the whole point is saying which one a
given consumer uses, and the names are the operator's. Many to many, so
not a claim: two machines sharing an account is ordinary, not a
collision.
Four gaps, written as gaps rather than design:
- a provider that is on no node, reached over the public internet, which
the reachability rule must not refuse
- a secret the mesh is GIVEN rather than mints. Every credential it
handles today it generated and discarded; an API key arrives from a
person, and accepting one must still discard the plaintext
- a consumer that is not a machine. Which licence a worker uses is a
binding to an agent, and the provisions model has no consumer identity
other than a node
- switching on exhaustion is a reaction to something observed, not a
declaration. It belongs with observability, changing a binding — saying
so is what stops the declaration language growing a conditional
The existing auto-refresh and switching is not being replaced because it
was wrong. It is being rebuilt because it lives somewhere that cannot
express the rest.
First end-to-end raise. A machine with a container runtime applied the
bundle its host carries and ended with a store, databases, schemas, a
broker holding a certificate it generated itself, and the control plane
serving. Then it took a token, checked the broker against the pinned
fingerprint, generated three keypairs and enrolled — the first node being
a node whose mesh is not up yet, observed rather than argued.
And a credential crossed. Declared the provider of a database for a
second node and pushed to over the broker, the machine ended with the
password in one file at mode 0600, and that password appears nowhere in
the declaration that crossed the broker, nowhere in the control plane's
database, and nowhere in what the node reported back. That is the whole
secrets argument, measured.
One fault, in the joining: the token did not say what the mesh calls the
machine, so enrolment needed a flag its own help said it did not, and
failed at the broker with an empty username. It is the fifth thing a
token carries now — the node cannot work its own name out, because the
broker account it authenticates as is named after it and exists before
the mesh has told it anything.
A password nothing was told to create authenticates nowhere. The mesh
generates one, seals it to both ends and cannot read it — so it cannot
tell the software to accept it either. Something on the providing machine
reads what arrived and makes it true.
That something belongs to the module, not to the mesh. The control plane
decides and never touches a machine; a provisioner runs on the machine
and touches it. What the mesh owns is the contract: a manifest of who
asked and where each credential is, and one file per consumer holding it.
It reconciles and is never told what changed, which forces three things
that are each a fault somebody has shipped: set the password every time
or a rotation changes nothing; remove what nobody asks for or a departed
consumer keeps a login for ever; leave alone what it did not make or it
cannot be run on anything that predates it.
Saying where the mesh stops is the point. It decides, delivers, and can
prove what it delivered; the last inch belongs to whoever knows what
`create role` means.
Written after looking at how the existing mesh does it, so this is a
reaction to a measurement rather than a preference.
There, credentials sit in a column encrypted at rest. Its own tooling
records what that bought: the tool for finding a secret matches by value
rather than by name, because the same password is in three tables, in
each node's environment file in plain text, and inside every connection
string composed from it — copies its documentation calls the ones usually
in use. And a query against the encrypted column returns zero rows and
proves nothing, so auditing moved to the decrypted copies.
Encryption at rest addresses neither fault. The control plane can read
what it stores, so a copy of its database is a copy of everything. And
composition is what mints the untracked copies.
So the value is sealed to the node that will use it before it is stored,
with a key that node generated. Nothing central is composed. What it
costs is auditing by value, which was never real anyway; what stays
answerable is which node holds what, which is what rotation asks.
What remains is a provisioner. The mesh generates the secret and tells
both ends; nothing yet acts on the telling.
Which turned out to be the useful way to cut it. A provider says what a
consumer needs in order to use it; a consumer says where it wants to be
told; the mesh adds which machine and what that machine is called on the
private network. So an app on one node reaches its database on another,
by a name the mesh also created.
The file says it carries no credential and why, because a missing field
looks like a bug and a stated absence looks like a boundary.
What remains is the secret itself, and the shape it will arrive in now
exists.
Also: two machines wired together across no private network is refused,
and that only became checkable when the network stopped being something a
machine has by virtue of holding an address.
0009 distinguishes presence from instantiation — what the edge hands
over. It never distinguished where the thing on the other end is, and
that turned out to be the half doing the damage: a shell and a database
were both written `requires`, so requiring a database installed one on
every machine that used one.
A provided name now carries a scope, as a claim already does. Scope
belongs to the name rather than to each provider, or one requirement
means two things depending on which module answers it.
A requirement answered from the mesh is never satisfied locally. Nothing
provides it, and it says which module to assign somewhere; two do, and it
says how to choose. Choosing is recorded per node, because two machines
may reasonably use two different databases.
And knowing which node answers is the first half of handing a credential
back — you cannot be given a database's password before it is settled
whose database it is.
0009 already said a consumer supplies a target and receives a name. What
it did not say is that those are two separate mechanisms.
Contribution — publish me at this name, on this port — now exists.
Binding — and hand me back a credential — does not, and is the larger
half: a secret has to exist, be stored, reach one node and not the
others, and rotate with every holder informed. That is the invariant set
found violated three ways at once, so it is not something to add in
passing.
The absence had a measured cost. Exactly two modules opened a direct
connection to the control plane's database, and they are the reason every
node permanently holds a credential to it. Both were doing by hand what
this edge is for. Neither needed a new kind of thing.
Two records, from building it.
0009 has a section titled "there are no domain modules", and `networking`
now exists. It is not a contradiction and it reads as one, so the
difference is written down: what was refused contains WireGuard and a
proxy and is assigned where half of it is unwanted. What exists contains
nothing — requirements and a name — so there is no half. Every artifact
it leads to is still an ordinary module assigned on its own terms.
With the cost stated, because it is real: adding a second implementation
turns a settled question into an open one for everyone using the bundle,
not only for whoever wanted the alternative. That is the refusing rule
applied consistently, and the alternative is a default, which is the
flavor field returning under a better name.
08-connectivity gains why the network stopped being code beside the
module system: a machine was on the private network because it had an
address, and there was no way to keep one off. A manifest can now say its
resources are computed, which is what a peer list needs.
And three modules rather than one, because WireGuard is one VPN of
several. Naming a module after the job and putting one implementation
inside it is flavor wearing a generic name — the second VPN has nowhere
to go.
jschoubben