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
A reference table goes stale the day somebody adds a field, so this one points at
mesh-catalog/modules/showcase — a module that uses all of it — and a test that
fails when it stops doing so. Read the module when the table disagrees with it.
Two rows in the coverage survey were stale because of this week's work: systemd
units were a file plus a service, which made every author write unit syntax and
is why "process" exists; and building from source was images only, where a
bundle now names a language and lets the mesh choose the toolchain.
And the two rows at the bottom of the resource table are the interesting ones.
"action" is refused to modules outright — the link may not carry a command, so a
module needing something done ships a program that reconciles. "service"
installs no unit by design, right for software shipping one and wrong for code
the mesh built, which has none until the mesh writes it.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
The first draft said the builder refused archives. It does not. An archive is
packed deterministically, hashed, published by digest, fetched by the machine and
unpacked — the whole path exists. Only the local builder used at genesis refuses
one, and deliberately: an archive is bytes that mean nothing until something
serves them, and at genesis nothing does.
What an archive cannot do is compile. Its source is a directory packed as it
stands, so shipping compiled output means compiling somewhere first, which means
a Dockerfile — the burden this document is about. The gap is not the artifact
kind. It is that no recipe both builds and packs.
Found by reading the builder rather than the manifest schema, which is where the
first draft's claim came from.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
12-a-module-repository says what a module may build and where it goes. Nothing
said how a build is MODELLED, and the model is the problem: a recipe is implicit,
singular and always a Dockerfile; a toolchain is not modelled at all, arriving as
two build arguments the module hand-writes; a language is not a concept; and an
archive is declared in the manifest and refused by the builder.
The cost is measurable rather than theoretical. Adding a module with its own code
means repeating an incantation - two ARG bases, a specific working directory so
the SDK resolves upward, the compiler invoked by absolute path because the usual
symlink is resolved away when the base is assembled, a second stage, an env var
naming the entrypoints. Most of the catalogue is unconverted, and two conversions
done in one session were each wrong twice with a working example open.
So: recipe becomes explicit with three kinds, and toolchain becomes derived from
a declared language rather than written by every author. A Dockerfile stays, and
stops being compulsory - it is right for software needing a particular base and
wrong for "compile my module's code", which is the same operation every time.
The cost is stated before it is chosen: every language is permanent, and the
contracts are already expressed twice - Go structs and TypeScript types kept in
step by hand. A second language makes that drift. So language-neutral contracts
come first, or the drift gets worse while hiding.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
The packet filter generates its rules from what modules declare they listen on.
The broker is not a module, so it declares nothing, so its port is not opened.
Every machine dials that port to enrol and to receive every declaration it is
ever sent.
Invisible where it is assembled and fatal on the next machine: a mesh of one
never dials its own broker across the network, so the ruleset looks right. The
first machine to join is refused at the packet filter during enrolment, several
steps from anything that reports it — and assigning the firewall before joining
machines is both the natural order and the one that breaks.
This is 051 in a second place. That issue says the substrate cannot be updated
because the mesh holds no record of it; the same absence means the firewall
cannot know it exists. Ssh is already a floor for the same reason — a machine
nobody can reach is a machine nobody can repair — and the broker may be the same
class of fact.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
The store and broker come from a bundle the installer writes once, with images
pinned in it, and nothing can change them afterwards: no build, no version to be
behind, no roll-out, and no way to report being out of date, because the mesh
holds no record of them as modules at all.
That is backwards. They are what everything else depends on, so their updates
matter most, and they are the only things with no mechanism to deliver one. A
mesh with a year-old broker reports itself entirely current.
The fix probably already exists: the control plane is carried, raised and then
adopted as an ordinary module pinned to what is running. Nothing in that pattern
is specific to the control plane. It would also remove a duplication visible on
any one-node mesh — the same postgres image running twice, because a store that
cannot be a module cannot provide a database to anything.
Recorded with the two hard parts stated rather than waved at: upgrading a store
the control plane is reading from, and upgrading a broker over the broker.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
A mesh raised from bare metal built six modules and its catalogue reported three:
exactly those built after it began running. Missing were the shared base, the
store, and the catalogue itself.
The hole is never random. On a fresh mesh the modules built before the catalogue
are by necessity the ones it needed in order to exist, so the foundation is
always what is absent, on every mesh, at the moment the graph is first populated.
It breaks the question the catalogue is for: build edges hang off the base, so a
catalogue with no record of it answers 'what must be rebuilt' confidently and
wrongly. And nothing reports the gap, because a catalogue cannot know what it was
never told — this was found by comparing its answer against what the mesh had
just been watched doing.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
A module's broker account is scoped to what it declares it emits and consumes. A
tool call needs a reply queue, which that scope does not cover and should not. So
the account is right, the request is reasonable, and no account exists that can
make it — asking a module its own question, from its own container, with its own
credential, is refused.
It matters because a module's tools are its operator-facing surface: the
catalogue serves the five questions it exists to answer and nothing can reach
them. It is also why a running module keeps being mistaken for a working one — a
test that cannot ask anything checks a container is up, and that substitution has
hidden two faults this week.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
The registry says every machine pulls from it and opens its port to the mesh for
that reason. A machine that tries is refused by its own container runtime: the
registry serves plain HTTP and anything but loopback is treated as HTTPS.
It has never failed, and that is the finding. Every proof that a machine can
fetch a mesh-built artifact was a proof about the machine that built it, where
the reference was loopback. The bed that uses a routable address gets away with
it because the harness writes the runtime's configuration before the mesh exists.
Kept separate from 042, which they are easy to confuse: that one is about not
being allowed to pull, this one about not being able to whatever the credential
says. Fixing 042 alone leaves a machine with a valid account for a registry its
runtime will not talk to.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
The report covered ports the firewall cannot close. The matching fault is ports
it does close and should not: rules come from what modules declare, a migrating
mesh knows about almost nothing, and anything listening on the host is dropped.
No module declares an ssh port and the generator has no allowance for one. The
session that loads the rules survives on conntrack until it drops, and then the
machine is reached from a rescue console.
Rehearsed on three lab machines before doing it on an anchor: loading the mesh's
rules refuses an ordinary host port and leaves a published container port
reachable, same prober, same second.
It is deliberate — no forward chain, because dropping there would stop every
container — but traffic to a published port never reaches the input chain, so
the firewall is silent about it. The anchor publishes 38 such ports today, all
filtered by the system being replaced, so the cutover would open every one of
them while reporting a firewall that is up.
Found while migrating the first module. Five approaches were tried and observed
to fail, including resolving the index and naming the platform at both ends; a
speculative fix was written and reverted rather than shipped, because it did not
make the mirror work.
One module uses this today, which is why it went unnoticed — and it is the shape
most of a migration wants, because the services being moved are third-party.
The design said how the builder arrives was unsettled and that nothing
installed it — the one gap stopping a fresh mesh from producing anything. Both
are now false. What is still true is narrower and worth keeping separate:
nothing asks a raised mesh for the rest of the catalogue, and no bed asserts
that it could.
The mesh could already say which modules a base change invalidated, and could
not do anything about it: each recipe named one particular copy of the base by
fingerprint, and rebuilding produced the old one. Worse, the copy each named
existed only inside a throwaway lab, so those three modules could not be built
anywhere at all — and the line each replaced had the same fault.
The builder's arrival was the one rule the document said nothing checked. It is
checked now, by both genesis beds — and so is the thing that distinguishes a
built control plane from a carried one, which every earlier assertion accepted
either way.
The section saying the change was decided and had not happened now contradicted
the section below it. It also records the argument that failed, because a reader
will otherwise ask the same question and reach the same wrong answer.
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.
Cloned alone, built by the mesh, every module moved onto it, and the base then
changed for real: all three went stale naming what moved. A comment-only change
correctly makes nothing stale, which found a second bug — staleness compared
commits where it should compare artifacts.
The runtime every module compiles against cannot be built by the mesh, so the
one rule that would catch it moving can never fire. And a container keeps the
values it was created with, so two good applies can leave it running on neither.
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
A build machine was refused the build queue, and this was raised as a gap in what
a manifest can express. It is not: `builder issue` creates exactly that account,
three lines from the code being read at the time.
Kept rather than deleted, for the one real thing in it — the wrong verb succeeds
and reports success, producing an account that authenticates and can do nothing,
so the failure surfaces a layer away as a permissions error that reads like a
missing feature.
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
The one complete account of standing a mesh up was an integration test, and a
fixture is free to invent what it needs — which is how a registry that exists in
no production hid two faults for as long as the lab existed.
Written from what the installer does, not what it should do: genesis and joining
are separate moments, the lab runs the installer rather than describing
installing, and three things that are not true yet are named rather than glossed,
including one rule nothing checks.
Claude-Session: https://claude.ai/code/session_01D6qtiYU3P9jk3pnAXyAFyx
Found by deleting the lab's registry and giving the machines a real path out:
public images fetched, the operator's own could not be fetched at all. There is
no provision for a registry credential, no manifest field, and no step in
enrolment that establishes one.
It applies to the mesh's own store too, which today asks for nothing — a
decision that has never been written down as one, and so reads as an absence.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
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
Nine references now name the digest their tag resolved to. What closed is the
immediate fault; the open questions stand, because a digest in a repository is
wrong the moment anybody rebuilds — which is the reason the design wants the
repository to name artifacts and the mesh to hold digests.
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
The doc's own rule is that the lab verifies capability by outcome, never by
reading a setting. A path out is exactly that kind of claim — a route and a
policy can both read correctly while nothing gets through — so it is asserted by
fetching something, in the table with the rest.
Claude-Session: https://claude.ai/code/session_01LrgweAeERJYBg88c5cKDzF
A container runtime on the same workstation sets the kernel's forwarding policy
to drop, and the virtualisation daemon's own accept rules do not override it —
both are consulted and a drop anywhere is the answer. The machines then get
addresses and resolve names, because the daemon's resolver is on the bridge, and
discard every packet to anything real.
The sharpest 'available is not adequate' yet: nothing is misconfigured, nothing
logs, and it presents as every image pull hanging. A workstation that runs
containers is the ordinary case, so it is a prerequisite — verified by reaching
something, never by reading a setting.
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
jschoubben