One commit, one plan: a commit off the trunk — a pull request's head, a branch built by
hand, a rebuild or replay of one — is for checking. The build seat reads from its clone
which branches hold the commit, and the controller records and never registers a build
whose commit is not on the branch the module follows (the repository's default for a
new one), so nothing off the trunk can be sent.
A pull request's check now carries its change plan, computed by the planner: what a
merge would build in which order, what each machine would receive, and what is not an
ordinary send — the bus step, a module waiting for a person, a provider's consumers.
touchedBy is now the only mapping of changed files onto modules — touched, added, and
read by no build — and reachOfMerge the planner's whole answer with the dependency walk.
The merge handler, the plan what-if, the merge gate's width and composition, and a pull
request's check all ask it, so planning and gating cannot disagree. The gate composes
the definitions of the modules a merge would rebuild or add, not every one in the tree,
and the check says the dependents a merge would build after them.
The builder reads a module's own directory (the repository for one built from its root)
and a repository its recipe packages, nothing else. A file in no module's directory was
read as shared code and rebuilt everything built from the repository: 103 modules for a
merge-check.sh added at the catalogue's root. It now touches nothing, in the merge
handler, the release planner and the pull request's check alike, and the gate says so.
de-spiegel's and link2pay's manifests already fail the module check on main; without
comparing against the base branch every pull request touching them would fail the gate
for a fault none of them made. The gate's own rule: what was already so is said.
Every pull request the forge announces is mapped onto the mesh's module graph by the
merge handler's rule (issue 278): touching a module — or adding one — runs the gate
(mesh/merge-gate), its judge chosen by the graph (the controller judges itself, the
node-engine by its validator); a repository of the mesh that touches none runs only its
own merge-check.sh (mesh/repo-check), a warning when it has none. Nothing is left pending:
a repository outside the mesh touching nothing is told so as a pass.
The gate moves out of the per-repository scripts into the build seat, so a script is the
repository's own tests and declares its toolchain (go or typescript). The controller's
manifest names every verb of its seat again (ADR 0132), held by a test.
mesh/merge-gate error: the check could not run: a throwaway postgres:17-alpine could not be raised: docker run --label mesh.build=build-1791318263948250337…
mesh/merge-gate error: the check could not run: a throwaway postgres:17-alpine could not be raised: docker run --label mesh.build=build-1791317509716888018…
An image is not byte-reproducible, so ADR 0236's 'same artifacts is no move'
never held for one: a catalogue merge that did not touch the bus rebuilt it,
and every send to the control node waited for a planned bus upgrade.
The builder now records a source fingerprint per build (module tree, context
trees, bases and toolchains by digest). A rebuild with the fingerprint of the
build it repeats is registered with that build's artifacts, handed to modules
standing on it, holds no push, demands no bus step, and a plan sends and
gates nothing for it. Identical artifacts remain a second way to be no move.
The live tests reached one shared bus and assert, read and remove the mesh's own objects by
their fixed names, so packages run in parallel deleted what each other read and the suite
passed only one package at a time; a red suite read as noise. internal/testbus starts a server
per test, linked in at the nats-server release go.mod pins, and a test holds that pin to the
catalogue's bus image and to the facts snapshot's bus when there is one, so the tests never run
a bus the mesh does not. The waiter test read a timing (the most connections held at one look)
and now reads the state it means (the fewest held across the wait). make check runs the packages
in parallel under the race detector, with a timeout.
Every check the mesh had ran after a merge, on a machine: a manifest the node-engine refused
(236), an identity a real machine's name made too long (263). merge-gate raises the mesh as the
facts snapshot says it is and the mesh with the change, each in a throwaway store through the
controller's own records, composes every machine twice and validates it with the node-engine's
validator, and fails what the change breaks, naming the machine's roles and the module - plus a
manifest the judging controller cannot read, a consumer left out of its grant, a module removed
while a machine runs it, a new module the node-engine would refuse; it warns on a wide rebuild.
The forge's new head of a pull request becomes a check the controller asks of the build seat:
the head and, beside it, the controller the mesh runs, the catalogue, the host and the lab; a
throwaway store and bus of the versions the mesh runs; the repository's merge-check.sh in the
mesh's Go toolchain with no container runtime socket; then mesh-lab's replays. The verdict is
said as checked, an error never a pass, and nothing is recorded or registered.
Every check the mesh had was right about the world it was given and none was given
the mesh's: a real machine's name made an identity too long (263), the node-engine
refused what the catalogue check passed (236). The controller now composes what a
check needs - every machine under a pseudonym of its name's length, its roles,
system, builds, capabilities, assignments, pins, settings and how its declaration
composes; every seat, module and source; the bus, store and node-engine versions it
runs - with no secret, no address and no name, and keeps it in the artifact store
as facts:latest when it moved, or daily. The replaced snapshot's manifest is let go
of, so the nightly collector takes it. S14 raises facts-stale past two days.
Two planned bus upgrades raised healer-wanted, though ADR 0236 never lets
the mesh roll the bus. Instead of naming one more cause, the hand-act
verbs are one table saying which record a person's decision (retire
approve/reject, cleanup delete, bus upgrade, upgrade release-backlog, and
secret rotate after a leak); S15 and `hand-acts` skip those, push and the
other repairs keep counting. Conditions already open for them clear on the
next tick.
The merge of mesh-catalog 7f99fb4a rebuilt 103 modules with the build agent in tier 0, and ADR
0236 recorded it as "a change to the build agent rebuilds most of the catalogue". The agent had
not changed: modules/showcase/index.ts had. showcase is the catalogue's reference module, held
by no machine, and its manifest was not in the merge, so whatTheMergeTouched read the file as
shared code and rebuilt everything built from the repository (88 came out byte-identical). The
agent stood first only because everything is built by it.
Whether a directory is a module is a fact of the repository at the merge commit, so the forge's
announcer now says it: module_dirs, the changed files' directories holding a module.json there,
with module_dirs_said. A changed file inside one is that module's business; only a file in no
such directory is shared. An announcer that does not say keeps the old rule. `plans` what-if
takes the same list as module-dirs.
And a regression for the open question: nothing depends on the build agent except by being
built by it, and built-by never widens a plan, so a change to the agent - manifest or program -
rebuilds the agent alone; what moved beside it is ordered after it.
A send carries the machine's whole declaration, so at the switch to roll the next send of
anything would have carried the old default's backlog, unjudged, to every machine. A gated
send now carries and judges everything waiting on its machine; every other send is refused
or leaves the machine; a release plan walks what waits one machine at a time, the control
node last, and one that fails holds the next until a person releases it.
A plan's send to the bus's machine for another module carried the bus's new build and
restarted it under every machine with nobody asking (2026-10-06). The guard is in the one
send everything uses; only the bus step passes it. A rebuild that made the same artifacts
is no move.
A build that reported applied was sent everywhere; one that then did nothing, served
no tools or broke its machine's word reached every machine. Now the first machine is
judged by the component's health (the core's definitions, as doctor probes H-*, or a
module's own) three times over two minutes within ten; a failing gate puts the previous
build back there once, marks the build, and says it as a condition and an event.
Upgrades roll out by default; the bus is a planned step; a module deleted at its
source is not built (the public-acme plan failure).
D2 raised a resolver urgent on one query that timed out while its machine was
loaded, and its summary carried the resolver's address and socket text, so the
operator channel withheld the whole alert.
- D2 asks every question up to three times, all at once; a resolver that
answers nothing is held for the next run and raised urgent when two runs
in a row find it silent. A wrong answer is still raised at once.
- Findings a single look can be wrong about carry Confirm: raised on the
second look in a row, kept while open, never cleared-and-reraised. Used by
D2 silence, D3 (also asks discovery twice), D6 behind, D9, D13 unmeasured,
probe-failed of the doctor, and blind watchdog rows.
- Probe seat asks (D8, D13) are asked again when the bus brought no answer.
- Summaries name machines and say things in words; addresses, paths,
domains and raw errors move to the evidence (D2, D5, D8, D9, D13, S12).
- internal/outward mirrors the messenger's content rule, allowing the mesh's
machine names; the keeper rewords a summary that would be withheld and keeps
it whole in the evidence; a TestMain lint fails the suite on any raised or
linted finding that would be withheld.
The night's backup of the bus takes each stream through JetStream's snapshot
API, run by the nats module under its own account. The module holding
mesh-broker is composed that account: stream names and info, the snapshot
request, its flow-control acks, its own inbox — no write, which the writers
table checks. A bus module declaring anything else to say on the bus is
refused by module check rather than silently granted nothing. The genesis
user list is unchanged: the controller's grants are.
Between assign and push a module's own secrets are not made yet; D1 composed
without making them and raised an urgent 'nothing can be sent' that the next
push resolved silently. D1 now composes as the push would (Foreseeing): a
secret the push makes gets a stand-in and is named, one the push is refused on
is refused with the push's words. Waiting is said only past 30 minutes, as a
warning. D3 and D13 expect a holder only once its machine was sent it and
reported or had ten minutes to.
A walk over a large library every hour loads the array that protects it. An item now says how it
is measured — a bounded daily walk, a dataset's counters, or its top level only — and a size that
is a lower bound is kept as such and never read as a shrink.
A module's data section says what it keeps and how precious it is; the backup holder's lines,
binding stickiness, retirement on unassign and D13's conditions follow from it, so issue 273's
empty replacement is said and an unassigned module's data is remembered, not forgotten.
grantsFor granted every consumer a pair credential from the provider was
ever made for, so a consumer pinned back to its own store was still asked
of the store it left, which then never retired it. A credential whose
consumer's resolution binds it elsewhere is now withdrawn like one nobody
asks for, kept on record for the login the provider keeps, and said on
plan and push.
Issue 258's fix let a mesh seat's holder elsewhere answer before this machine's own provider. Right
for the resolver, which any provider answers alike; for the store's seat it re-bound every database
consumer on a machine running its own store to the holder on another, each was given a fresh, empty
database there, and nothing said so for twenty hours.
- An offer says whether it keeps its consumers' data (`keeps-consumer-data`); unsaid, a provider
that grants each consumer a credential does. For such a provision the seat's holder no longer
overrules a provider beside the consumer; a pin still does.
- Where each such consumer was sent is recorded (migration 0071). A resolution that would bind it
elsewhere keeps the recorded provider and says the move; one whose provider is gone is refused,
never answered by another.
- A push says a kept move and raises it as an urgent condition at once; the self-check's D12 raises
it every run, with a pinned move not yet sent as a warning and any unasked move as urgent.
networking required mesh-wireguard and nothing else; machines are assigned the network directly.
module forget refuses a provided module, so a retired one is removed at start once no machine has it.
Guard route-proxy's public account directory against a reissue.
Approving a retirement and deleting what was retired are a person's act by
design (hq ADR 0230); counted by S15 they would ask for a healer that must not
exist.
A provider that cannot disable keeps the consumer reachable until a person
deletes it (hq ADR 0230); the listing and the approval say so instead of
claiming access was disabled.
A provider now waits for a person before retiring more than three consumers
or half of what it holds, and deletes only when asked. The controller is that
person's way in: it keeps waiting and rejected sets as conditions, answers them
with retire approve|reject, lists and deletes retired consumers through the
provider's own tools on its machine, records each act in the hand-act log, and
probes for anything retired longer than thirty days (D11).
mesh-admin
jochen
jschoubben