The node's own composer reads the catalog half through meshcatalogclient's Entries cache over its own calls of
mesh-catalog's verb entry, and no longer watches or reads the indexer's modules, manifests and verdicts or the
builder's builds: a module's manifest is its entry's, each artifact's reference is made from the digest its want's
part names (artifact-store://<module>/<artifact>/blobs/<digest>, A-50). A want whose entry is asked or not found is
held, its have the phase held with a detail opening entry-asked: or entry-missing: naming the fingerprint. The
entries and wants it composes from are kept in composed-<node> (entry.<digits> and want.<module>), written before
the composition that uses them, read at start before anything is asked, a kept want standing until the wants'
live read answers, and dropped when nothing names them. The shadow keeps its reads (G §628).
The node grant gains the verb entry and the running stream's consumer calls; OverNATS hears the running stream
as the link's session does.
Builds: design 47 G §421, §683, §614, §628, A-47, A-50
A want that names a node in kept-on, for an artifact its parts name, is fetched through kept-artifact of that node's build seat and
read as an object on the bus, checked against the digest; with no kept-on it is fetched from the store, and the builder's builds
are read no more (the apply hook, the fetch and the object read from bc8fbc7, its node and fingerprint from the want).
Builds: design 47 G §271, §526, §421 (A-50)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The seat list's five-minute finding and the finding of a held module are raised as they happen, so they use the held
writer (seeded at Start, Held() answers what stands) over the contract of the seat mesh-conditions; the keepers keep no
set or sweep of their own, and the writer is started before the seat list reader writes or clears any. The engine's own
finding stream stays its foundation stream mesh-host-findings.
Builds: design 47 G §529, A-29, A-30, A-37, A-38, A-40
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The join with mesh-contracts' K6 (7d730a4a) removes the event and its body; H-3's second source of the same fact goes (haveState.capsSaid, capabilitiesToSay, sayCapabilities), the test asserts none is said.
Builds: design 47 G §590
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A finding is written in the node's state findings and followed by finding-written or finding-cleared on the engine's own
stream mesh-host-findings; the engine states every finding it observes at its start (the seat list's finding only while the
reader waits for the list, else not) and when the keeper asks, and answers an ask by a new statement. The node's grant in
the golden file gains the stream's publish, the ask stream's consumer, and the whole of its state findings.
Builds: design 47 G §383, §384, §390, §461, §485, §503
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The node follows the node-engine seat's running events (an ordered read of the last message of each subject, then a durable
consumer), keeps what each node runs by `at`, acts on none until the first read ended, and hands the composer each node's
heard build with the manifest read by the event's digest; compose resolves the passes through catalogue.SeatsToPassOn
(heard, entries) with the trunk's entry for a node that said nothing.
Builds: design 47 G §199, §249, §459, §536, §556
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A new node-engine process cleared nothing at its first full read of the seat list, because the Waiter remembered in memory whether
it had written the finding; ace and g14 kept the condition for ever. The Waiter now asks the node's state findings whether the
finding stands, clears it at the first full read and writes it at the bound only when it does not stand.
Builds: design 47 H-7, ADR 0636 §1 (the finding is cleared at the first full read), issue 1966
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/delivery-group group fix/1966-call-made-once rejected: composed together: the checker's verdict at the members' heads (c54b617d200d): mesh-delivery-checker
mesh/check failed: mesh-delivery-checker: no published of the tagged version of go-meshconditions of mesh-health-and-re… | 9cd03a7430cc on f50442ac7a2d
A failed say was only a line in the engine's log; now it is said again when a link opens, beside the
statement made while no link was open (proved by a step of the node key test, with H1's publish).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
One publish of manifest-reads-changed with a 10 s deadline was the likely loss behind note 155; the
stream takes the id once, so publishing again is safe.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/check checking 5c353cfc7bf0 on f50442ac7a2d | novox/mesh-host#221 in group fix/1966-seat-list-unread
The node-engine seat's definition names "the operator and the healers" for
both seat-list verbs, and design 47 G keeps it; the read was opened to every
proved caller on a ruling the review withdrew. node-tools stays on both: every
console call is proved as its machine's tool runner (issue 627, issue 1966).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
seat-list-again answered person.<name> and the healers, but every call through
the mesh MCP server is proved as its machine's tool runner (issue 627), so no
operator's session could make a node read its seat list again. node-tools now
stands for that session; the read seat-list is open to every proved caller, as
ADR 0636 §7 limits only the read again. The verb names come from the contract.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A request that reached no one is asked again by mesh-sdk's busretry only when
it can see the bus's own error; the node-engine's ask dropped it, so the seat
list's first read, which fell into its holder's re-serve, was never asked
again (issue 1966).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/check pass 7b42c94a4d78 on 87bdc95a3d65 | unit 1679 run, 333 kept, 31 skipped | integration 1692 run, 333 kept, 26 skipped | over bound: unit (sl…
The link serves the seat node-engine when it opens, before the node agent
starts the reader of the seat list; since ADR 0635 catalogue holds no seat
until then, so the engine registered no verb at all. Nobody answered
manifest-reads, compose, composition or the hand acts on novox, shanks and
now ace, and the assigner kept ace's old statement, holding its own
successor on seats the new engine states.
Quick and thorough fix are one here: the only seat-list read at start is
this one (the composer waits for the list, the reads statement restates).
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh/check failed (2): mesh-delivery-checker: no published of the tagged version of go-sdk of mesh-delivery…; and 1 more | 9ab85ef92e6c on 875b4ee5ce38
manifestReads froze the statement at its first call, before the engine gave
catalogue even its bootstrapped seats (ADR 0635 empties mesh-sdk's table
until the list is read), so ace's engine stated no seat and its own next
build, naming mesh-catalog, was held on it.
The words and forms stay the build's, computed once; the seats are those of
the list held now. The reader of the seat list restates after each change:
the node record carries the new statement at once and the node seat's event
manifest-reads-changed says it (ADR 0618 §4); an unchanged statement writes
and says nothing; the verb manifest-reads answers when it was made.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
The review of fe9533c1: a new test function raised a count already over the
budget, so two broker-pin tests are one; KVWatch no longer yields every
change, and its doc says so (issue 1966).
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
Live fault 2026-11-08: node-engines on two machines logged only "nats: slow consumer, messages dropped" and applied nothing. The
client's watch sends into its channel from inside the subscription's callback; KVWatch forwarded that channel to its reader
blocking, and the reader (composing for minutes) stopped draining, so the subscription's pending limit overflowed.
Quick fix: a larger buffer on the watch's channel; it only moves the limit. Thorough fix (chosen): the forwarder always drains the
client's channel and, while its reader is busy, coalesces the pending updates by key (newest value and revision, first-seen
order, Caught kept in its place), so no reader's pace can back the client up, whatever the burst. A slow consumer error now
names its subscription and what it holds pending.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
The state, events and types of the seat mesh-module-assigner say assignments; the legacy node key form is gone from the
reads; the dead seat-verb list of the retired name is deleted. The placement guard and the placed-file sense stay.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The engine's Composed keeps the catalogue's Plan and the declaration types and goes through nodeengineclient's codec; the fingerprint key part is meshcatalogclient's. The candidate answer is nodeengine.CandidateAnswer; the have still holds it through an interim JSON bridge (link.interimAnswers) until the Have group moves.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The provision artifact-store retires, so the store is found by the holder of
container-image-registry and the listen it opens to the mesh, on the port its
node put it on; every name of the store, oci-artifact-registry's too, answers
that address. The engine states form:seat-reach with its contract.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The image, the seat and its holder, when the pull began, and the last retry with what triggered it ride in the failed
have's pull-wait (mesh-sdk delivery.Have.PullWait, asked of the lead in j1-sdk-requests.patch), from the agent through
the link's module fact to the have bucket; the agent's own read view of them goes with the verb. Rows in the agent's
pull test and the have bucket's test; each failed without the code.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
ADR 0627 §1, §4: mesh-sdk keeps no asks, asker, vault answers or compose-candidate checks, so the node-engine takes them
from mesh-contracts: operator-channel's contract and client (asks, warrants, the asker's client), mesh-vault's sign verb
and answer, and node-engine's own contract and client for compose-candidate's ask, answer, verdicts and bound. The
lead's ruling on pull-waits (no record gives the seat that verb): the verb and its handler go; the row that read the
waits now holds the seat refusing the verb, and failed before the change. The go.mod requires of mesh-contracts' modules
come with the group's pin.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The verb's answer and the event's body are mesh-contracts' nodeengine types, not mesh-sdk's, and the seat's verb and event
names are the contract's. The two reasons of the event, started and returned, are the engine's own words beside it.
Requires mesh-contracts and the registry milestone's mesh-sdk (a go.work, uncommitted, until the group pins them).
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
mesh/check failed (2): unit: mesh-provisioner-client/cmd/mesh-provisioner-client.TestNatsANodeComingBackFil…; and 1 more | 32a99144f65e on 798d73eb11e9
recordOf writes catalogue.ManifestReads in NodeRecord.ManifestReads, once per build, so the assigner moves a want only
to an engine that reads its manifest. A row holds the engine's own manifest to the baseline's words, so its new build is
never held for a word only it reads.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
mesh/check failed: novox: not composed on novox: these assignments cannot be applied: | b8158affa7af on 27a956448c0a
An endpoint's subscription hands its asks over one at a time: two checks asking
one node at once queued, and the second answered at about 100 s, past the
checker's bound and the bus's one-minute reply window. compose-candidate is now
answered in a goroutine of its own, at most four at once, one more refused in
words. The new cases are rows of the existing tests (test budget).
Issue 1126, review round 1.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
jschoubben