The review found §794's backlog and §796's tag at another commit proven
only by unit tests. Both become steps of the builder's whole test on the
embedded bus: a reservation skips 0.2.0 whose tag stands at another
commit, takes 0.2.1 with the finding version-tagged-elsewhere, the
release writes modules/m4/v0.2.1 at the build's commit and the
version-released heard from the bus clears the finding; a message the
handler asks again is kept on its consumer's ledger in the state
consumers on the bus, and the health held six minutes names it.
The health judgement moves from the tools program's main into
internal/versions (BuilderHealth), which the tools program calls, so the
whole test can prove it. The test's comment no longer says the old state
is dropped through the bus module (F1).
Builds: design 47 H2 §794, §796, H-A-453 (3), H-A-435, H-A-462
The tools bundle follows the seat-holders view of the claimant of
mesh-module-assigner, names each build seat's holder per node from it, asks
each holder's current at its node one after another and answers them
together, a holder that does not answer listed with why; it reads no presence
and chooses no node. The manifest lists the tool and the three build seats'
current among the module's invokes, outside the process's bus.
Builds: design 47 H2 §801; H-A-446 (1)
The work queue publishes is declared in the manifest's old streams section
(retention work-queue, duplicates 24 h, max-deliveries 10, 64 MiB, 7 days), the
module's own stream consumed on the module's account, so the bus binding asked
of the broker seat goes; the consumers per holder and class are still made at
run time. The seat mesh-builds gains the read verb publish-queue, served by the
builder from its own stream (counts per holder and class, oldest and newest, a
list capped at 50, capped past 4,000 items read), and builder_publish_queue asks
it in place of the runtime's stream and work-queue reads. Two whole tests of a
gone artifact are folded into one, so the builder's integration count stays at
18 with the publishes whole test.
Builds: design 47 H2 §793, §819, §820; H-A-440 (1), H-A-446 (2), H-A-454 and its addendum
The bucket publish-queue listed and reread itself before each send; its items
now ride the builder's work queue publishes, pulled per holder and class by
consumers the builder makes as it learns a holder, the live class first, a
failed send handed back with the pause, an item acknowledged when its publish
is heard and terminated when its build is withdrawn, superseded or dropped.
builder_publish_drop and builder_publish_first act through reasks after the
hand-act log; builder_publish_queue reads the queue through the runtime. The
old state's items are moved once at the start, then dropped through the bus
module.
Builds: design 47 H2 §793, §820; H-A-290, H-A-291, H-A-440, H-A-446 (2), H-A-448
The builder said nothing when its events backed up or one handling stood still; its health
now reads each durable consumer's oldest unacknowledged message from its state consumers,
and an event handled over 100 ms is said with the part that took it.
Builds: design 47 H2 §794, §820; H-A-435 and its addendum
A line per module drowned the journal at 6,500 plans; the plans live on the bus, where
builder_unbuildable reads them, and the builder says them once per start and per seat list change.
Builds: design 47 H2 §795, §820; H-A-425
A caller whose step ended by its bound could not stop the builder working on
its ask, and nobody could see what waited on the seat's work queue. The
builder now serves the seat's verb drop: an ask it runs is cancelled, one not
pulled yet is kept dropped in its state taken (dropped.<ask>, for
ask-taken-bound); both end the step could-not-run naming the drop and who
asked it. builder_asks reads the waiting asks through the runtime and the
taken ones and the drops from the state taken; builder_ask_drop asks the
seat's drop once the hand act is recorded.
Builds: design 47 H2 §777, §778; H-A-401, H-A-404, H-A-410
The review found two faults. The tool paged on through a state of any size, one replay of the bucket per page; it now
reads one page of each of its own states and, when the state holds more, answers that page capped, with the keys read
and the state's size from its info, never asking a second page and never refusing. And the retiring of superseded done
builds keyed the newest reservation and release by module alone, so a standing 0.2 would retire a done 0.1.x; both are
now found per module and series (MAJOR.MINOR).
Builds: design 47 H H-A-292, H-A-293, H-A-291 (1), H-A-290 (5)
The tool runner's release waited forty minutes behind old versions: every start asked a publish again for every done
build, and each send relisted and reread the whole queue. A publish whose build ended in this run now goes before any
asked again at a start (the class kept on the key's value, a live key of a build before the start made a start key at
the start); a pass lists a holder's keys once and sends them four at a time; a withdrawn or taken-over build is dropped
with its publish item ended at enqueue, at the pass and at the start; a send that returns unsent says why, and every call
it makes honours its one bound; builder_publish_queue reads its states in pages and answers counts per holder and class,
the oldest and newest queue time, a list capped at fifty and one module's items with their class.
Builds: design 47 H H-A-290 (1)-(5), H-A-269, H-A-271, H-A-275
A module value carries no source stamp (C1d): ModuleOf takes none from the
entry's inputs, a build's claim is numbered per fingerprint naming no source
revision and no merge, and the log's started entry names none. The seat list's
parts take no callers and resolve no caller's holder (mesh-sdk's NewPart without
callers, SDK-C). internal/... is green; cmd/mesh-delivery-builder still hears
manifest-accepted (names.go:21, versions.go:192, module.json:57), which the
seat mesh-catalog no longer defines: its floor's wake is H-A-49's, asked of the
lead.
Builds: design 47 H §764 (the indexer: the field pull, the source numbering); lead answers H-A-48, H-A-66 (hq f599b761)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
Asked when the build ends and at the start, deleted by the release, ended loudly after publish-tries
tries and readable in builder_publish_queue; settings publish-pause, publish-pause-max, publish-tries.
The publish no longer waits on running or manifest-reads-changed.
Builds: design 47 G A-457, A-457a, §588
A trunk build reserving while its series is reserved by an older build never
released drops the older reservation's bound and finding and says so in the
journal. builder_reserve_done (a hand act, reasks mode reserve-done) reserves
the series' next version for a done trunk build that reserved none, with the
done key's digests, and queues its publish as builder_publish_again does.
Builds: design 47 G A-455, §369, §588
Its states' buckets, its finding observer, its build log entries and its step asks' asker take the credential's module (tools: the module the runtime serves them as) instead of constants of its name; builds.StateBucket/BuildsBucket replace the Bucket constants. The readers of the builds bucket (checker, view, grants) still name the builder: the next D1 subtask finds it by the seat mesh-builds.
Builds: design 47 G A-296, §428
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The builder, its tools bundle, the assigner (process and bundle) and the view's tools bundle answered
their seat list's again tool to the modules node-tools and mesh-healers. Their lists are now the
operator, seat:node-tool-runner and seat:mesh-healers, resolved by a holder function:
- meshcatalogclient's Branches.Claimant (moved from the builder's versions.HolderOf, which now asks it)
names the claimant of mesh-module-assigner from the trunk's half, and Branches.SeatHolders follows
that claimant's seat-holders view through the part's wiring. It drops the view and follows the new
claimant's on a catalog change, asks nothing on a timer, and answers seatlist.HoldersIn.
- the assigner answers HoldersIn from its own seat-holders picture (its world in the process, the
watched copy in the bundle).
The builder's and the view's manifests consume seat:mesh-module-assigner.views.seat-holders. The
assigner's seat-holders first-view tool answers both, by module as A-131 has it.
Builds: design 47 G A-298, A-295, A-296, A-210, A-131, §717, §616, §614
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The delivery repository's contracts and SDK live under modules/ as
mesh-delivery-contracts and mesh-delivery-sdk, their Go module paths
renamed to match the folders (the TypeScript names already did), and
every importer in mesh-delivery, mesh-catalog, mesh-host, mesh-tools
and mesh-sdk follows in the same group: a design as it should be, one
path per module, checked by the group's check building every importer.
Builds: design 47 G §427, §88
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
module.json reads seat:node-build-<language>.steps as the checker and the watchman do; the
tools bundle reads <holder>.steps, the holder found by seatlist.HolderOf where it is read.
Builds: design 47 G A-116, A-126, A-128
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
serves beside the process in the manifest (no bus), the catalog half's carriers and seat mesh-catalog.versions used in the process; builder_build stays the bundle's.
Builds: design 47 G A-115, A-127, A-87
The builder finds a build module as the indexer's seed does: the build seat whose acting rule
`build` has an `on` matching the artifact's language or kind:image, and that seat's holder, read
from the seat list's definitions and the claims of the catalog half's manifests. Task.Module, every
lane key, the unpublished modules, the toolchain sources, the ends watched, the sayer and
builder_stuck (through `<seat>.steps`) derive from it; the floor's sources are the holders of the
seats git and mesh-catalog. buildrun.GoBuildModule, NodeBuildModule, LanguageOf,
buildstep.BuildModuleOf use, build.ImageBuildModule and StuckBuildModules are gone from the builder
(the checker keeps its own names in two small files until its part reads them from the list).
Builds: design 47 G §7, §428, §704, §204, A-82
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
builder_build reads the builds and asked states, and the tools bundle follows the catalog half itself (registrations for the
watched branches, revision per branch) for `newest`; the manifest names the tool and the two seat verbs.
Builds: design 47 G §589, A-87
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The builder adopts meshcatalogclient's branch follower on its views reader (A-5, A-15): the modules, their manifests (kept from the
entry that holds them, by its digest), their packages and each package's name in its ecosystem are read from the catalog half's
entries, never the indexer's modules, manifests, candidates and packages views; an entry of a branch other than the module's first
builds its fingerprint and supersedes only its own branch's older one; the start's sweep runs once every followed half's first view
is in; the builder's tools bundle reads package names through the same follower. A run's candidate is no input of the builder
(A-84): its read of the indexer's candidates view goes, and the entry point for a run's build ask is not built (no line gives the
ask's form). The seats' events the builder's own handler takes are named per seat so the follower's events have one taker.
Builds: design 47 G §348, §479, §614, A-5, A-15, A-84
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The reservation keeps the fingerprint of the catalog entry it was made from (versions.Source, A-76). The state sections
publish-queue (max-mib 4) and acted (max-mib 1); a consumer acting once per message of the node-engine seat's running and
manifest-reads-changed (why returned) by its stream sequence, the running of every subject read at the start with
lastPerSubject; one queue per holder sent one at a time as the registry seat's publish with the digest and no parts, the
seat and the package name read from the entry (A-73, A-76); builder_publish_queue and the new builder_publish_again, a
hand act. The old form goes: build.NewPublishAgain / ModePublishAgain, versions.Acts.PublishAgain and the
Again* decision, and the integration step that refused it.
Builds: design 47 G §588, §369, §695, A-73, A-76 (G-16, G-18, G-20)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The builder's tools bundle serves reserved given a package's name: it follows the catalog half's packages view itself, finds the
module from the name, answers the reserved version, or for a version no reservation holds asks the seat mesh-catalog's versions
and answers released with the release's digests; a name no entry carries, two entries carry, or a version neither reserved nor
released is answered error. builder_versions by name calls reserved through its invokes. The builder's process serves no call.
Builds: design 47 G §572, §582 (A-70)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
An entry whose fingerprint names a toolchain version of its language other than
the agreed one is held in the state waits (one key <module>.<fingerprint>), says
wait-held and wait-ended (ran when the agreed toolchain names the version,
released by builder_release) on its stream waits, says wait-held again for every
key at its start; builder_waits and builder_release; its setting
toolchain-wait-minutes. Not built: the bound's finding toolchain-waits (its
scope is not in the design).
Builds: design 47 G §672, §267, §276, §633
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The builder agrees a language's toolchain from the three build seats'
toolchain-reported, kept in its state toolchains-heard (read back at its start),
forgets a node's on the node-engine seat's running with no fingerprint (stopped)
and never on a placement, says toolchain-agreed with its digests list only when
the agreed version changes; builder_toolchains and builder_toolchains_drop;
the manifest's reads of the build modules' toolchains are gone. The gate over
the assignments is now PlacedNodes, which only counts nodes.
Builds: design 47 G §492, §604, §546, §472, §574
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The state tag-asks and its bounds, builder_tag, builder_tag_asks and builder_untag, the checker's tag-wanted
consumer, the branch end's removal of tagged versions, the numbering of a check's build under a tag and the
event version.numbered (its contract in builds, its emit, its manifest line) are removed; a forge message still
wakes a floor. builder_publish_again refuses a held build, since nothing says a version numbered.
Builds: design 47 G §373, §321, §349, §413
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The builder keeps one key per module and MAJOR.MINOR in its state reserved, written by compare-and-set when a trunk
build is claimed; nothing is reserved, and the build says wait-held, before the catalog says versions-filled; the key goes
on version-released; a built reservation unpublished past ten minutes is the finding version-unreleased; builder_reserved,
builder_reserved_drop (hand act, recorded) and the seat mesh-builds' verb reserved.
Builds: design 47 G §369, §372, §52, §85 (reserved)
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The put carries the ask's id as its message id, so it is made again through a deadline and no responders within the bound;
a put or a note that still meets one fails the key as a transport fault, which the builder does not retake by itself
(ADR 0664 §3) and the hand tool does. A key left building with no look holding it (its end never written: "claimed", no
ask) is taken as it stands by builder_build_again, from another builder too.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
LogEnded copied the outcome, manifest, bases, artifacts and imports of the build's key but not its error or step, and the log entry had no field for them: a failed build read back as only "failed".
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
Every state a module keeps has a read tool (the lead's ruling). Each key of bounds now holds what it bounds
and when it ends, and builder_bounds lists them with the time left.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
mesh-delivery's sdk/firstview moved whole into mesh-sdk go/views (J3): the builder, checker, watchman, teller and view
follow each owner's state with views.ProcessReader.Follow(owner, state, tool, stream), naming the owner's read tool as
their invokes name it and its views family as the owner's manifest's `streams` names it ("indexer-views",
"assigner-views"); a refused Follow stops the start. Their beds serve the owners with viewstest.Serve and grant with
viewstest.Grants by the same names. The view's tools ask the owners' read tools by their literal names. The assigner
names its own views family and the indexer's itself (views.AssignerStream and IndexerStream are gone).
Every views.StatusTools call takes handacts.Recorder{Ask: stdio.Ask}: a first view asked again by hand is a hand act,
recorded through seat:mesh-hand-acts.hand-act, which the builder's, checker's, watchman's, teller's and view's
manifests now invoke (a row in each manifest test, each failed first).
Row (A2j's written-time switch): TestRunNeedsItsCredential "a want stood since the assigner's bus wrote it": the
watchman's want time (createdOr) is the time the owner's bus wrote the key, through ProcessReader's first view (failed
with createdOr ignoring the entry's time). sdk/go.mod: nats.go is indirect now.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
L3 (710c4032) arrives with it. Conflicts, each kept from both sides:
- assigner module.json, manifest_test.go, main.go, tools.go, integration_test.go: A2's first-view tools,
read tools and states beside L3's assigner_engine_reads and engine-reads; the two "consumes" lists
joined into one (the indexer's views and node-engine.manifest-reads-changed).
- builder module.json and manifest_test.go: A3's tools, states, consumes and invokes beside A2g's first
views (DLV-M1); one findings state (A3's, history 5) carries both; reads keep A3o's
mesh-delivery-indexer.packages until the builder follows it (join item).
- builder cmd main.go: rebuilt from A3's with A2g's changes applied (the reader follows modules,
manifests, candidates and placement; one calls connection; one findings bucket, A3's
health.FindingsBucket; FollowPlacement on a goroutine once placement's first view is answered;
buildFromModules in place of the start's settle). Tools main.go: A3's tools then the first-view tools.
- builder run.go: A2k's out-of-step on the end entry, then A3's numbering of a done build.
- view module.json and manifest_test.go: node_manifest_reads and seat:node-engine.manifest-reads beside
A2j's first views and read tools; one invokes list.
- indexer judge rules_test.go: L3's manifest-reads verb and event, without pull-waits (J1 dropped it).
- sdk/go.mod and builder go.mod: both requires kept.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
ADR 0615 §1: the builder invokes indexer_read_modules, _manifests, _candidates and
assigner_read_placement, consumes their views families and reads none of the four buckets; it keeps
first-views and findings. The lead's ruling: an owner that is down never holds the start; the modules
are built once the modules', manifests' and candidates' first views are answered, the placement is
followed into the toolchain gate once its first view is, and until then the gate passes every report on
as before the placement settled. Tools builder_first_views and builder_first_view_again.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The window opens while a counted node-engine reads, of some kind, none of the newest form sdk/forms knows, and closes
when none does: said once as version.old-form-closed, and a node holding it seven days is old-form-held about it. It
steps on a statement's change, the placement, a hand drop and each build key written, never on a timer. A retired node
dropped by hand leaves the count of the forms a build key keeps (design 51). internal/versions' own Forms goes: which
forms a key keeps is sdk/forms'.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The builder keeps both: the registry's versions, floor, tag asks and findings, and ADR 0618's engine-reads state,
its tool builder_engine_reads, the node-engine seat's manifest-reads verb and manifest-reads-changed event.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
Each outcome is a seat event <seat>.<event> on the shared events stream (the
lead's ruling of 12:4x), heard on the builder's events consumer through its
consumes; the subjects are mesh-contracts'. A restarted builder takes its
floor waits in again, so its silent sources' events still wake them;
builder_floor_waits reads them.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
A holder that failed leaves a version to publish again by hand: said numbered again while the build is held, built
again under the same number when its file is gone, numbered anew after a published; refused for other bytes. The
floor is asked again by hand; a branch's tagged versions are removed through each seat's remove, per repository.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
A counted node is one the placement names; a node with no statement reads as the baseline's first forms, and no node record is read in its place. builder_engine_reads lists, asks again and drops.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
The builder keeps one key per name, tag and version (design 51 *The versions*, ADR 0606), in its old state
section (ADR 0619), history one, 64 MiB with the 80 % mark. A package is numbered per source revision and keeps a
version held published, publishing nothing; one numbered but never stored is published again under it. An image,
bundle or archive keeps its fingerprint's number. An unknown floor builds an image, bundle or archive unnumbered and
makes a package wait (ADR 0617 §1). Merge order and tagged versions below the next version are the package's
helpers. builder_versions answers a name's versions and the holder's question (ADR 0605 §4).
Four claim tests merged into two, so the new table test adds no function (test budget).
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
buildrun leaves mesh-sdk for the builder's own Go module (ADR 0593 §1); the
checker, assigner, indexer and watchman require it by a replace to the sibling
directory. Issue 1930's bound was already on mesh-sdk's main and moves with it.
The builder copies the imports a build module answers to the done key and the
log's end entry, where the indexer keeps them (§2; D7).
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Recovery of photos and link2pay on novox, 2026-10-29: their composition and grant were lost and no tool wrote a done key again.
Quick fix: rewrite the done key unchanged as a new revision. Thorough fix: ADR 0547, the node agent composes its own and the
assigner stops waiting on controller compositions (design 47).
Review round 1 (issue 1126 note 72): the inflight check, the watch and the look were three locks, so a second request or the builder's own look could build the key twice; the answer said asked when the look wrote nothing or the build failed before its asks.
The builder retakes a failed key only when it looks at its module again, which nothing does for an unchanged module. The tool makes it look now, with the claim's own rules, and refuses a key that is done or waiting.
mesh-grants moves to the same mesh-sdk (3203358c), so its shadow user list grants the checker and the builder the
DEAD_LETTERS reads the controller's list will, and the take-over is not blocked by a difference. The checker and the
builder hear DEAD_LETTERS through buildrun.DeadLetterHearing: in the background, again at every reconnect and every ask
waited for while deaf, and keep whether they hear it in their new state hearing, which checker_status and builder_stuck
say as dead_letters. checker_try says that a try ended by the not-taken bound leaves its ask in the stream.
Not done as asked: a module health tool. The catalogue refuses a tool check on a module with no runtime of its own and no
other check beside it (ADR 0227 rule 8, ADR 0366 §1), so the deaf state is said in the status tools and the log.