G starts: grants composes the first rights from the trunk, the indexer answers at once and knows its build, unpublished packages build at the source commit
#217
Merged
mesh-toolsmerged 152 commits from fix/47-G-start-2 into main2026-11-16 14:15:13 +00:00
Placing the steps of a pipeline that has none says its pipeline-started, a
set-aside one included. The seat's acts entry for step `pipeline`, the wait step
value `pipeline` and the wait's `clock-from` go.
Builds: design 47 G A-446, A-446a
A take-in that changes the runner starts its pipelines as any take-in does; the
state bounds' held keys, the wait's `on` value `runner version`, the finding
pipeline-waits, the verb release (and the tools runner_release and runner_waits
that only served it), the Held field and the runner's build-end and running
hearing go. A pipeline a previous runner left set aside is placed and asked
at the runner's start as a newly started one.
Builds: design 47 G A-446 (replaces A-195, A-206, A-218, §266's hold)
The list giving every account every subject was held at 23:20:32: the guard reads each seat's holders from the text of
the subjects users subscribe to, and > names no seat, so it saw every seat lose its holder.
The operator, 2026-11-15 23:12 CET: the mesh accounts should have all the permissions, otherwise it becomes a mess;
the operator can do everything, always. Permissions derived per module refused the indexer's chunks, seat asks and
seat verbs again and again while G is delivered. The approval mark's and the desk's denies stay; an enrolling node
keeps its few subjects. The research after G replaces the accounts themselves (mesh, agent, operator).
The validation of mesh-delivery #219 failed twice: the indexer answered seat:mesh-catalog.input, and the bus refused
its chunks, since only an account whose process repeats the verb in serves was granted to send them. The indexer's
and the assigner's programs answer their seat's input verb on the module's own account; their claims say they serve it.
Every step of the first pipelines ended 'could not run (the request names a module directory and no path)': the runner put
the entry's directory in the fetch without paths, which the build seat refuses. It now names the directory with the
artifact's named paths from the entry, and else neither, so the commit is fetched whole.
The validation of mesh-delivery #219 failed: the check program on shanks asked seat:mesh-catalog.input, whose answer comes
in chunks, and the bus delivered none, since only an account naming the verb in invokes was granted its chunk subjects.
Inside the mesh any module may call any seat verb (the operator, 2026-11-15).
An ask's end read 'passed' when its end was said, whatever the step's outcome: five pipeline steps read passed after 0 s
and nothing said whether they ran tests, carried them or had nothing to run (failure must be loud).
The bus refused SEAT_MESH_BUILDS's and SEAT_MESH_WANTS's create: the part's account is granted each stream it creates by
name, and the ask streams were not named. They join the declared streams, as the kept-event streams do.
A build module pulled a waiting delivery before any check, so 89 queued deliveries held every pipeline's check back on
shanks for an hour and more. The pull now starts at the kind after the one its last ask came from; neither kind starves
the other (the operator's yes, 2026-11-15).
The pipeline runner's step ask to mesh-builds timed out: it publishes on the seat's accept subject through JetStream and
no stream held mesh.seat.mesh-builds.accept.>. Before G the controller made SEAT_<SEAT> for every seat that accepts work;
the seats whose queues came with G (mesh-builds, mesh-wants) had none. The grants part now makes each such stream, from
the seat list, at every pass, as it makes the seats' kept-event streams: only where the bus holds none.
The bus refused the indexer's validation ask and the pipeline runner's step asks: a named ask was granted only from the
seat definition's callers and a composed manifest, and both accounts were kept from the list in force without them. The
operator, 2026-11-15 about 18:10 CET: no fine-grained permissions inside the mesh; every account of a module, its
processes, its node and the tool runner is granted mesh.seat.*.accept.>, composed or kept. A seat's events and verbs keep
their own rules until the research after G decides the whole shape.
The indexer leaves out a package's call word on a cycle (withCallsLeftOut, A-361), but the node check refused a left-out
word the module's own invokes do not name: once the node-engine's manifest stopped naming mesh-catalog.registrations
(mesh-host #229), the trunk revision the indexer wrote could not be read back, and the indexer stopped at its start.
The pipeline runner heard every take-in.taken with branch "" and source 0 and started no pipeline, asking the bus
for each again for ever: the indexer's event body never gained design 47 G §476, §493 and §648's fields. Every taken
event of one take-in now names the watched branch, the source revision it makes (one above the newest seen or said,
A-431) and one entry per module written; a taken event with no entries starts nothing and is journaled (§493).
Bootstrap circle on novox: the node-engine's grant to ask a registry's versions follows the node-engine build wanted
there, which needs a passed pipeline, which needs the pipeline runner that this want composes. With the parts beside
the version, the node-engine (8a0c392) composes by its parts and asks no registry. Recovery (lead-47).
The user list in force on novox stayed at 14:53 UTC while the shadow list changed, and the pass said nothing of why;
each refused create of a state's object store waited out its bound, one after another. Recovery mode (lead-47).
A state of the field state had no <state>-objects store: the grants part made and granted object buckets from
the manifest's objects alone, so the indexer's state inputs (A-237) had no inputs-objects, and every validation
failed on a refused STREAM.INFO of OBJ_mesh-delivery-indexer_inputs-objects; the assigner's inputs likewise.
objectsOf now gives each such state its store (its size, 16 MiB when it names none; none for a state only read),
so the grants part creates it and grants it to its owner as the module's own bucket.
And a state asked in requires by the template state: the bus module's process may now make its store too (the
grants part's BusStateStreams names it, as mesh-sdk's does), and a kept binding's objects is among its streams.
Builds: design 47 G §58, §607, A-237
The rules' hold at F rebuilt the want without its version, so novox's want of mesh-delivery-pipeline stood as
{fp, self}; 98875eba names the release after the self derivation. This row fails without it.
Builds: design 47 G A-430a
A want of a released build names the module's version and no parts (A-430, A-430a): the node-engine resolves each
artifact at that version through its kind's registry seat. Every want that names a build gets it, whichever rule made
or kept it, a self want held at its fingerprint included: novox's want of mesh-delivery-pipeline stood as {fp, self}
since the rules held it at its fingerprint and that path never named the release. A candidate's added module carries
the version alone. A build the assigner knows no version of (another branch's, a kept one: build.ended names none) is
named by its parts as before, until the builder says the version it numbered it under. The release lookup says why it
misses when no assignment graph revision is kept.
Builds: design 47 G A-430, A-430a, §379, §526
The want of mesh-delivery-pipeline on novox stayed with no version and no parts after d1ed5fd5, and the assigner said
nothing. Each want the pass writes or keeps at a fingerprint with no parts is now said once per reason: whether the
node's engine reads parts (its record's capabilities) and whether a release or kept build of it is found. A want held
for its dependency's delivery (§376) is said when the hold starts.
Builds: design 47 G §379, §525, §376, A-420 (a path that says nothing)
The want of mesh-delivery-pipeline on novox named no version and no part digests, so novox's node-engine of G composed
nothing: the assigner looked for want-parts among a node's manifest reads, while a node-engine of G states it among its
record's capabilities (design 47 G §525), so no node ever read as able to take parts. Both are read now.
Builds: design 47 G §525, §526, §379
Every proposal whose verdict passed waited on the operator's yes. Which are asked is now the assigner's setting
ask-operator-for, a list of change kinds (place, take-off, pin, unpin, hold, unhold, decide, config, add-node,
remove-node, and stops for a verdict whose answers say a node would stop a module), its default take-off, remove-node,
stops. A proposal none of whose changes is listed is applied at once, said in the log and by assignments-changed,
naming who proposed it and why. The key is declared in the manifest as the two others, filling a file the process reads
at each pass; a change of it is said.
Builds: design 47 G A-425, A-425a
The pipeline runner's want on novox waited under rule 8 for seat:mesh-hand-acts.hand-act, which mesh-hand-acts serves on
novox from a healthy build the catalog holds no entry of: its have stands in phase held (entry-missing), and rule 8
counted only phase running. A node that runs a healthy build of a part serves what that build serves, whatever its
version and whether the catalog lists it: a healthy have held or waiting for a yes counts, unless its running statement
says it stopped or is not healthy, and the rule says which build it counted.
Builds: design 47 G A-426a, A-426, ADR 0614 §2
Rule 1 held mesh-delivery-pipeline's want as being built after a restart lost its build end, though 0.1.6 was released.
A build whose version is released counts as done for rule 1, and its build.ended counts only for a build not released.
Builds: design 47 G A-429, §379
The pipeline's want on novox waited for an assigner build not yet built, while a working assigner ran. An acting wait now
holds only while the acting part runs nothing serving the edge's seat on its node: a running build whose manifest claims
the seat (or one whose manifest cannot be read, the part holding the seat by its claim) ends it at once, whatever its
newest is, and the end says which build it took. A wait on a version not yet built is never held for that alone.
Builds: design 47 G A-426, §266, §268
Rule 1 held mesh-delivery-pipeline's want on novox as being built, though its build ended at 00:57:36: the assigner
reads the builder's stream of build ends whole at its start only when the bus keeps it last per subject, and the bus
keeps MESH_DELIVERY_BUILDER_BUILDS by age (168 hours, no max per subject), so after each restart it heard only the ends
said since. It now reads the stream from its first message whichever way it is kept (A-233), the last end of each
fingerprint standing. The integration test's stream is kept by age, as live.
Builds: design 47 G A-181, A-233, §556
The pipeline runner could never be wanted: a want follows a passed pipeline (§29), and nothing passes one before a runner
runs. While no module claiming the seat mesh-pipeline-runner runs on any node (read from the node-engine's running
statements), a module claiming it skips that gate and is wanted from the trunk's entry by the other rules, said once.
Every other module still waits for a passed pipeline.
Builds: design 47 G A-428, §29
The want of mesh-delivery-pipeline on novox was not written and the assigner said nothing: a module whose fingerprint no
pipeline passed at on its branch gets none (design 47 G §29), and that return was silent. It is now said as any rule's
reason, once per node. A put refused because the builder holds no build of its fingerprint now names the build the
node's running statement says it runs, so the put of A-427 can name it.
Builds: design 47 G §29, A-427, A-420 (a path that says nothing)
A put of ace's assigner at the build it runs (a hand start) was refused, since the builder's store holds no build of
it, so the trunk's newest build would replace it. Where the store holds none, a put naming the fingerprint the node's
running statement names is held there with the composition the node's have names; the store is not asked. A have
naming no composition at that fingerprint refuses it, saying so.
Builds: design 47 G A-427, A-426
The want of mesh-delivery-pipeline on novox was not written after the operator's yes, and the assigner said nothing:
its entry's acting edge place waits for the assigner's own catalog version to run on ace. Each acting wait the
assigner starts is now said, naming the entry, the part, the step, the version waited for and how it ends.
Builds: design 47 G §266, §268, A-420 (a path that says nothing)
The operator channel keeps no count of an asker's open asks (A-424), so the teller waits for no room: each of the
assigner's asks is asked when it is written, and the asks it could not hand the router are asked again at each end and
at its start. A key whose ask lapsed (its proposal ended) or that is of no kind the teller knows is dropped with a line,
never asked. The view's answer only fills an answer given there; one it cannot read is said and the ask is raised all
the same (it never held the raise back; its line said so wrongly).
Builds: design 47 G A-424
The ask of proposal p-20261115t143855-09a7bc was written and the assigner's journal said nothing of it, so where it
stopped was found only in the teller's. The assigner now says each proposal it asks, with the ask's id and key.
Builds: design 47 G A-420 (a path that says nothing)
The ask of proposal p-20261115t143855-09a7bc never reached the operator: the teller kept it waiting for room, since two
old asks of the assigner (mailu's restore and hold, open since 2026-11-14) filled the two places it allowed, one being
kept free for the merge ask, which went with the Approve per merge (design 47 G §653). The assigner's are the only asks
the teller makes at the router now, so they take every place it gives (occ.MostOpen).
Builds: design 47 G §653
The candidate placing mesh-delivery-pipeline 0.1.6 on novox carried no parts, though its version-released was heard:
the assigner's module value of a catalog entry dropped the entry's inputs, so every trunk entry had fp-src 0 and A-419
never matched; and a release's parts were keyed <module>:<artifact> by prefixing the module to the artifact the event
names, which the contract gives whole (mesh-delivery-pipeline:program), so every release's parts read
mesh-delivery-pipeline:mesh-delivery-pipeline:program and A-421's match by digests missed too. The value now carries
the entry's src and fp-src, and a release's part is keyed by its reference as named.
Builds: design 47 G A-419, A-421, A-96
The two package-published of mesh-delivery-pipeline 0.1.6 sit on SEAT_mesh-builds_package-published, where the
indexer's program durable novox_mesh-delivery-indexer holds them pending and was never pulled: the program bound only
its durable on EVENTS, and a kept seat event is never there. So no version was released. The same holds for
toolchain-agreed (2 pending) and node-build-go's step-ended (10 pending). The program now binds, once its EVENTS
durable is bound, its durable on each kept stream whose event it handles, as the builder does.
Builds: design 47 G A-397, §349
A package-published of mesh-delivery-pipeline 0.1.6 left the catalog's versions empty and the indexer's journal
silent: the hearer said nothing when it heard one, waited for the module's other artifacts, took it as another
branch's, found the version released already or released it. Each of those ends is now said, naming the artifact,
the version, the branch and source, the artifacts still awaited and the fingerprint of the entry compared.
Builds: design 47 G §349, §371, A-420
The builder leaves an event it cannot take for the bus to give again, and the bus drops it after the fifth delivery its
durable allows without a word: the 0.1.6 published was lost so. On that last delivery the builder now says it gives the
message up, naming its subject, what it carries (a registry outcome's name and version, a version-released's module and
version) and the last refusal.
Builds: design 47 G A-420
An entry whose fingerprint moved with no merge names no fp-src (ADR 0601 §1), so A-419 found no release for it and its
wants and candidates carried no parts. Such an entry now takes the release whose parts' digests are the digests its
fingerprint's build.ended made; one found by neither way is said, naming which was tried.
Builds: design 47 G A-421, A-419
The indexer leaves an event it cannot take for the bus to give again, and the bus drops it after the fifth delivery
its durable allows without either saying so: a package-published lost so leaves a version unreleased silently. On
that last delivery the indexer now says it gives the message up, naming its subject, what it carries (for a
package-published the artifact and version) and the last refusal.
Builds: design 47 G A-420
namedBuild matched a trunk entry's version-released by the kept assignment graph's source revision, so after any merge
that does not rebuild a module its wants and candidates carry no version and no parts. It now matches by the entry's
fp-src, the source revision its fingerprint was built at, which is version-released.source for that build; an entry
whose release is not heard, or that names no fp-src, is said once, naming the module, the fingerprint and the source.
Builds: design 47 G A-419, A-96, A-67
oci-artifact-registry's published of mesh-delivery-pipeline/program@0.1.6 was refused as 'not numbered by the builder'
and left for the bus, up to its consumer's five deliveries, and no package-published was said: the version is the
reservation's (§369), never an entry of the state versions, design 51's numbering by name, which G keeps only until
it is removed (§372). Such an outcome is now taken: the reservations mark it and say package-published (A-415). The
log of an outcome left for the bus names its delivery count.
Builds: design 47 G §369, §372, §52, A-415
distribution's tag of mesh-delivery-pipeline 0.1.6 was refused: reserved answered 'no entry carries the name
mesh-delivery-pipeline/program', since the lookup matched only an artifact's package name and a bundle's entry
names none. A bundle's or an archive's name is <module>/<artifact>, read from its reference through NameOf; a
name no entry carries is said in the builder's log with the entries looked at.
Builds: design 47 G A-418, §478, §572
The registry refused builder_publish_again of mesh-delivery-pipeline 0.1.6 with 'transfer ... is no transfer id':
the name carried the version's dots, which mesh-sdk's parts receiver refuses. The transfer stays the builder's own
name (the registry part's publish), with the refused characters made dashes.
Builds: design 47 G §588, A-73
builder_publish_again for mesh-delivery-pipeline 0.1.6 queued nothing after a restart: the holder was learnt only
from node-engine running events heard since the builder started. The Republisher now reads each registry seat's
holder from the catalog's SeatHolders, the holder function the builder's FollowSeat of the assignments takes;
publish again and a node's returned queue at it. A running still queues the versions waiting on its module.
A seat with no holder is logged by name.
Builds: design 47 G §588, A-416, A-416a
A trunk build's registry publishes during the build, before the builder records the digests on the reservation;
such a published was acknowledged and lost, so package-published was never said. It is now kept on the reservation
(name, version, digest) and marks its artifact when the digests are recorded; one that never matches is logged.
Builds: design 47 G §588, §52, §334, A-73, A-415
The seat event's own stream is heard from its first message through the SDK's HearSeatEvents with FromFirst; nothing of EVENTS is read and no new grant is needed.
Builds: design 47 G A-411 (replaces A-408's read of EVENTS)
The builder kept the catalog's versions-filled in memory only and its durable
consumer took the event once, so a restart left every build wait-held. It now
reads the last message of the subject from the events stream at its start (a
read that fails is retried with backoff in the background).
Builds: design 47 G A-408, §369, §372
Live on novox 2026-11-14 15:20 the operator channel refused the seed's ask: its lines named paths (modules/<m>/module.json),
which the messenger's check refuses as a file path or a random-looking run. Checked against the messenger's Check: the
seed's whole words pass.
Live on novox 2026-11-14 15:01 the operator channel refused the seed's ask for a 64-hex digest in its words, as a secret;
the Approve binds the whole digest through the option. The tool runner reads the catalog half for the indexer's own tools
on its account and was refused every minute.
Live on novox the Go and Node build seats refused every seed steps call with 'the request names a module directory and no path': the trunk entries name no paths, and the seed sent the directory alone.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Live on shanks the build runtimes' seat list and catalog-half reads were refused on their caller-named subjects, since the modules' manifests do not invoke them, and the no-caller fallback is answered by nobody.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Live on novox the tool runner's first view of catalog-revisions was refused 54 times: the revision verb answered only the parts declaring it in invokes, and the runner reads it for every module it serves.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Holders subscribe and callers publish on mesh.seat.<seat>.tool.<verb> (and its call form) beside the machine-named subject, so a verb asked of no machine is answered by any one holder in a queue group.
Builds: lead answer A-393 (design 47 G)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Later artifacts of a seat that did not answer stay unanswered without a call; one log line says so when the seat first goes unanswered. Per lead answer A-392.
Builds: design 47 G §704, lead answer A-392
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Covers busstate.go keeperConsumer for the seat definition subject catalog-revision.<branch>: restoring the old body fails the table test.
Builds: design 47 G (seat definition subjects with parts)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
mesh-catalog's seat.json gives catalog-revision the subject catalog-revision.<branch> (likewise source-moved.<branch>,
group-ended.<branch>.<group>, owed.<repository>.<number>). watchedEvents widened only a node seat the node-engine holds
(machineEvent), so the consumer and subscribe grant of a module consuming mesh-catalog.catalog-revision (node-tools,
builder, teller, pipeline) stayed on the bare subject and never took the branch-form event. One rule now
(watchedEventSubject, used by the grants and the bus module's consumer): each part is *. A held consumer gains the
filter through AddFilters (grants/consumers.go).
Builds: design 47 G (seat definition subjects with parts)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The bus refused the bare subject mesh.seat.mesh-catalog.event.catalog-revision: the seat definition
(mesh-delivery-contracts/seats/mesh-catalog/seat.json) gives the event the subject catalog-revision.<branch>, and the
holder is granted catalog-revision.*. The publish now uses meshcatalog.SubjectCatalogRevision(the event's branch); an event
with no branch is refused loudly, never given a bare subject.
Builds: design 47 G, the seat definition mesh-catalog (event catalog-revision, subject catalog-revision.<branch>)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
lead answer A-391: the start replay keeps the view once (measured: 50 min CPU per start).
applyBuildEnd reports whether it wrote; ApplyBuildEnd keeps the view only then;
ApplyBuildEndsAtStart applies every entry and keeps the view once if any wrote.
Builds: design 47 G lead answer A-391
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The verb revision's callers rule matched seat:mesh-catalog.revision by exact text only; the grants part reads a single * as every seat verb, so node-tools was refused. Match that reading; the holder stays off the list.
Builds: design 47 G, lead answer A-390
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Before its catalog half is read, a tool runner's gate lets through the
verbs its held definition classes read; these five read only and are
now classed so.
Builds: design 47 G lead answers A-387
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
The merged message and the catch-up asked the seat's groups with all=true, whose answer outgrew one bus message
(2888983 bytes, live 02:06), so no take-in answered. Read asks the open groups, MergedOf the one group its branch
names (group=<id>); the list a person asks (bridge_groups) keeps ReadAll.
Builds: lead: principle 7, live 02:06
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
lead: principle 7, live hold 02:0x (nats_users for the whole list was 1086642 bytes, over 1048576).
Builds: design 47 G, lead's handover (no design line beyond principle 7)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Builds: design 47 G A-372 (G-22-lead-answers-2.md, hq design/47-G-lead-answers-31)
Also: Ready reads what the first views hold, since a first view closes before the follower's change is taken into the branches followed; Branches.FirstViews says when both are in.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h