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
Design 47 *A module's registration*, "No part fixes a name" (b229f5c4 lines 482-485), and *What ADR 0656
leaves to this design*, "The indexer reads it [acts] into the catalog graph" (line 1583): ToolchainModules,
ToolchainStreams, BuildGoModule and BuildNodeModule go. The build module of a language is the module whose
manifest claims the seat whose acting rule is step build on it (index.BuildModulesOf), its stream the one of
its manifest carrying toolchain.reported, its bucket <holder>_toolchains as before. Found again at start, at
every change of the seat list and at every catalog revision; the watch, the gate and the hearing follow.
A language with no build module found reads as no toolchain reported.
Rows: TestAgreedToolchainChangesAGoModulesReasonWithNoMerge/the build module of a language is the holder...
(fails without: no BuildModulesOf, and go-builds' report refused); TestToolchainWatch's later build module.
Left: six integration tests of cmd/mesh-delivery-indexer hold no build seat with acts nor a holder manifest
and fail (TestIndexerWhole, TestIndexerWritesAGoModuleAgain..., TestIndexerGivesACandidateItsFingerprint...,
TestIndexerJudgesAnUnreadableVerdictAgainAtStart, TestIndexerCountsOnlyThePlacedBuildNodesToolchains,
TestAMeshSDKMergeMovesOnlyTheModulesImportingAChangedPackage): their fixtures are to be given both.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Builds: design 47 G §364, §428, §535
graph.NodeOf moves from the indexer's own package to sdk/graph, since the assigner reads its verb edges from the catalog
half too (a second reader).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Builds: design 47 G §47 to §50, §121, §122, §378
ADR 0663: the catalog half is read through revision, answered by the
tools bundle from the catalog graph revisions before it asks anything,
to the parts whose entry in the newest trunk revision declares
seat:mesh-catalog.revision in invokes, the holder refused too; none kept,
every reader waits. The first view modules, its callers list and its
tool leave the manifest; the assigner declares revision instead (its
code moves in W5.10); the grants part keeps its read of modules until W8.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The node agent calls a restored part's catch-up tool with no arguments, so
indexer_take_in with neither module nor repository takes in the head the
gate's source names for every repository, as the first start's catch-up
does, and keeps the view as the next revision when an entry moved. The
manifest names it beside the catalog graph revisions, form 1 (design 47,
Reverted).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The tools bundle reads the objects catalog-revisions through its runtime (index.TrunkStates over busstate): the first
views, read tools and seat verbs of modules and graph, indexer_config_readers, indexer_bus_state and
indexer_package_users; the marks of modules and graph are the revision's number, and a view asked before the first
revision says view-waits, never an empty catalog.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
main.go reads the view and the module graph's nodes (NodeOf) from the newest trunk revision instead of the modules and
graph states; every write outside a recompute (the judge pass's verdicts, a build's imports, the source stamp, the
modules the trunk dropped, the indexer's own entry a held revision writes) ends in KeepView, so a restart loses nothing;
a revision lists every module the view holds, one whose manifest cannot be read too. At a first start whose objects keep
no trunk revision the indexer takes in the head the gate's source names for every repository by the forge bridge's
merged message and keeps revision 1 (the lead's ruling A). The whole-module tests read the view from the revisions.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The lead's ruling on finding 1: modules becomes an in-memory view of the newest
trunk revision read from catalog-revisions at start, one form of the fact
(ADR 0656 §6). TrunkView gives that view as a store in memory and the revision
it was read at; an empty bucket is an empty view. Not yet wired in main.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The lead's ruling on finding 1: an entry gains inputs (what its fingerprint is
computed from, the toolchains used, problems, warnings, named paths, the source
revisions it was taken in and its fingerprint moved at) and packages becomes
name to version, so the indexer's view of the trunk can be read back from the
newest trunk revision alone (ModuleValueOf). A revision's entries are the values
the recompute wrote, not the values it started from.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Design 47 (Two graphs) keeps a catalog graph revision for each source revision taken in and for a fingerprint moved
with no merge. The recompute that writes the module graph now keeps the revision it makes, numbered after the newest
kept trunk revision, and says the seat mesh-catalog's event catalog-revision naming the changed modules; a recompute
that moves no entry keeps none, and objects not yet on the bus keep none, said in the log.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Design 47 (Two graphs) keeps each revision as the list of its entries' digests and each entry as one object, so an
unchanged entry is stored once. The bucket is declared in the old objects section with a 512 MiB cap (ADR 0619 §2:
the indexer is the foundation), opened at its first use; a revision's list is written only after its entries, and a
revision read back with an entry lost is an error naming the module.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The entry is the seat mesh-catalog's CatalogEntry filled from the module graph's
node a recompute computes (build edges, provisions, verbs, the edges left out on a
cycle) with where the trunk holds the module and its manifest's digest; a revision
lists one entry per module by the sha256 of its bytes, so an entry two revisions
share is one object (design 47, an unchanged entry is stored once). Rows in the
graph test: one entry per module, and only a moved entry's digest changes. Kept
as objects, numbered and said with W4.2's store and W4.3.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The catch-up of the heads the source names runs before a source-moved is read, once per branch; the gate's source bucket is no longer read.
internal/index compiles against the new contracts again (SourceByMainValue is gone); tests of earlier subtasks that
the broken compile hid are fixed (the validator's MAJOR.MINOR version in their manifests).
Builds: design 47 G §15, §115, §362
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The refusal of a status in a manifest (IX20) relies on strict decoding: a manifest
field naming a status is an unknown field and is refused. The `*` emitter of
consumes (design 47 G §386) is held for the design writer's answer on model-usage's
`*.usage.*`.
Builds: design 47 G §89, §368, §387
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/check failed (3): mesh-build-go: the go.mod of novox/mesh-sdk at 396511b7a276, beside the change, cann…; and 2 more | 00232c71977b on 9c6bf236b6b1
Every recompute wrote every key of its modules state again with identical bytes: recomputeLocked
compared the package versions with reflect.DeepEqual across two map types (the contract's
map[string]string and sdkstep.PackageVersions), which never match. Each write was a views
event, and a reader of the views took each as a gap and asked the whole first view again.
The comparison is by maps.Equal now, and write itself skips an answer whose bytes are the
bytes held, for every caller and every state it writes; the merge's written list counts only
real writes.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
The operator, 2026-11-04: logic one module owns lives inside that module as a package of its own, so the take-in step
changes no mesh-sdk. TakeIn, its keys and event names, PendingWait and PendingAnswer move to
modules/mesh-delivery-indexer/takein (its merges a takein.Merge of its own); Forge whole, with its group fields,
ForgeGroupEnd and GroupMerge, moves to modules/mesh-delivery-forge-bridge/forgemsg. Both are nested Go modules that their
readers take by replace, as buildrun is. Every importer moves in this commit: the indexer, the bridge, the watchman and
the view. Forge's private checks (the repository name, the version) are copied small; mesh-sdk main still holds an unused
Forge until its own fold removes it.
The indexer, bridge, watchman, view and checker pin mesh-sdk 5a38a12 (origin/main), no feat/take-in commit.
The bridge's integration bus now grants the state end-asks that bridge_group_end_again added: without it every test of
the package failed on a permissions violation. Test budget: two new test functions (TestForge, TestTakeInEvents) and the
two the join added are paid by merging four pairs in internal/bridge; the repository stays at 806 unit tests.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
The assigner reads which edges are left out from the graph, so one place decides the cycle; the whole graph is
judged, a module past the graph's cap included, before the first GraphMost are written.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
ADR 0573 §2: a module builds against the latest published version of
each mesh package it uses, and a package's trunk merge moves every
dependent (issue 1585, S13). The bridge reads the packages each go.sum
names, the indexer keeps them and takes each one's trunk merge into the
fingerprint, a candidate takes a group member's head instead, and the
builder's ask carries the versions so the build module writes them.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The operator, 2026-10-25: only if a module has a new version in commit x is a new value written naming commit x. A merge
wrote every module of the repository at the new commit, fingerprint unchanged (about 186 keys per mesh-catalog merge), and
the controller's take-in took each in again: twice tonight it flooded. Now a merge that gives a module the value it holds
but for the commit writes nothing, and the value keeps the commit its fingerprint first appeared at. The kind: a write
that differs only in where it was computed. The other instance, a source revision re-stamping Src on every module at its
repository's head after any merge anywhere, now stamps Src once. The trunk head stays the gate's source.main.
mesh/check failed: integration: mesh-delivery-indexer/cmd/mesh-delivery-indexer.TestIndexerCountsOnlyTheAssignedBuildNodesToolchains: within 5 s of n2…
On 2026-10-25 20:56 assigning mesh-build-node to ace again brought ace's report of 2026-10-23 (v26.12.0) back into the
agreement before the module ran there; the indexer blanked every module's fingerprint to "builders disagree" and the
controller's take-in refused modules already computed and built.
- The toolchain gate (indexer, builder, checker; still one file in three) leaves a report out, said once, until its node
writes it since its current assignment: the assignment's own time, or the revision that opened it.
- The indexer keeps a module's held fingerprint while a language has no agreed toolchain, when its kept inputs with the
toolchains it was computed with give it again; only a new fingerprint (a merge, a base, an agreed toolchain moved)
needs the agreement.
mesh/merge-gate head 61fea8314fe8 merged into trunk 318064ce206d — pass: builds mesh-delivery-indexer → novox; no bus step; every machine composes with the…
mesh/merge-gate head a5da6079de01 merged into trunk 340bf643883d — pass: group feat/r4c3-build-modules composed: every machine composes with the change as …
mesh/delivery covered after its failure: a later walk that contains it was delivered: plan-1792582266656254343 (novox/mesh-delivery at d68d7ef3 into main…