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
A revision changing the indexer writes only what the named build needs and is
taken in when that build starts; a gathered group stays across the restart and
its end makes one take-in.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
The consumer tells a group's end apart by its name before the forge decode,
a merged message goes through the pending state, and the pending revisions
naming this build are taken in before the consumer starts; take-in events go
on the indexer's own stream. Until the pending state is on the bus every
merge is taken in at once, as before.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
A merged message is kept in the indexer's own state pending and the merges
there are computed together on overlays of the indexer's states: taken in
whole when the indexer's fingerprint is the running build, otherwise only
the indexer's own entry, graph key and verdict are written and the rest
waits for the named indexer. A group's members are gathered until its end.
Waits are said on the stream take-in, read by indexer_pending and steered
by indexer_take_in_pending and indexer_drop_pending.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
mesh/check checking 70ade15c36eb on dd90a26251b9 | novox/mesh-delivery#149 in group feat/l3-final
The indexer judged only the manifests its modules state holds, so a digest refused by an older judge (node-tools
48c729f6 and mesh-provisioner-client a382b9df, refused for words then unknown) kept that refusal once it was back on
the trunk or in a candidate. Each such digest is now judged alone, in the catalogue the trunk holds, writing its
verdict and nothing else; an accepted verdict stays accepted under any later judge (ADR 0579 §1), so only a refused or
waiting verdict by another judge, or none, is judged again.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
dropRemoved hands the module over from the other trunk, read from the forge at the head the source names, so no wait in memory is needed. indexer_take_in with a repository takes in a head with no module.json as a merge that leaves none, and answers by name for a head with modules.
The check of 2026-11-02 22:35 on a loaded shanks failed two buildrun tests on wall-clock bounds of 60 ms, the
second failure of this kind tonight (the first: mesh-sdk's busretry tests, f02f142).
Quick fix: give the two buildrun asker tests bounds of seconds.
Thorough fix (chosen, because the kind recurs across every module of the repository): every test that raced a
wall clock now waits on an event, steps a clock it is given, or has a bound only a failing test reaches.
- buildrun asker tests: the steps' own times stand for the clock (bound an hour, starts and phases written in the
past), no goroutine paces phases with sleeps.
- builder: the decline bound reads Builder.Now; the decline test steps it; the stream-wait test waits for five
check reads instead of 100 ms.
- checker: the lost-ask try bounds an untaken ask at 200 ms only where that is what ends it; the program says when a
merge turn changes, and the integration test waits on that line instead of 300 ms.
- teller: the bed makes the teller's own ends bucket (a 10 s stall at start the tests raced by 0.5 s), waits for its
warrant subscription instead of 500 ms pauses, and the package-level now is restored with t.Cleanup.
- watchman: the recount, run in its own goroutine, is waited for instead of 10 ms.
- polls of 2-5 s become 10 s (unit) and 20 s (integration).
Under a throttled cgroup (half a core shared with two busy loops) beside the node-engine's suite, the buildrun
tests failed 5 of 20 runs before and passed 30 of 30 after; no test added.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Review finding 2 (ADR 0593 §2, §3): after the release no build has reported imports, and a module that
replaces a mesh Go module of its own repository by a directory had no go.sum line and no version for it,
so a change of buildstep or buildrun alone moved no importer. The indexer now reads go.mod's directory
replaces, publishes those modules, and puts their content digest on the version line; builds are pinned
to the other repositories' versions alone.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0593 §7: compose and frombus left mesh-sdk for the node-engine, so how a composer holds a refused
module is the engine's to test; the indexer's test keeps proving the verdicts and refusals it writes.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
mesh-delivery is over its budget (810 unit, 73 integration on main), so this
branch adds none net: its new cases are subtests, the three work-queue process
tests and the consumes test become one, the history stream bounds join the
stream values table, and the go/-only mesh-sdk merge check joins the one new
integration test. buildrun's 20 unit and 1 integration tests moved in from
mesh-sdk with it.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The indexer whole: two catalogue modules report their imports at their builds'
end, mesh-sdk merges a change of buildstep alone, the packages key names the
merge with a digest per package, the buildstep importer moves and the other
keeps its fingerprint though the version line would have moved it (ADR 0593
§2-§4; D11).
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
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
The fingerprint, named tree and package versions move from mesh-sdk into the
indexer's internal/fingerprint (ADR 0593 §2). The indexer publishes each reached
mesh Go module's package digests from the forge's listing in its new packages
state, keeps the imports a done build reports, and writes the import form only
when an input differs from the build that reported them (§4), so the form change
costs no rebuild and a package merge moves only its importers (§3). Candidates
take the same form against the group's member heads (D1-D4).
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
mesh/check checking edf7b3c0bf3a on b1f262d7d06e | novox/mesh-delivery#129 in group feat/judge-once
Issue 1585, R1: fresh holds the digests a take-in newly wrote (a new module, a changed manifest); a registered or
taken-in module is judged too; a start pass that fails is judged again at the next message heard, no timer.
R2: a module's verdict, problems and warnings are carried while its digest stands and are no part of the compare.
R3: the checker judges the trunk beside the change and fails on every problem the change case adds, an untouched
manifest's included. R4: the judge reads the indexer's kept predefined keys, or the catalogue member's when it
changes them. A log line names a package no held module publishes. Every module pins mesh-sdk 62cddf4.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Issue 1585: the indexer keeps an accepted manifest accepted when a stricter judge re-judges it, its problems as
warnings carried in its value; a merge's new manifests are judged in full. The watchman raises manifest-outgrown as
a warning until mended. The checker fails only on a manifest the change adds or changes and names the rest as
warnings. Every module pins mesh-sdk 19c2042.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
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
Issue 1585, S5-S8: the indexer's fingerprint reads go through judge.ForInputs, so judge/ is the one place in
mesh-delivery that calls mesh-sdk's rules; the command's request judging moves into judge.Request, tested in the
judge's table; two predefined tests become one.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Issue 1585: the indexer's merge test takes in three manifests more against the real bus; it writes their
verdicts, says each refusal, and mesh-sdk's bus reader composes the valid one and holds the refused, named; before
any verdict it takes each as accepted and names the field set aside. One test extended, no test added.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Issue 1585, S9: verdicts keyed by digest, written only when they differ; a module's value carries a refusal; a
new refusal is said as manifest-refused. Judged at start (a new judge judges again) and after every merge.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Issue 1585, S5: judge.Shelf judges each manifest alone and the rules between them, giving a problem between
manifests to each module whose removal ends it (naming alone blamed the module mounts for f1-circuits' problem).
cmd/mesh-judge answers a composer request so a Go build module runs it as it runs the composer.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The holder writes a node's changed module keys, then its node key as the commit by compare-and-set at
the revision the proposal was laid over, then the placement.<node> mirror for readers not yet delivered.
Every reader assembles through mesh-sdk's PlacementKeys, which reads both forms.
Deletes the placement and candidate writes, the reread tool and its forge reads, the two buckets and the placement-changed event.
D1a and D1b are one commit: the index package, the named reads, the tools and the manifest do not compile apart.
Asking whether the merge contains the held commit refused forward merges
forever on a diverged or rewritten main or a squash-merged registered head,
failed every later merge when the forge no longer had the held commit, and
cost one forge call per held commit. The question is now the reverse: a merge
is kept back only when it is in the taken merge and differs from it. A merged
message asks it once, before it writes anything, and a late message writes no
module, deletion, placement or taken key. A take-in asks it once for its
module. A 404 is not known to be older and is written, and said; any other
failure to say leaves the message for redelivery.
The indexer wrote mesh-catalog's merge 9cdac09a back over the later 69056948,
rolling kernel and mesh-build-image back on the nodes: a merged message's
commit was written whatever the held one was. Before writing or deleting a
module's value, the indexer now asks the forge whether the merge contains the
held commit, keeps the held value when it does not and says so, and leaves the
message for redelivery when the forge cannot say. Each applied merged message
is said in one line with the modules written.
No tool showed what the indexer heard of FORGE nor a modules key's history
with times, so the cause could not be proven: the indexer now records each
FORGE message it handles (sequence, deliveries, times, commit, outcome) in its
state heard, read by indexer_heard, and indexer_module_history reads
modules.<module>'s kept values with their times.
Quick fix (Src order in writeOverlay) rejected: Src is 0 once the source moved
on, as here. Thorough fix chosen: the order is the forge's, on every path that
writes a module (merged message, take-in).
A merge whose merged message never reached the indexer had no recovery: take-in answered
held for mesh-vault, whose record stayed at 78d7ef7 after mesh-catalog#535.
A drained forge stream does not say a merge was applied: the forge bridge was still
retrying mesh-catalog#535's merged message when the indexer started, and the source's
head moved taken.mesh-catalog to 9fd9569, so the view and show said the merge was on the
machines while mesh-vault's record stood at 78d7ef7 (ADR 0556 clause 1).
jschoubben