Rows in the judge's table: two modules invoking each other, or a verb closing a cycle with a provision, are not
refused; a cycle of provisions and one of bases are. The refusal's words name what a part waits on now.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
What the running builds serve is read when a changed build module is met: the node-engine's haves, the builder's
build of each, its manifest. The modules it builds are composed from their manifests, as for any changed build
module, and the composition's warnings name the verb.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
A failed candidate build no longer needs mesh-delivery.recheck, an exception verb (ADR 0421 §12): the checker hears the
builder's keys the failed verdict names and, once each is written done (builder_build_again), retakes the check of the
same heads as it does after a failure of the nodes, at its start too. When the tests have already failed the check
composes at once and names the builds it did not wait on, so its red status is never held back by a build.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
Worked out from the builder's builds state at each read, stored nowhere: of each running check, the candidate's builds
that are neither done nor failed.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
The builder already builds every candidate fingerprint into its own key. The check now watches exactly those keys on the
bus and composes once each is done or failed, within the new setting build-wait-bound (1800 s). A failed build fails the
check, naming it; a build not ended within the bound is named and the check composes without it. A module built by a
build module the change alters (mesh-build-go or mesh-build-node), one standing on a base the change alters, and one the
trunk holds no record of are not waited on, and the composition's warnings name each with why.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
D13b: the composition's warnings, and so the check's line, name the changed modules built with the Go build module the change alters and those standing on a changed base. D9: builder and verdict rows. D15: the assigner's integration test gets a source revision changing a provider, its new consumer and a build module.
Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
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
mesh/check failed: mesh-build-go: the indexer's judge at the group's mesh-delivery head 3b3b096be8cdf8f49246ed3961ece3eb5c349d62 refuses the manifest …
The composer command moved to the node-engine (ADR 0593 §7); a catalogue check had only mesh-sdk beside it. Pin the build runner to the catalogue head whose composer test skips aloud until then.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Review finding 4: an unreadable placement gave no nodes, so nobody was asked and the check passed. The
placement read now answers its error, and the verdict says the composition was not made and why.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Review finding 3 (ADR 0593 §9): only a node that does not answer may let a check pass with a warning; a
sandbox composer that cannot be built or run was filed with the silent nodes. It now fails that node's
composition, naming the head.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0593 §1: the step runner left mesh-sdk for mesh-build-go's buildstep package, so the examiner's unit
tests and the module's integration test require it by the group's version (as the builder does) rather
than an imitation of its plan and kept-verdict rules.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0593 §7, §9: the checker asks compose-candidate of each placed node in parallel within its own setting
compose-answer-bound (60 s); a refusal fails the check naming node and module, a silent, unready or
unreachable node passes with "not composed on that node: no answer". A group with a node-engine member is
composed by the composer at that head on the Go build seat over each node's input. The wait on the
assigner's compositions goes, and the checker invokes the one engine verb it needs.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0593 §11: the controller's repository holds nothing that builds once route-proxy moves, so its
sibling rule goes; its tests fold into the sibling test, which now uses mesh-sdk's rule.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Review of 650a1d0: a member that only failed its tests got no preview, so the indexer held no manifest
of it, the assigner's composition only said it waits, and the check waited for ever. The preview is now
written whenever every build module's fetch merged; a member with no preview of a merge (a conflict, a
fetch that did not answer) makes the composition not run, said with the members it names, unawaited.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The operator, 2026-11-02: "A failed check shows only its first fault". Group feat/1126-removal's
check failed four times, each verdict naming one fault, so three faults took three runs of 20-40 minutes.
The checker no longer returns at a member's first failure: every member is asked, every level's
failures and every member's refused step or conflict are joined in judging order, the manifests are
judged and the composition is read also when the tests failed, and a failed composition is added after
them. The status line says how many failed and names the first, cut so "; and N more" stays whole.
checker_verdicts and checker_pull answer the whole list (failed). Needs mesh-sdk's buildstep change
(every level runs after a failure) pinned at the group's head.
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
mesh/check being carried to d26781b403df on 4dd7dcfc1411 | from be88162cde31
A group's check judges each member against the group's heads (ADR 0482 §2,
ADR 0573 §4), but the checker placed mesh-catalog at its trunk beside every
other member even when the group held a catalogue pull request, so
mesh-delivery's judge tests in group feat/1126-removal judged the new judge
against catalogue main and failed. The catalogue now follows the rule of
every other sibling: the member's head and ref in a group, the trunk alone.
A catalogue change's own trunk copy (ADR 0529 §2) is placed separately and
is unchanged.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The controller ended by its unassignment (ADR 0590) and mesh-sdk drops
its seat, the grants seat and placement's adopt and hand-back. The
indexer's judge still carried the controller's seat entry, the old
removal words and the two placement verbs, and counted 58 seats; the
checker still placed mesh-controller beside the node-engine's and the
catalogue's tests and failed a node-engine check without its trunk; the
assigner still asked a compose of the running controller, served adopt
and hand-back as stubs, and kept a policy-split finding that could no
longer arise.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
mesh/check pass d739516056dd on 55a506975b01 | unit 7056 run, 243 kept, 16 skipped | integration 7194 run, 275 kept, 25 skipped | over bound: unit, in…
A check ask named no toolchain, so a node on the minority version could run it and KeepPasses record the pass as run on the agreed one. The check ask now carries the agreed toolchain and a node off it declines (mesh-sdk), for the checker's decline loop to put again. Pins the builder, indexer and checker.
Known limit: the checker puts a declined check ask again at once, with no pause; the declining puller skips that kind until its next pull wait, which paces a lone minority node.
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 §4 makes refusing a member for its mesh-sdk pin a fault of the
check (issue 1585, S14). Every Go check ask now carries the latest
published version of each package, or the head of the group member that
publishes it, so the pin check, its forge reads and its tests go, and
the controller's tests read mesh-sdk at the version it is built with.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Issue 1585, S12: grants, builder and checker read with catalogue.Decode and grants holds a module the indexer
refused; the checker asks the Go build seat to run mesh-judge, built at mesh-delivery's trunk or at the group's
mesh-delivery head when it changes the judge, over the catalogue as the change leaves it, and fails on a refusal.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The forge carries the operator's own repositories (jschoubben/hal, jschoubben/novadock) beside the mesh's; the
indexer writes their pull requests as candidates, the gate never names their trunk, so the checker said each
"cannot be judged" and owed the status error on mesh/check, which the forge module wrote into repositories the
mesh's forge account may not write: 403, 119 times each (sweep 2026-10-30 item 7).
Quick fix: name the two repositories as not the mesh's in the checker.
Thorough fix, chosen because any repository on the forge can meet it: a candidate every pull request of which is
of a repository the gate's source.main names no trunk of and touches no module is outside the mesh. The checker
writes it no verdict and owes it no status, takes away a status owed before, says it once in the log, and lists it
in checker_status as outside_the_mesh. A new mesh repository's first pull request adds a module, so it is still
said untrunked and owed error (issue 1422). Nothing is outside before the gate's source is read.
mesh/check pass 2d8ad2e66a8f on 38f89c6976c9 | carried from 1f6f4eda3208 | unit 470 run, 4760 kept, 13 skipped | integration 503 run, 4844 kept, 22 sk…
Once the gate's one-hour ask bound goes (mesh-catalog fix/1733-no-hour), the
gate asks the operator's Approve only on a verdict, so every path that left
a candidate without one now writes one, from an event and with no timer:
- a group with a member whose repository's trunk head is not known (a pull
request the checker cannot claim) is not judged in part: its verdict fails
naming that repository, and every status of the group says so (a pass of
the others would have let it merge with that member never checked);
- the verdict of heads that cannot be judged replaces whatever the key held,
such as a running verdict left by a check that can no longer run;
- a verdict key deleted is judged again on its delete (the checker watches its
verdicts), and a check running when its key is deleted writes it anew.
Tests: three rows at the end of the no-trunk test; counts unchanged.
mesh/check checking e7888024a826 on 769815ac0fea | novox/mesh-delivery#103 in group fix/lost-check
Re-review of fix/lost-check: writing an empty list at the end kept one key
for every verdict key ever checked. Tests: every row of the lost-ask table
proves the key gone at the end; one row more (deaf after its build module
replaced it, asked again by hand); counts unchanged.
Review of fix/lost-check: an ask no node took yet has no step key, so
asking again then would run the check twice (issue 1791's kind). The
checker keeps the asks each running check put in a new state asks, and the
tool refuses unless every one of them has a step key. Asking again under the
same ask id is not done: the stream drops a second put of an id within its
de-duplication window and keeps both past it, so the waiting copy is never
replaced.
Tests: the hand row of the lost-ask table refuses while its ask waits;
counts unchanged.
The checker's HeardEnds moved to mesh-sdk buildrun (fix/lost-check,
6aeb401): both modules pin it, and the builder, which waits on the same
Asker, now reads an ask's step at each step end of its build module too.
checker_ask_again and checker_status's lost stay in the checker.
Tests: one row in the builder's ask table (its watch deaf, read at a
nudge, done with no new ask); counts unchanged.
mesh-catalog #553's verdict stayed running after its step mvsgthyjed5786
passed on shanks at 19:09:31Z: the watch of that one step key, opened at
18:46:05Z, heard neither its start nor its end and never closed, so the
wait had no bound running and nothing would ever judge it.
- HeardEnds stands between the asker and each ask's watch: at every step
end of the same build module (one watch per steps state), at a
reconnect, the asks waited on read their keys now and are judged; a
watch that closes is opened again, at most 3 times.
- A step the build module's next process ended (RestartReason, a delivery
replacing it) counts as a lost ask: asked again under the claim, at most
LostAskTries, then failed as not heard, judged again when the nodes change.
- checker_ask_again {group, why} ends a running verdict as asked again by
hand; the checker hears its verdicts, ends the wait and claims it anew.
Refused while a step of its pull requests runs, or for a verdict that
is not running.
- checker_status names such checks under lost, and each step's what.
Tests: one table replaces the given-up test (rows for given up, replaced,
deaf and nudged, deaf and asked by hand); counts unchanged (147 unit, 10
integration in the checker).
A repository never merged through the gate has no entry in source.main, so its pull requests were dropped without a log
line, a verdict or a status, and checker_status counted them as new. Now the verdict says it cannot be judged and why,
the forge status is error with those words, checker_status names it under cannot_be_judged, the log says it once per
head, and the heads are judged whole once the trunk head is known. Quick fix: this. Thorough fix (the gate seeding
source.main from the pull message's base head) is still to do: it needs mesh-sdk, the forge bridge and the gate.
Only tries a build node holds count toward the 4 at once; tries waiting for a node are bounded
apart at 8, and the refusal names both counts (issue 1890: at 19:53 four tries queued behind 44
checks for up to 47 minutes refused every further try).
The checker wrote a try's asks only at its end, so a 31-minute try could not be told queued from
running (issue 1890). The try's record now names each ask as it is put, then the node that took
it, its state and the phase its step says, written on the step key's own changes. build_usage
names a running try's asks and takes a build module's capacity.<node> (configured bounded by its
node's build CPU, issue 1806's follow-up) when it is current with mesh-config.
A failed test with no cause, or a passing line as its cause, was worded as if it were the failure: a try of mesh-catalog
failed with a passing '✔' line of the first test file. Under mesh-sdk's failure contract the verdict's words and its
status line now start with 'the build module hid the failure (<module> at <version>)', and the check still fails.
jschoubben
mesh-tools