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
The builder numbers, asks the floor of, records and tags a package under the name the indexer's series key names for
its <module>/<artifact> (a Go module path, an npm package name; design 32), so builder_versions answers the holder's
question by the name a registry seat publishes it under. A package the state names nothing for, or a state not readable,
is built unnumbered and said; the rest of the claim is numbered.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The builder's forms gain seat-reference for bundles and archives (oci-artifact-registry/<module>/<artifact>@<digest>);
the old-form window writes the older artifact-store reference beside it (catalogue.InArtifactStore), the oldest kept
in the single reference (ADR 0624 §1, §3), so mesh-build-go's seat-form answer no longer strands an engine stating nothing.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The builder reads out-of-step from each lane's step result, a failed step's too, and writes it on the build's end
entry (BuildLogEntry.OutOfStep, J3's mesh-sdk field), where the indexer raises generated-types-out-of-step for a trunk
(the lead's ruling on A2i's finding 2). BuildAnswer names no such field yet, so it is read by name and shape.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
Design 51 says the build log tells a version kept and why, a tag ask unanswered after five minutes,
and the old form's window closing; the process log alone was read by nobody. Each numbering decision
is now a build.said of its module and fingerprint, its why one of numbered, kept, waiting, unnumbered;
the tag ask and the window, no one build's, are said under the builder's own key, the ask's or the
window's key standing for the fingerprint. Needs mesh-sdk 7efb173 (build.said). Each failed first.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
A check's build of the ask's head writes the ask numbered in tag-asks by compare-and-set. An unanswered ask has a key in the
new key-ttl state tag-ask-bounds (five minutes); the bus's marker of its end, heard on the watch, narrates the ask with what
it waits on (the check's build not claimed, names waiting on their floor) and writes the bound again; no timer (design 51
*Tagged versions*). A restart arms each unanswered ask whose bound ended meanwhile.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The builder reads tag-asks for the candidate's repository and head (versions.TagFor), numbers only its packages under
the tag below the next version, keeps each with its repository for the branch's end, and says numbered with the head;
the same fingerprint claimed again keeps its number (design 51 *Tagged versions*).
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
The builder hears the forge bridge's stream on a durable consumer: merged, a
pull's closed and branch-deleted ask remove of each tagged version still
held, and the bus gives the message again each minute until all are removed;
past 15 minutes its finding tag-removal-not-done names the seats. The builder
gains the state findings (ADR 0564 §2). Every forge message also wakes a floor
whose bridge_tags was silent.
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's one consumer of EVENTS now consumes the indexer's manifest-accepted and
the condition keeper's condition-cleared; a heard source revision stands for the
forge's merge. Numberer.Heard marks a version published, refused or removed and wakes
a floor whose seat was silent; its consumer of each <seat>.events waits on the seat
family's stream form (asked of A1).
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
A package whose floor is not known waits and nothing is claimed; an image, bundle or
archive is built unnumbered. The state versions is bound when the bus holds it, every
new number is said as version.numbered, and the manifest invokes the floor's three
sources (design 51 The versions, The floor; ADR 0606, ADR 0617 §1).
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The floor asks the seat's versions, bridge_tags and indexer_requirements; a silent source is named in the
floor finding and woken by an event of that source (the holder's published, the forge bridge's next message, the
indexer's next manifest-accepted) or its module's condition clearing, or by hand (builder_floor_again), never by
a timer (design 51 *The floor*, ADR 0617 §1). The old form's window (ADR 0624 §1-§3, ADR 0618 §2): each counted
engine gets the newest form it reads among those written, the new form only once every present engine states
it, the oldest kept in the single reference; a node not serving manifest-reads counts as the baseline; a node
dropped by hand counts no more; old-form-closed is said once, and a node holding it seven days is a finding.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
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
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
The named tree is the indexer's computation (ADR 0593 §2): the builder reads it
from the module's value, so its test gives one as the indexer would write it.
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 build runner leaves mesh-sdk for mesh-build-go (ADR 0593 §1), another
repository the builder does not import; the tests answer asks as the builder
needs them answered.
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
ADR 0590 supersedes the lease write: mesh-grants says no lease event any
more, so the watchman's lease-altered and removal-failed findings and
their row had nothing to hear. The ended controller's consumers are
removed as hand acts, so their deletion is no consumer lost; only the
one on DEAD_LETTER_NOTICES is still read, since its removal hands the
notices to the watchman. The builder hears artifacts-gone from the
artifact sweep seat alone, and mesh-enrolment's tokens no longer report
an issuing gate that nothing can turn off.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
A build node off the agreed toolchain declines the ask and, idle, takes the ask put again at once, so three declines used the three askings and failed the build. A decline now uses no asking: the ask is put again after a short pause, and past a bound from the first decline the key fails naming the declining nodes and their reasons. A person reads which node sits out in the key's step text, which the view's builder_builds answers as stored.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
One build node on another version refused every build of the language. The builder and the checker now follow mesh-sdk's majority agreement; the key's step names a node that sits out ("ace sits out: typescript 26.12.1, agreed 26.13.0"), and a build ask a minority node declined is put again for a node on the agreed version. Check steps compare no toolchain, so clause 2 already holds for them. Pins builder, indexer and checker to mesh-sdk 71be15f.
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
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).
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.
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.
The waiter moved to mesh-sdk buildrun (fix/734-lost-ask, d96b9fd6), where every asker of the build seats shares it.
The builder hears DEAD_LETTERS for the go, node and image seats and asks a given-up part of a build again at once, with
the seat's words as why, instead of waiting for the check bound. The checker keeps what it does with a lost ask (a try
ends as error, a check is asked again) and its verdict words.
mesh/check pass 3e8bd9b0def2 on bcb5aafdbe80 | unit 168 run, 472 kept | integration 178 run, 538 kept | composition passed on shanks
mesh-delivery #51's re-ask never fired for mesh-build-node.4e7b118e: messages the
SEAT_NODE_BUILD_GO workers had given up on stay on the work queue, and each counted
as waiting ahead, so the ask was never asked again (note 62). A message ahead now
waits only when no node wrote its step and no ask of its kind asked after it was
taken; the old message of an ask asked again is deleted from its stream; and
builder_seat_queue (answered by the builder through its queues state, as only its
account may read the stream) shows each message and whether it still waits.
builder_stuck, checker_kept, checker_verdicts, checker_status, checker_previews, checker_pull and checker_try listed a
bucket of thousands of keys and read each value (builder_stuck also its history) one round trip at a time before
filtering, so they timed out at 30 s. Now keys are narrowed by what the key carries first, history is read only for rows
kept, values are read sixteen at a time, at most 1000 a call with the answer saying it was cut, steps newest first, a new
try reads only the newest 64 tries, and each tool logs the values read and the time taken. Same kind as the view's fix
(PR 52); the helper is copied, not put in mesh-sdk, as a few lines do not earn a group delivery.
A build ask its seat's stream gave up on, dropped or ended with no step was
waited for with no end, and on restart taken over and waited for again: no
build node would ever answer it, and nothing could ask it again (trunk key
of mesh-build-node at 4e7b118e, building since 19:06; seven keys at 18:44).
At the step bound, a context's bound and no timer of its own, the builder
reads what the bus says of an ask with no step: gone from its stream, or
first in line while a node took a later ask. Such an ask is asked again
under a new id the key names, logged and said as build.asked-again, at most
three askings, then the key fails with why. The bus's max-deliveries
advisory would tell it sooner, but the builder is granted neither it nor the
workers' information, and the controller is not changed here.
builder_stuck reads every waiting key with its asks, askings and steps;
builder_ask_again writes the new reasks state, which the builder watches
and answers.
mesh/check failed: integration: mesh-delivery-indexer/cmd/mesh-delivery-indexer.TestIndexerCountsOnlyTheAssignedBuildNodesToolchains: within 5 s of n2…
Candidate.Base was never written non-zero (review round 2): the builder's base lookup would have found no trunk for any
live candidate and built no preview at all. The indexer now writes, whenever it writes a candidate's fingerprints, the gate's
source revision whose trunk heads it read the previews onto; that head of the pull request's repository is the preview's
trunk. No mesh-sdk change: the field and its meaning were there. The builder's test now takes the candidate the indexer's
own test writes (testdata/candidate-written.json, kept equal to it), not a hand-made one.
The builder took a candidate build's merge base from the module's value commit. Now that a value keeps the commit its
fingerprint first appeared at, that commit can be many merges behind the trunk, so a preview would be built onto a stale
trunk (review round 1). The base is now the head of the pull request's repository in the gate's source revision the
candidate names (Candidate.Base): the builder reads mesh-delivery-gate.source, watched with its history (a direct get of
one revision is not what a read grants), and a candidate heard before its revision is built once the revision is heard.
The store's nightly collector deleted the bare blobs of node-tools', mesh-host's and the forge bridge's done keys, which
the controller had never taken in, and the builder kept them done: their next delivery would refuse. The builder now asks
the controller's held before it leaves a done trunk key done (once per key per process, at its start and on each module
value), claims a key whose blob is gone building as a new revision and says so in its log and the key's step, and hears
artifacts-gone, said when a take-in found a gone blob. No timer: a start, a value or an event.
mesh/check failed: novox: novox/mesh-catalog#464: checks/health/go.mod names mesh-sdk a9560890ee26, which is neither on mesh-sdk's trunk nor the group…
jschoubben