A judge pass that failed lost the set of manifests the take-in newly wrote, which
lived only in its memory: the next pass found no verdict, took each as accepted
before any verdict (ADR 0579 §1), and wrote a refusal as a warning. The take-in
now keeps each new digest in the indexer's own state unjudged before it writes
the module naming it; every judge pass judges those as fresh and ends a key once
its verdict is written or no module holds it. indexer_read_unjudged reads the
state; indexer_judge judges it again by hand.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The look arms a key of the state bounds with its own time to live; the bus's marker of its end raises call-unanswered
about the build key while it is still building and no look holds it, naming builder_build_again; the key's next written
end deletes the bound and clears the finding. A bound that cannot be written raises at once. No timer.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The put carries the ask's id as its message id, so it is made again through a deadline and no responders within the bound;
a put or a note that still meets one fails the key as a transport fault, which the builder does not retake by itself
(ADR 0664 §3) and the hand tool does. A key left building with no look holding it (its end never written: "claimed", no
ask) is taken as it stands by builder_build_again, from another builder too.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A failed publish deleted the kept record for "the next writing of the key", which nothing made.
The ask now joins the teller's wait for room, kept on the bus, which asks again at the next end
and at the start. The gate's asks are left as they are (ADR 0666 §6 replaces them).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The recovery of an ask that ended at the channel unheard needed a restart. The tool writes the key
again under a new asking by the ask-ended rules (a proposal's authorising ask and a put's record ask
are refused), refuses an asking the operator answered, naming the answer, and records the hand act
through the seat mesh-hand-acts. One tool added to the manifest.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The indexer's call for a merged message is a kept call (CallKept). A load that met a deadline or no
responders keeps the entry as unread (Pending.Unread, indexer_pending's unread), with a line that
names the read again rather than a drop; after every message heard on FORGE, of any kind, an unread
merge is evaluated again. The bridge's own answer stays lost. indexer_take_in_pending already reads
the message again (Q8).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
An ask-ended the teller said before the assigner's consumer existed is not delivered to it (note 155).
At each bind the assigner calls the seat verb operator-channel.asks once, a kept call, for the ended
asks since its earliest held asking, and writes again each one that ended expired or cancelled, by
the ask-ended rules and their compare-and-set. A seat verb: no module named. One invoke added.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
#650, its repository's head, was dropped on the advice of a line that named a drop where nothing
covered it; its changes would never have been taken in. The rule lived only in the record and the
advice; the tool now checks it, and the advice names the read again.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Note 155: a manifest-reads-changed said before the assigner's consumer existed was never heard, and
ace's bootstrapped statement stood. Each part registers the engine-reads keeper's AtBind on its hub
and calls the verb as a kept call (CallKeptOn). The asks-at-bind half (Q7) is not in this commit.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
An ask that ended at the operator channel without an answer while its key stood was never
written again, so nothing asked it (note 151: the nats yes on novox, three backup asks). The
assigner now consumes mesh-delivery-teller.ask-ended on its durable consumer and writes the key
again by compare-and-set under AskingSeconds (above the old asking), an id of its own and a new
expiry (AskLasts), one row per kind: hold, restore, backup and a record ask of the rules are
written again; the authorising ask of a proposal is not (ADR 0567 §5, §11: its lapse ends the
proposal, again asks it again), nor a record ask a put asked (ADR 0546: its lapse refuses the
put), nor an asking the key no longer holds.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
An assigner ask that lapsed or was cancelled at the operator channel while its key still stands
was never asked again: the router drops a second ask under an id it keeps, and the assigner
only writes a new asking when the key is absent (note 151: the nats yes on novox and three
backup asks). The teller now says ask-ended (its module event, emits; body in its own package
tellerevents, ADR 0634 §1) when the router records such an end, and when a watch hands it a key
standing under an asking that ended so, which covers its start and the asks standing now. A
refusal, a replacement, an answer and an end by hand (teller_end_ask, kept in its own record)
are not said.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/check checking b1343147e80f on 8d9683d844ec | novox/mesh-delivery#197 in group fix/1966-seat-list-unread
Their seat_list_again answered person.<name> and the healers, but every call
through the mesh MCP server is proved as its machine's tool runner (issue
627), so no operator's session could ask it; node-tools now stands for that
session, the kind met a second time in the node-engine (issue 1966). The
assigner's silent stand-in keeps the bus's retry bound short, since the
seat list's reader now asks again while its ask reached no one.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The assigner took record.novox.nats away as stale in the same second the operator's yes
was written (the yes counts only while the ask is there), and rewrote it under the id the
router already kept, so the teller never asked it again. The stale test compared a module's
ask with the whole node's composition; it now reads the module's own have, and restore and
record asks are numbered by their writing's time as hold asks are.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The hold named the same commit twice when only a mesh package moved. It now names both builds
by fingerprint with their commits, the input that moved when the commit is the same, and says a
newest build that failed as failed with its first words.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A put with hold false waited for the node's composition digest to equal its own and never ended when the node composes its own,
pinning the module's want (issue 1966, F21).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A large node's request is 1.2 MB, over the bus's 1 MiB message limit, so the compose ask failed once keeping no longer did.
The checker packs the request (a JSON string "gzip+base64:..." in the part's `any` request, no contract change) and
buildstep.ComposeJob.RequestBytes reads either form. A kept request of another version is refused naming it, and listing
no longer reads pieces. An older mesh-build-go refuses a packed ask loudly (it holds no request object): the check fails
naming it, so the catalog change is delivered with this one as a group.
Quick fix (chosen as the recovery): pack the request, ~290 KB for the largest node. Thorough fix (milestone G): the request
named by digest and served on a seat verb, as RunInput is.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh-host #219's check failed: the request kept for g14 was 300127 bytes compressed, over the 262144 the state takes.
The key of the check and node now holds a pointer (digest, size, pieces) and each piece of the gzip bytes is a value of
its own under pieces/<digest>/<i>, at most 128 KiB; checker_composition_requests follows the pointer and checks the digest.
Quick fix: a larger cap or a better compression: still a limit, met again as the mesh grows. Thorough fix (chosen, and
bounded only by the bucket): pieces. Not an object store: the checker is a foundation part, whose data stays in the old
sections (ADR 0619 §2), and the tool runner serves no old-section object read to its tools. No manifest change.
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh/check failed (6): integration: mesh-delivery-indexer/cmd/mesh-delivery-indexer.TestIndexerWritesModule…; and 5 more | 3a63813b3d75 on fb69c8dee691
mesh/delivery closed: a newer head of the same pull request
An indexer that started while the bus's storage was down read no published package,
took every import for unpublished and wrote the version line over ~130 modules'
fingerprints, which a later pass kept through the build's shortcut. A failed or empty
read of the packages state, where held modules were computed from published ones,
now fails the pass and says why; a version line clears the import form it was
computed beside; indexer_fingerprints_again computes the import form again by hand,
acting first and recording the hand act after, with a bound.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The verbs ask mesh-delivery-indexer.modules and .manifests through the tool runner
since the asks moved off the indexer's state, but the test's stand-in answered every
ask {ok:true}, which is no answer: propose failed on main with 'unexpected end of
JSON input'.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The grants part wrote a binding into a membership on the bus module's
created, but granted its subjects to no account but the ledger's, so
distribution's program and the tool runner carrying it were refused their
own key-value, object store and stream entries (issue 1966, ADR 0595 §1,
design 25 § 12.2). The kept bindings now reach Declared in the same pass that
writes them into the membership, and one function, boundOf, grants each to
the module's own account and the tool runner, as its old section is granted;
a process's bus.state names a key-value entry as an old state. The composed
membership carries the bindings, so grants_membership's two sides agree.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check failed (2): integration: mesh-delivery-assigner/cmd/mesh-delivery-assigner.TestAProposalIsAnswer…; and 1 more | af3b678a92f0 on 042b8d1fc160
holdLost builds a pending take-in naming neither WaitsFor nor Unknown, which TakeIn's check refused, so the hold stayed
in the journal and the watchman never saw it.
Quick fix: set Unknown on holdLost's event. Thorough fix (chosen): the check accepts a pending revision that names Lost,
the field made for it; one naming nothing is still refused. Tests: a row accepting Lost, the refusal row kept, and the
lost-hold scenario encodes its said event.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
Quick fix: none smaller than this (the bridge fix alone leaves a covered merge needing the bridge).
Thorough fix (chosen): loadNewest asks takenAlready (already used for group members) before loading the newest merge of
a repository; at or behind taken it is marked Covered, said once in the log, and never read. A merge whose ancestry
cannot be asked is read as before. Test: a row pair behind/ahead of taken with a bridge that cannot answer.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
Both ForgeMade and Tags took the tool's wrapper for the pull request / the tag list, so every forge-made merge read
"#0 is not merged" and bridge_tags failed, leaving builds unnumbered.
Quick fix: unwrap "result" in forgeAPI (the one Go reader of gitea_api).
Thorough fix (chosen, the kind was met twice and hid behind a drifted fake): bridge.API answers what the tool answers,
forgeBody is the one reader of that shape and refuses an answer without a result, naming it; ForgeMade refuses number
0 as "no pull request"; the unit fakes answer the tool's real wrapped shape, with rows for the bare body.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The operator, 2026-11-09 09:07: raise only. The grants part's assert raises it, says it in its asserted event and the
journal with the old and new cap, and grants_assert asks it by hand.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
jschoubben
mesh-tools