The store's address is the registry seat holder's listen, so the provision
artifact-store retires; the sweep and the read tools read the store through
list, never by its files or its read side.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
mesh/check failed (9): ace: novox/mesh-sdk#167 does not merge into the trunk: files in conflict: go/catalog…; and 8 more | ae5266e77639 on 2923c82001a5
A store that cannot remove a tag alone (another tag names its digest) kept the tag and the holder said
removed, so the builder and every follower believed the tagged version gone while it stayed pullable:
work reported done that was not done. The shared registry part gains the outcome kept: in remove's
answer (kept, the reason shared-digest, the tags still naming the digest) and as its own event, in all
four registry seats. Regenerated with mesh-catalog's seatgen (join-mesh-catalog ccb7eaff); seatgen
-check passes; fixtures for the answer and the event in each seat.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
A client requires the seat it speaks and mesh-sdk; the seats and the seat list require nothing. The pins are the commits
before this one, as a module cannot name the commit it is part of; the group's check builds against its members' heads.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
seatgen writes a string's enum as Go constants, so the sweep names
RemoveInputReasonRetention instead of the literal. release of what is not
held is no error. The package seats' fixtures name a package as design 32
does (git.novox.be/novox/mesh-sdk/go, @novox/mesh-sdk), name says so, and
module equals name for a Go package. Generated with mesh-catalog's seatgen
(wip/registry-a3); seatgen -check passes.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The assigner, the teller, the view and the gate each wrote the same two checks over the ask a value of theirs carries;
the fourth copy made it a kind, so it lives once in the seat's client (Carried) and the four call it.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
Each change of an artifact's series the checked head makes is {module, artifact, from, to} under series; the
fixture carries one and the regenerated round trip reads it back (seatgen -check clean).
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The lead's ruling (14:2x): the pull waits ENG-9 shows are carried in the engine's have, read through node_haves, as
ENG-9 allows, so the seat serves no verb of its own for them. Regenerated; seatgen -check finds nothing out of step.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The checks the node-engine, the checker and the assigner share over the verb (ADR 0593 §7, §9) leave mesh-sdk's delivery
package, where they sat over aliases of the seat's types, for go/clients/nodeengineclient, a Go module and package
artifact of its own over the seat's generated contract: the verdicts, the module states, the asker's bound, the no-answer
words, Encode/DecodeAsk, Encode/DecodeAnswer and CheckAnswer. Its test holds the cases mesh-sdk's held, and that an
open input's number keeps its digits; each fails without the code. seatgen -check finds nothing out of step.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The forge began mesh-contracts with a commit of its own (ADR 0428), unrelated to the branch's history; the README keeps
the branch's text and the forge's paragraph on how main is reached.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The node-engine seat, defined whole by lead-47's L3 (ADR 0618, ADR 0613 §3), is regenerated in A1's layout (ADR 0625 §1,
§3): go/seats/nodeengine a Go module of its own, ts/seats/nodeengine an npm package of its own, and module.json declares
go-nodeengine and ts-nodeengine. seatgen -check finds nothing out of step.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
One @novox/mesh-contracts would rebuild every TypeScript user at any seat's change, since the indexer
counts the npm package a lock resolves. Generated by seatgen c1fdd7c; ts/ keeps the seat list alone.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
go-package-registry's published says the module (field module), so a reader fetches the version by its path. The
four registry seats take their events whole from the part registry, which refuses a seat redefining one, so the
field is the part's, optional and empty for every other kind.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The messenger keeps them in its own stream asks (its manifest's streams, design 55 § 11), so they ride that
stream's subjects, as the checker's verdict does.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
Their output schemas replace mesh-sdk's hand-written SignAnswer, EnrolmentTokenAnswer, NodeAccountAnswer and
Signed, on the same wire names; signed's time is RFC 3339 text, as the generator writes every time.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
verdict, tag-wanted and preview.written ride the checker's own streams, renamed, retiring and retired
mesh-config's config-events stream, whose subjects are mesh.stream.<module>.event.<event>.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
mesh-sdk's asks logic (an ask's check and digest, the act's digest, a warrant's For and Says, the condition's
encoding) and its asker client move here, over operatorchannel's generated Ask, Option, Warrant and Condition; the
digests are those asks gave (a row holds two of them), so a router and an asker of either build agree.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
Each client binds mesh-sdk's seat-neutral parts sender to its own seat's publish and publish-part and types, so no
client uses another seat's; each is a Go module and a package artifact of its own (ADR 0625 §3).
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The lead's ruling of 13:0x replaces 12:5x's said-subject: mesh-broker's events follow every seat, on
mesh.seat.mesh-broker.event.<event> in the shared events stream, 7 days; the bus module keeps a longer history on a
stream of its own module. So every seat family is that stream filtered to the seat's subjects: a definition says 7
days and no cap of its own, and the reader refuses anything else.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
ADR 0625 §1-§3 (accepted 12:45): the seat set stays the root Go module's package; each seat defined whole is a Go
module of its own, since the fingerprint counts a Go module's go.mod and go.sum, so a change of one seat moves only
its importers. A seat's module imports nothing: the registry part's types are written into each registry seat whole,
so no seat module requires a sibling's version.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
The lead's ruling of 12:5x (design 25 § 12.3): mesh-broker's events ride the bus module's own stream, 30 days,
on mesh.seat.mesh-broker.said.<event>, outside the shared events stream's mesh.seat.*.event.>, so the bus never
refuses the two as overlapping.
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
jschoubben