ADR 0592 §5, §6, §9, §25 (the grants part's share by ADR 0593 §10): an object bucket and a stream that
keeps its history are created create-only with their caps and no age limit, refusing new data when full;
the owner alone writes its bucket and a module its grant names reads it; the part raises
mesh-grants.history-full once any passes 75 percent. Each engine also reads the indexer's manifest
verdicts by one key for compose-candidate.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0593 §6, §7: the grants part creates composed-<node> per node, grants each engine the writes of its
own bucket and a direct get of the indexer's candidates for compose-candidate, and grants
mesh-host.composed read-only, with the stream listing and creation advisory a per-node read needs, to
the provisioner, the assigner and the view alone; memberships issue it per node.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0593 §1: buscompose's engine moves from mesh-sdk into mesh-grants/internal/buscompose, its golden
user lists unchanged; the controller's raise and assert code is left behind, and the merged switch
test runs inside the part's whole integration test.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The counts change every pass, so a consumer that stays behind was said to the operator again at each pass. The
condition names the consumers alone; how far each is stays in the last pass.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The self-check's bus-consumer probe judged consumers the controller defined, and with the controller ended it judges none
(issue 1126). The grants part owns the account consumers, so it says when one stays more than 1000 messages behind its
stream's head at two passes running, the probe's bound and confirmation, and clears it once none is.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
mesh-sdk's group B member renames the grant to Grant, drops the
controller principal, the signer and the lease fields from the user
list key, and the keeper's and config's take-over verbs. The grants
part, the assigner and the view use the new names; the judge's copies of
the health and config seats follow; the tests' fixtures lose the
controller's user, and the refused principal is one removing what the
closed set does not name, the controller's mirror being gone.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0590 §3 ends the choice of a condition source by the keeper's
take-over key: the controller is unassigned and the keeper is the one
source. The part still read the take-over key, fell back to the
controller's condition events when it was absent, and read and watched
the controller's lease, which nothing writes any more. It now hears the
keeper's events alone and reads neither.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The controller ended by its unassignment (ADR 0590), which supersedes the
lease write and its hand-back whole and makes the grants part's list the
only one the bus node follows. The lease write, its check, the removal
marker, the take-over and hand-back asks, their five tools, the seat
claim and the lease and takeover states had nothing left to act on; the
membership notice on its subject had no reader. A whole-module test on
the bus replaces the lease write's: with no controller at all, the part
publishes a membership in its state alone, makes the module's consumer
and writes the key the bus node follows.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The controller ended by its unassignment and its lease key expired, so
every gate on "the lease is handed away" stayed shut for good: no account
consumer was created or matched, grant-missing was never raised, the
enrolment users kept the controller's subject, memberships were still
said on their subject and the sealed key never said LeaseHanded. The
handed branch now always runs, and a hand-back is gone.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
mesh-sdk 22c03af (feat/l3-enrol) requires overlay-key on every enrolment; the fixture of the enrolment-user subtest carries one.
(issue 774, review round 1b)
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0583 §1 (insight 2026-11-01) and §2: the grants part reads the vault's node-logins as its named reader and
composes a node user per login, both while the node rotates and the new one alone after rotation-done; each node's
engine reads its own node's key alone (composed by mesh-sdk while the vault serves rotate-node). The part follows the
state by watch, so a rotation begun or ended composes again at once. A second login is judged its node's when the
list is compared (Placed), and a hash changing by a rotation node-logins names is not counted as a writer changing
several hashes at once (sealed.go). The state not on the bus yet is a standing gap: every node logs in as node.<n>,
as today. Tested as a subtest of TestGrants, no new test function. Built against the group's mesh-sdk head 5000774
through a go.work outside the worktree; the pin is the lead's.
(issue 774, 774/S53)
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0587 §4: the grants part composes enrol.<node> from a key at live and the vault's hash, granted only the seat's
redeem and its own inbox; §5: when the key ends the user goes. The part reads mesh-enrolment.enrolments as its named
reader and follows it by watch: a key reaching live composes the user, and its end (by the holder, or by the bus at
its expiry) is a change of the key that takes it out, so no consume of the seat's ending events is added beside the
watch (the subtask named one; the watch already hears every end, the bus's expiry included). From the lease write the
user loses the controller's enrolment subject (§6) and the standing gap of the controller's tokens is dropped; before
it the gap stands. Tested as a subtest of TestGrants, no new test function. Built against the group's mesh-sdk head
5000774 through a go.work outside the worktree; the pin is the lead's.
(issue 774, 774/S52)
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0549 §5, ADR 0564 §9, ADR 0571 §10: the part no longer reads mesh-controller.assignments (the union with
placement goes; placement not read is a failed read that places nothing) nor consumes the controller's
condition-raised and condition-cleared, the keeper's being the one source. ADR 0571 §13's test for this
repository replaces the guard of ADR 0549 §6 and its list ../assignments-readers, now empty and deleted: at
phase write no module.json here depends on mesh-controller beyond the lease and the three kept to the write
(catalogue.ManifestRefusals).
Tests: in TestGrants the fixture places through the seat's placement (assign, unassign), the subtest of the
stale assignments becomes 774/S29/AC1 (placement alone gives the fixture's three memberships with no read,
consume or call of the controller; the fixture is the suite's own, whose assignments named the same modules,
not a copy of live memberships), and the assignments-readers subtest becomes 774/S24/AC1. No new test function.
(issue 774, 774/S29, 774/S24)
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0582: lease-write is refused, naming each check, until the lease check at phase write and every condition
of clause 2 read green (design 47's earlier issues resolved, every node trusting the vault's key, mesh-config's
compare empty from the indexer, the upgrade compare empty, the keeper's take-over done, the vault minting with
nothing held back, the bus module listing its expiry and removal tools). It asks the operator once; the yes
binds the act, the holder, the ask and the checks' digest, and the checks are read again on the warrant (one
that reads otherwise refuses it, named). Then, each step read back: node-service-manager.stop on the
controller's node, the list composed with Records.LeaseCut and waited on until the bus node's have says it
loaded it, the expiry removed through nats_bucket_expiry, the key holder written (with the ask and the time),
lease-handed said, the act recorded through mesh-hand-acts.warranted, and removal-due written with a day to
live. lease-altered is said once when the key or the expiry differs; lease-hand-back reverses it for a day
and is refused once the removal ran; removal-due's marker alone starts the removal of the controller's
consumer and the stream ASSIGNMENTS, recorded as hand acts, and a failure says removal-failed and writes the
key again with a day (grants_removal_due is the recovery for a missed marker). From the write memberships are
written to the state alone (ADR 0571 clause 10), the user-list key says lease-handed, and the user list's
hand-back answers gone (ADR 0547 §11). Every step waits on an event, never a timer.
Bent rule: S14-S21 are one commit, not eight: the steps are one state machine in leasewrite.go, and splitting
it by hunks would review the same file eight times for nothing it protects.
Tests: TestTheLeaseWrite (new, one function with a subtest per subtask), paid by the deletion of
TestTheKeepListNamesItsRecordsAndThreeToTheWrite (moved to mesh-sdk, S4/S30) and the assigner's two deletions
(S25). Fixtures shaped from each writer's code: systemd's Manager.Status, nats-tools' BucketExpirySet,
ConsumerRemoved and StreamRemoved, mesh-issues' Brief, mesh-delivery-view's TrustsAnswer, mesh-config's
Comparison, the assigner's upgrade compare, mesh-vault's vault_status.
(issue 774, 774/S14, 774/S15, 774/S16, 774/S17, 774/S18, 774/S19, 774/S20, 774/S21)
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
Issue 1585, S6-S8: 192 judge-only declarations (the strict parser, every manifest problem function, the tables)
moved whole by an AST mover, methods becoming functions; 163 helpers composition keeps too are copied. 245 tests
move with them, reading the captured catalogue through a copy of internal/beside. judge.WithTemplatesFrom and
ReadTreeManifest judge what mesh-sdk's readers now only fill. mesh-delivery's own tests decode. Pinned to mesh-sdk 25a2cec.
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 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.
mesh-wireguard is provided by the mesh itself (mesh-sdk delivery.ProvidedByMesh): no repository builds it, so the
indexer holds no fingerprint of it and the assigner never wants it by one; the controller composes it with the
node's own. The grants part named it "no fingerprint of the new form: the indexer holds no value of it" on every
pass, a warning nothing could clear (sweep 2026-10-30 item 6).
Quick fix: drop mesh-wireguard by name from the missing list.
Thorough fix, chosen as cheap and covering every module the mesh may provide later: MissingGrants passes over any
module ProvidedByMesh names. The controller says the same of it in mesh-control (its grants and switch lines).
Met twice on 2026-10-30 (the engine's acts, then mesh-host_health): the composer refused one principal as a
second writer, and that refusal ended the whole composition, so the list in force stood for every principal
while changes piled up; the next list that composed moved 52 users at once (16:37:28).
Quick fix: leave the refused principal out and write the rest (locks that principal out of the bus).
Thorough fix, chosen because the kind was met twice and it hides every other change: ComposeEach composes each
principal on its own; a refused one keeps its grants of the list in force (the part's sealed grants), written as
the composer writes an entry; one the list in force does not carry stays out; a list that would drop one the
list in force carries, whose grants were not kept, is held, said. The refusal is raised as the warning
mesh-grants.principal-refused, by name and why, cleared once it composes, and listed in the pass
(principals_refused). The entry renderer mirrors buscompose.ComposeAccounts's lines and is pinned to them by the
test (the entry the bus loaded is written back byte for byte); it belongs in mesh-sdk as an exported entry writer.
The sealed-grants value held every user of the last two lists in one value; at 61 users it was refused with
"message size exceeds maximum" (JetStream 10054, 16:37 and 16:55 on 2026-10-30), so the permissions of the
list in force were never kept and a grant judged against them waited.
Quick fix: keep only the newest list in the one value (halves it; it still grows with the mesh).
Thorough fix, chosen because the value grows with every placed module and the kind was met twice: one key per
user and list, sealed-grants.<digest hex>.<user>, and an index sealed-lists naming the kept lists and their users,
written only once every user's key is there; the third list's keys are taken away, and the old one-value key read
as a fallback and removed. mesh-sdk's kvlimit guard (7b02f63) refuses an oversized value before the put through
process.State.Put, which the part's own store (cmd/mesh-grants ownStore.Put) writes through once pinned.
mesh/check checking 57193ab2e474 on f7cd4698db7e | novox/mesh-delivery#115 in group fix/runner-carries-no-engine-state
Second meeting of the engine-state-in-runner kind; thorough: the runner
never carries the engine module. The live placement names mesh-host,
and with its health state the runner was a second writer, so every
grant since 13:54 was held back. The seeded-placement test now carries
the engine's manifest and refuses any runner write of mesh-host_*.
The state memberships held the part's wrapper ({node, module, digest, membership, published, retired}) while every
reader takes the bare membership the assignments stream carries: novox's tool runner decoded an empty one and served
no tool. The part now writes membership.EncodeGrantsMembership of a membership published or unchanged, never one held
back. The pass record moves to the pass the part already keeps (user-list at pass): published or why not was there, the
retired seat verbs are added as Result.Retired; grants_membership reads both, so no new state is declared.
Issue 774 (the bus login list, L3). Quick fix: marshal m.Membership here. Thorough, chosen: one codec in mesh-sdk the
writer and the tool runner share, because a hand-built fixture agreed with the mistake. Pins mesh-sdk at the head of
its fix/memberships-shape.
Why: after the lease write the assignments are frozen, so the part's union kept a module the catalogue takes off a node
with its account, grant, membership and user-list entry. The lease is read as the assigner reads it
(buscompose.LeaseHandedAway); before it, or with no placement on the bus, the union stands. The read stays declared until
the first change after the lease write, named in modules/assignments-readers with the assigner; a TestGrants row refuses
any other module.json that reads mesh-controller.assignments, as mesh-catalog's checks/placement does for the catalogue.
Why: the controller's grant writer skipped link2pay, snake, f1-circuits and txt-game in silence (it read the fingerprint by
their unchanged commit, a build of the old form) and the take-in said 'its grant written'. The part already reads the
fingerprint from the indexer's value, so those four are granted at their src2 done key's fingerprint; a row proves it.
What a pass leaves without a grant (no fingerprint of the new form with the indexer's none2, a refused manifest, no
hashed account) is now said rather than passed over. Quick: log it. Thorough (chosen): the pass's Missing, shown by the
grants tool, and a warning condition raised and cleared on change, as consumer-differs is.
Why: the assigner switches its grants source when the controller's lease is handed away; until then the written keys are compared with the controller's, so the switch rests on a measured equality. Needs the group's mesh-sdk pin (PartGrantKey) before it builds.
Live since the take-over the shadow list was written every 3 s (revision 99270
to 99357 in 90 s, nothing differing): the sealed key read back says [] for the
nil left-out a pass composes, so every pass wrote the key again; each write
moved the bus node's report of what it loaded, whose revision and time were
inputs of the shadow list, and woke the next pass. The bus itself reloads only
when the list's bytes change (its log: two configurations alternating with the
placement, f8567865 and 473f9ad3), so this loop was load, not the reload.
Review round 1 of issue 1945's fix found two ways the guard could let a loss
through or hide one from the operator.
A publish deny was never compared: grants.User carried no deny, so a widened
allow that brings a new deny (a lost publish right) passed as an added right.
The user now carries its deny; a deny the shadow adds is a loss, and while
nats_users answers no deny for the loaded list, a user whose grants differ
and whose shadow grant has a deny refuses, named as a deny the loaded list
cannot be compared on. None of today's 41 differing users carries a deny
(only the four node-tools users do, and they are the same in both lists).
The ask said "takes none away" whenever rights were added, whatever else
differed. Its words now name the added rights only when nothing else refuses
(Refusals: unread list, Removing, a missing hash), and otherwise say what the
yes would be refused for, naming the users; Decide uses the same Refusals.
issue 1945
The operator's yes to take-over was refused twice (2026-10-26 13:48 and
2026-10-27 21:19 UTC) because 41 users differ, every difference a right the
grants part's list adds. The controller's list is not changed (ADR 0547 §1),
so ADR 0547 §8's equality could never hold again. ADR 0562 changes it:
take-over is allowed when no user is added or dropped, no loaded user loses a
publish, subscribe or responses right and every hash is held; the ask names
the added rights by user and kind, and a yes after they changed is refused.
Tests: rows from the captured live comparison of 21:19 (testdata, no
hashes) in the compare table, and rows in the take-over table for a right
removed, added rights changed since the ask, and only added rights recorded.
issue 1945
jschoubben