The owner's watch of a state keeping one value per key is never handed a
revision that a later write of the same key replaced first, so every such
skip read as "changes missed" and the reader asked its first view again,
over and over in a burst. Each event now names the revision said before it
(prev), and a reader takes it when its picture holds that one.
A change whose event never reached a reader was only noticed at a later
change, and a lost last one never. The owner now says its mark as an event
each time it writes it; a reader short of it asks its first view at once
and holds what it read, never the event's body.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A restore point is taken before a module moves across a change of how
it writes a kept data, so each declaration (the old sections and every
entry in requires by template) names its form, 0 where none is said,
and the catch-up tool a restore is followed by. KeptDeclarations reads
them in the one walk KeptStates now uses; FormChanges is what differs
between the running manifest and the one applied (design 47).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The literal server=1.1.1.1 broke on every pull request when the upstreams became settings; the dnsmasq configuration is checked in the module that owns it.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A definition carrying acts was refused as an unknown field by every reader (design 47, What ADR 0656 leaves to this
design). seatdef reads acts at either grain: each rule's step one of the seven, on every or a language, a bound as a
duration, one step on one language once; ActsOn and Wait give a rule's match and its bound, the step's default when it
names none. W1.2: the node-engine copy follows mesh-contracts a6c5eaa6 (its compose rule).
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The seat mesh-delivery's and mesh-delivery-checker's owed events are kept on each owner's stream owed, 7 days and at
most 64 MiB (design 47, What the gate owes the forge); the format could only keep an event last per subject.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The node-engine's running.<node>.<module> is kept last per subject with no age limit, which the shared events stream cannot say; the reader takes keep last-per-subject.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The seat git's events (pull.updated, forge.status) and design 47's events with parts
(owed.<repository>.<number>, running.<node>.<module>, milestone.<milestone>.asked) could not be
written in a definition: an event's name was one word and its subject the name. An event now takes
dotted words and an optional subject whose words are its name with parts among them; the seat list
says it by its subject with * for each part, so a consumer naming it is heard as the seat's.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A key-value entry asked in requires (ADR 0595) was bound but never issued
as a state, because each reader read the old state list alone (issue 1966).
Membership.StateIssued and StateNames, and catalogue.KeptStates, are the one
rule the tool runner, the grants part and the manifest validator call.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check checking f6a48060d1c7 on 7798fb481e9d | novox/mesh-sdk#176 in group fix/1966-forge-made-merge
Merges made on the forge directly since 2026-11-09 07:04 UTC have no gate merge record,
and the indexer holds them; an empty merge through the gate covers this repository's head.
Second of its kind: the group's mesh-delivery member removes the need (the forge bridge
reads a forge-made merge's pull request from the forge).
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check checking 38659c3937e7 on 55f71b78114f | novox/mesh-sdk#174 in group fix/1966-views-durable
A consumer is made from now (ADR 0443): an event said before it was made is never delivered to it and no later event
shows the loss. The event hub tells a part when the bus made its consumer, and a follower asked before that asks
again. Events passed over, at or below the held revision, or of a state not followed are now said in the log. The
status carries the highest revision heard and why the picture is behind it.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
A follower passed over an event heard before its first view and dropped a gap heard while an ask ran, so a quiet
owner left the reader behind until a further event. It now keeps the highest revision heard and, when an ask ends
short of it, asks again on a pause of its own until it holds it. The status a read tool shows followed only the
asks, so it said a reader was behind that was not; it follows the events too.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check checking f56e94d87ea9 on 62ec14abdd8b | novox/mesh-sdk#173 in group fix/1966-gate-record-2
mesh-sdk #172 (c861bd8a) was merged on the forge and has no gate merge record; the indexer holds its take-in. An empty merge through the gate covers it, and the forge-made wait is dropped (indexer_drop_pending).
Claude-Session: https://claude.ai/code/session_01NcbgfFfQ4Z9UE5jK32PmE5
mesh/check pass 070e1030d887 on c1b50e99a639 | carried from a0e5530aef1b | unit 310 run, 350 kept, 1 skipped | integration 309 run, 383 kept, 1 skippe…
B0's first group check failed because the running indexer, built at this trunk,
refused mesh-build-go (invokes seat:mesh-delivery-checker.input) and
mesh-delivery-checker (serves input), and the check's composition took that
refusal. Publishing the verb first rebuilds the indexer (ADR 0573 §3); B0 then
checks as reviewed. The lines are B0's own, reviewed at 1b073c23.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
mesh/check being carried to 1b073c23c2a5 on 54aff0a3d920 | from c1b50e99a639, hq 885099eebbbd, mesh-catalog 980ade933853, mesh-delivery 2d201f781ced
The operator's answer: a build module reading the checker's state broke ADR 0596 §2. RunInput names the seat verb it is
served on (seat:<seat>.<verb>), RunInput.Read asks it and takes only the bytes named, and the seat
mesh-delivery-checker defines the verb input (RunInputAsk, RunInputAnswer).
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
The verdict gains judged-by: where the checking program was built, the
key its build is kept under and the heads it read, so the checker's
reads say who judged a check (ADR 0644, design 47's bootstrap delivery).
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
A build module that fills a request re-marshals it, and without the
field it dropped the seat list the trunk's manifest validator reads, so
the validator never got one (milestone 47/B0, ADR 0636 §4, ADR 0644).
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
A run ask has a program built at a commit and run in the build seat's
sandbox over named heads, its input kept by the asker by digest and
measured against a bound before asking (ADR 0644, ADR 0656; design 47's
ask run). The checker asks it for the checking program in milestone
47/B0.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
jschoubben
mesh-tools