Compare commits

..
Author SHA1 Message Date
jschoubben 3a969c6b99 Designs 33 and 08: the converged answer carries the node's links (H-A-325) 2026-11-21 21:14:05 +01:00
jschoubben da58998af4 Designs 08 and 33: no interface never opens an outward link (H-A-324) 2026-11-21 21:11:18 +01:00
jschoubben 41005dfc1e Designs 08 and 33: the build's forms (H-A-317 to H-A-321) 2026-11-21 20:53:37 +01:00
jschoubben 8fc4f09d8f Designs 08 and 33: the reviews by domain (H-A-314, H-A-315, H-A-316) 2026-11-21 20:43:11 +01:00
jschoubben 654a08b5a0 Designs 08 and 33: an adopted node refuses allowances; expired ones removed at the holder's start (H-A-312) 2026-11-21 20:39:52 +01:00
jschoubben 9dcbc0fef1 Designs 08 and 33: ADR 0683's number, the allowance's identity and expiry, the verbs required from the start 2026-11-21 20:34:01 +01:00
jschoubben 9e99e9562b Designs 08 and 33: the firewall seat's control verbs beside the derived ports (the operator, 2026-11-21)
The operator decided that the packet filter's module must have tools to
control it beside what is derived from the modules' declarations: deriving
every open port from listens left no way to open a port no manifest names.

Design 08 section 4 keeps the derived ports as the automatic part and adds
allowances set by hand; the node-engine stays the one renderer, asks the
holder's allowed verb at every compose, refuses the compose and keeps the
loaded filter when it gets no answer, and renders allowances only on a
converged node. Design 33 gives the seat allow, disallow and allowed,
widens rules, and adds the event allowances-changed and the holder's
state: one allowance per protocol, port and interface, removed and
journalled at its expiry. Both follow ADR 0683 and design 47 H's lead
answers H-A-304, H-A-305, H-A-307 and H-A-308.

Both designs are put in playbook 02's shape: their new lines live in
milestone parts, and design 08's adoption lines move to a delivered part
so the main file stays under 80 KB. The glossary gains allowance.
2026-11-21 20:33:52 +01:00
mesh-tools dfaabb6c77 Merge pull request 'Design 47 H: lead answer H-A-128 (nats_consumer_remove judges a module's durable by prefix)' (#795) from design/47-H1-answers-57 into main 2026-11-20 05:22:09 +00:00
jschoubben 00d6446fea Design 47 H: H-A-128, nats_consumer_remove judges a module's durable by prefix
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 06:21:49 +01:00
mesh-tools 1540d297b4 Merge pull request 'Design 47 H: lead answer H-A-127 (nats_consumer_remove refuses by durable name; why is the warrant)' (#794) from design/47-H1-answers-56 into main 2026-11-20 05:15:28 +00:00
jschoubben f3d57662c6 Design 47 H: H-A-127, nats_consumer_remove refuses by durable name; why is the warrant
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 06:15:13 +01:00
mesh-tools f1164f49ff Merge pull request 'Design 47 H: lead answer H-A-126 (the bus module's drop tools exist; H-A-121 (1) withdrawn)' (#793) from design/47-H1-answers-55 into main 2026-11-20 05:13:00 +00:00
jschoubben f0559b3a71 Design 47 H: H-A-126, the bus module's drop tools exist; H-A-121 (1) withdrawn
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 06:12:46 +01:00
mesh-tools d42bbc6138 Merge pull request 'Design 47 H: lead answer H-A-125 (the knowledge base's delivery block goes whole)' (#792) from design/47-H1-answers-54 into main 2026-11-20 04:54:45 +00:00
jschoubben 7c21fef8e8 Design 47 H: H-A-125, the knowledge base's delivery block goes whole
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:54:30 +01:00
mesh-tools 9b16d3fa07 Merge pull request 'Design 47 H: lead answer H-A-124 (the all form of indexer_pin_drop drops every pin)' (#791) from design/47-H1-answers-53 into main 2026-11-20 04:53:38 +00:00
jschoubben 53f72a6740 Design 47 H: H-A-124, the all form of indexer_pin_drop drops every pin
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:53:26 +01:00
mesh-tools ffcf21d5bb Merge pull request 'Design 47 H: lead answer H-A-123 (the all form of assigner_want_drop drops every want)' (#790) from design/47-H1-answers-52 into main 2026-11-20 04:53:06 +00:00
jschoubben a653f39e2e Design 47 H: H-A-123, the all form of assigner_want_drop drops every want
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:52:52 +01:00
mesh-tools 6d7217619f Merge pull request 'Design 47 H: lead answers H-A-121 (the clean state's drop tools) and H-A-122 (the mail reader's callers list goes)' (#789) from design/47-H1-answers-51 into main 2026-11-20 04:49:20 +00:00
jschoubben fbff9aa306 Design 47 H: H-A-121 corrected (module tools, not a seat verb); H-A-122 the mail reader's callers list goes
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:48:44 +01:00
jschoubben 198d44382a Design 47 H: H-A-121, the clean state's drop tools (mesh-broker.drop, consumer delete, all forms)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:45:49 +01:00
mesh-tools 4bebb4c1ed Merge pull request 'Design 47 H: lead answer H-A-120 (the build seats' kinds take turns, A-440; H-A-8 corrected)' (#788) from design/47-H1-answers-50 into main 2026-11-20 04:29:59 +00:00
jschoubben 1fe45361c1 Design 47 H: H-A-120, the build seats' kinds take turns (A-440); H-A-8 corrected
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:29:33 +01:00
mesh-tools 3af51ed4a1 Merge pull request 'Design 47 H: lead answer H-A-119 (pipeline-ended's mover entry gains commit and widens; the teller)' (#787) from design/47-H1-answers-49 into main 2026-11-20 04:12:00 +00:00
jschoubben ef446d6d8b Design 47 H: H-A-119, pipeline-ended's mover entry gains commit and widens; the teller hears no record-opened
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:11:32 +01:00
mesh-tools 3176501000 Merge pull request 'Design 47 H: lead answer H-A-118 (the held take-in's value carries the event; later ones wait in order)' (#786) from design/47-H1-answers-48 into main 2026-11-20 04:06:38 +00:00
jschoubben 6ea3bcd416 Design 47 H: H-A-118, the held take-in's value carries the event; later ones wait behind in order
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 05:06:14 +01:00
mesh-tools 0afc44ea6f Merge pull request 'Design 47 H: lead answer H-A-117 (the check and recheck kinds stay in H1; the runs' input is mesh-wants')' (#785) from design/47-H1-answers-47 into main 2026-11-20 03:57:41 +00:00
jschoubben c06d4dd03a Design 47 H: H-A-117, the check and recheck kinds stay in H1; the runs' input is mesh-wants'
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:57:10 +01:00
mesh-tools 162c721f08 Merge pull request 'Design 47 H: lead answer H-A-116 (the builder's floor wakes on take-in.taken in H1)' (#784) from design/47-H1-answers-46 into main 2026-11-20 03:44:05 +00:00
jschoubben 508faf0cf6 Design 47 H: H-A-116, the builder's floor wakes on take-in.taken in H1
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:43:38 +01:00
mesh-tools 853b3ae819 Merge pull request 'Design 47 H: lead answers H-A-114 and H-A-115 (a held take-in is acknowledged and kept; code dependents by go.sum)' (#783) from design/47-H1-answers-45 into main 2026-11-20 03:42:21 +00:00
jschoubben b93bf48739 Design 47 H: H-A-114 a held take-in is acknowledged and kept in state held; H-A-115 code dependents by go.sum in H1
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:41:28 +01:00
mesh-tools 0cf71a5515 Merge pull request 'Design 47 H: lead answer H-A-113 (grants_why_refused stays; every account gets every seat's verbs)' (#782) from design/47-H1-answers-44 into main 2026-11-20 03:22:27 +00:00
jschoubben f599b76165 Design 47 H: H-A-113, grants_why_refused stays; every account gets every seat's verbs
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:22:01 +01:00
mesh-tools 3c2cd089e8 Merge pull request 'Design 47 H: lead answer H-A-112 (the composer drops its verdict branch and Case.Verdicts)' (#781) from design/47-H1-answers-43 into main 2026-11-20 03:17:42 +00:00
jschoubben 49f08c9e83 Design 47 H: H-A-112, the composer drops its verdict branch and Case.Verdicts
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:17:16 +01:00
mesh-tools 8c94adf744 Merge pull request 'Design 47 H: lead answer H-A-111 (the node-engine reads a taken module without a verdict)' (#780) from design/47-H1-answers-42 into main 2026-11-20 03:16:17 +00:00
jschoubben 307168a778 Design 47 H: H-A-111, the node-engine reads a taken module without a verdict
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:15:42 +01:00
mesh-tools 7319cc5e1b Merge pull request 'Design 47 H: lead answer H-A-110 (the node-engine asks entries again on catalog-revision)' (#779) from design/47-H1-answers-41 into main 2026-11-20 03:15:06 +00:00
jschoubben 6fde0ff09d Design 47 H: H-A-110, the node-engine asks entries again on catalog-revision
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:14:55 +01:00
mesh-tools 13be7087dc Merge pull request 'Design 47 H: lead answer H-A-109 (the view says a held take-in without naming modules)' (#778) from design/47-H1-answers-40 into main 2026-11-20 03:14:11 +00:00
jschoubben 3433db0efb Design 47 H: H-A-109, the view says a held take-in without naming modules
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:12:57 +01:00
mesh-tools 996cd3c391 Merge pull request 'Design 47 H: lead answer H-A-108 (the validate step's run, input, baseline and verdict)' (#777) from design/47-H1-answers-39 into main 2026-11-20 03:11:30 +00:00
jschoubben 38fdb88994 Design 47 H: H-A-108 in the glossary's word
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:10:58 +01:00
jschoubben 9c7fa01851 Design 47 H: H-A-108, the validate step's run, input, baseline and verdict
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:10:33 +01:00
mesh-tools 0f8c0812d9 Merge pull request 'Design 47 H: lead answer H-A-107 (the pipeline keeps the movers superseded out of it)' (#776) from design/47-H1-answers-38 into main 2026-11-20 03:08:57 +00:00
jschoubben fabb927f26 Design 47 H: H-A-107, the pipeline keeps the movers superseded out of it
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:08:33 +01:00
mesh-tools eea237e45b Merge pull request 'Design 47 H: lead answer H-A-106 (the queue's over-cap start, the approved pipeline, the findings' form)' (#775) from design/47-H1-answers-37 into main 2026-11-20 03:07:36 +00:00
jschoubben e428f47913 Design 47 H: H-A-106, the queue's over-cap start, the approved pipeline and the findings' form
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 04:06:46 +01:00
mesh-tools 2cc1da4357 Merge pull request 'Design 47 H: lead answer H-A-105 (indexer_take_in goes; the run candidate's check stays)' (#774) from design/47-H1-answers-36 into main 2026-11-20 03:00:30 +00:00
jschoubben a000df7dbe Design 47 H: H-A-105, indexer_take_in goes; the run candidate's check stays
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:59:44 +01:00
mesh-tools 91266f3007 Merge pull request 'Design 47 H: lead answer H-A-104 (the assigner's state engine-reads; RefusedCycle into the sdk)' (#773) from design/47-H1-answers-35 into main 2026-11-20 02:59:06 +00:00
jschoubben 0d18ad2daf Design 47 H: H-A-104, the assigner's state engine-reads; RefusedCycle into the sdk on the contracts branch
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:58:36 +01:00
mesh-tools 0cef3fc826 Merge pull request 'Design 47 H: lead answer H-A-103 (the indexer's derived state between take-ins)' (#772) from design/47-H1-answers-34 into main 2026-11-20 02:56:53 +00:00
jschoubben 50e6d16381 Design 47 H: H-A-103, the indexer's derived state between take-ins
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:56:12 +01:00
mesh-tools 4b4802df52 Merge pull request 'Design 47 H: lead answer H-A-102 (indexer_pending's answer form)' (#771) from design/47-H1-answers-33 into main 2026-11-20 02:53:33 +00:00
jschoubben c836029155 Design 47 H: H-A-102, indexer_pending's answer form
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:52:35 +01:00
mesh-tools 960cc77a7c Merge pull request 'Design 47 H: lead answer H-A-101 (the runner's hand verbs name the module; a not-asked step says its end)' (#770) from design/47-H1-answers-32 into main 2026-11-20 02:52:10 +00:00
jschoubben 69a8854835 Design 47 H: H-A-101, the runner's hand verbs name the module; a not-asked step says its end
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:51:31 +01:00
mesh-tools b59a5893dd Merge pull request 'Design 47 H: lead answers H-A-98 to H-A-100' (#769) from design/47-H1-answers-31 into main 2026-11-20 02:50:49 +00:00
jschoubben 62d60bf43d Design 47 H: H-A-99 the placement's build is the want's target; H-A-100 the forge module's merge-check removal
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:50:17 +01:00
jschoubben 922f7017b8 Design 47 H: H-A-98, an old-form event is refused by each handler's strict decode
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:49:56 +01:00
mesh-tools 0bebde8bfe Merge pull request 'Design 47 H: lead answer H-A-97 (a held take-in is kept in the indexer's state held)' (#768) from design/47-H1-answers-30 into main 2026-11-20 02:48:58 +00:00
jschoubben 55da6d1af7 Design 47 H: H-A-97, a held take-in is kept in the indexer's state held
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:48:31 +01:00
mesh-tools f6b2837e0a Merge pull request 'Design 47 H: lead answer H-A-96 (the queue's running keys)' (#767) from design/47-H1-answers-29 into main 2026-11-20 02:47:59 +00:00
jschoubben c55ea604d1 Design 47 H: H-A-96, the queue's running keys
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:47:33 +01:00
mesh-tools d17fa4efa7 Merge pull request 'Design 47 H: lead answer H-A-95 (a held want is not timed; Owed's ask and group go)' (#766) from design/47-H1-answers-28 into main 2026-11-20 02:46:19 +00:00
jschoubben 77b05c2381 Design 47 H: H-A-95, a held want is not timed; Owed's ask and group go
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:45:35 +01:00
mesh-tools 0cef71532a Merge pull request 'Design 47 H: lead answer H-A-93 (step-unread is one finding per unread step)' (#765) from design/47-H1-answers-27 into main 2026-11-20 02:42:34 +00:00
jschoubben 33208b3cd9 Design 47 H: H-A-94, the watchman's bus-state and take-in readers
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:42:26 +01:00
jschoubben d33b17351c Design 47 H: H-A-93, step-unread is one finding per unread step
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:41:50 +01:00
mesh-tools 8c0b04e33c Merge pull request 'Design 47 H: lead answer H-A-92 (a catalog revision carries its movers)' (#764) from design/47-H1-answers-26 into main 2026-11-20 02:38:42 +00:00
jschoubben c8fc134444 Design 47 H: H-A-92, a catalog revision carries its movers
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:38:20 +01:00
mesh-tools f3e205fa72 Merge pull request 'Design 47 H: lead answer H-A-91 (an adopted branch says start; delivery-waits goes; the owed act merge leaves)' (#763) from design/47-H1-answers-25 into main 2026-11-20 02:36:11 +00:00
jschoubben 7425b11b34 Design 47 H: H-A-91, an adopted branch says start; delivery-waits goes; the owed act merge leaves the contract
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:35:28 +01:00
mesh-tools 6cacd527e3 Merge pull request 'Design 47 H: lead answer H-A-90 (a named event consumer in mesh-sdk; bound on registration-changed)' (#762) from design/47-H1-answers-24 into main 2026-11-20 02:31:08 +00:00
jschoubben e95cc598a0 Design 47 H: H-A-90, NamedEventConsumer in mesh-sdk; a consumer bound on registration-changed
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:30:41 +01:00
mesh-tools 81e1964d0c Merge pull request 'Design 47 H: lead answer H-A-89 (the record's pipeline families)' (#761) from design/47-H1-answers-23 into main 2026-11-20 02:29:00 +00:00
jschoubben 287871724b Design 47 H: H-A-89, the record's pipeline families
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:28:34 +01:00
mesh-tools 2c6c7c03d3 Merge pull request 'Design 47 H: lead answer H-A-88 (StepRun gains asker)' (#760) from design/47-H1-answers-22 into main 2026-11-20 02:26:46 +00:00
jschoubben 1a25538c0a Design 47 H: H-A-88, StepRun gains asker
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:26:19 +01:00
mesh-tools 76cc85de24 Merge pull request 'Design 47 H: lead answer H-A-87 (the forge module's whole test is Go; the source key's escaping)' (#759) from design/47-H1-answers-21 into main 2026-11-20 02:25:33 +00:00
jschoubben 4dabb3015a Design 47 H: H-A-87, the forge module's whole test is Go; the source key's escaping
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:24:48 +01:00
mesh-tools e19d61a4b0 Merge pull request 'Design 47 H: lead answers H-A-85 and H-A-86 (bridge_heads, a run's source; the assigner's pins go)' (#758) from design/47-H1-answers-20 into main 2026-11-20 02:22:57 +00:00
jschoubben 1bea97bed2 Design 47 H: H-A-86, the assigner's revision pins go whole
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:22:25 +01:00
jschoubben df47ba6a22 Design 47 H: H-A-85, bridge_heads; a run's pipeline source
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:22:08 +01:00
mesh-tools a6b6c6aab9 Merge pull request 'Design 47 H: lead answer H-A-84 (a mover keeps the publishes it heard)' (#757) from design/47-H1-answers-19 into main 2026-11-20 02:20:10 +00:00
jschoubben 1d8f7f06a2 Design 47 H: H-A-84, a mover keeps the publishes it heard
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:19:47 +01:00
mesh-tools dd3d82905c Merge pull request 'Design 47 H: lead answer H-A-83 (a run on request is one pipeline; kind-3 edges wait on the build step)' (#756) from design/47-H1-answers-18 into main 2026-11-20 02:19:13 +00:00
jschoubben ae7fc40ad4 Design 47 H: H-A-83, a run on request is one pipeline; kind-3 edges wait on the build step's end
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:18:48 +01:00
mesh-tools 0631f43f25 Merge pull request 'Design 47 H: lead answer H-A-82 (the indexer reads nothing in place of version-delivered)' (#755) from design/47-H1-answers-17 into main 2026-11-20 02:17:12 +00:00
jschoubben 0cbff22c6e Design 47 H: H-A-82, the indexer reads nothing in place of version-delivered
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:16:21 +01:00
mesh-tools 43aac6b785 Merge pull request 'Design 47 H: lead answer H-A-81 (severities of head-untaken and merge-partial)' (#754) from design/47-H1-answers-16 into main 2026-11-20 02:16:10 +00:00
jschoubben 92592c9375 Design 47 H: H-A-81, the severities and resolvers of head-untaken and merge-partial
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:15:28 +01:00
mesh-tools 9d0d941190 Merge pull request 'Design 47 H: lead answer H-A-80 (indexer_consumer goes; stopped work is committed in progress)' (#753) from design/47-H1-answers-15 into main 2026-11-20 02:14:59 +00:00
jschoubben 94fe792721 Design 47 H: H-A-80, indexer_consumer goes; stopped work is committed in progress
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:13:53 +01:00
mesh-tools c2b897e28c Merge pull request 'Design 47 H: lead answer H-A-79 (the verdicts state and its readers go; oldsections stays)' (#752) from design/47-H1-answers-14 into main 2026-11-20 02:13:38 +00:00
jschoubben cc20946308 Design 47 H: H-A-79, the verdicts state and its readers go; oldsections stays
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:12:58 +01:00
mesh-tools 1c9a24a06d Merge pull request 'Design 47 H: lead answer H-A-78 (applied compares the fingerprint; the readers' writer)' (#751) from design/47-H1-answers-13 into main 2026-11-20 02:11:19 +00:00
jschoubben 372123e744 Design 47 H: H-A-78, applied compares the fingerprint; the view, watchman and grants readers' writer
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:10:47 +01:00
mesh-tools 6179eb5a7d Merge pull request 'Design 47 H: lead answer H-A-77 (artifacts from the store; the kept-artifact fetch retires)' (#750) from design/47-H1-answers-12 into main 2026-11-20 02:10:16 +00:00
jschoubben 0bf11925c4 Design 47 H: H-A-77, artifacts come from the store; the kept-artifact fetch retires
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:09:54 +01:00
mesh-tools 2a0222bdbb Merge pull request 'Design 47 H: lead answer H-A-76 (the source say-again's body; the seed after the drop)' (#749) from design/47-H1-answers-11 into main 2026-11-20 02:06:22 +00:00
jschoubben 2f1512bd1d Design 47 H: H-A-76, the source say-again's body; the seed after the drop, refusing old keys
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:05:25 +01:00
mesh-tools ba1265100c Merge pull request 'Design 47 H: lead answer H-A-75 (pipeline-wanted's outcome and body; the node's own want unchanged)' (#748) from design/47-H1-answers-10 into main 2026-11-20 02:03:32 +00:00
jschoubben b4563e5256 Design 47 H: H-A-75, pipeline-wanted's outcome and body; the node's own want unchanged
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:03:10 +01:00
mesh-tools 49c216e299 Merge pull request 'Design 47 H: lead answer H-A-74 (nothing of the builder changes in H1)' (#747) from design/47-H1-answers-9 into main 2026-11-20 02:00:50 +00:00
jschoubben 18feae5df3 Design 47 H: H-A-74, nothing of the builder changes in H1; the toolchain wait stays
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 03:00:23 +01:00
mesh-tools a0cb55cfd3 Merge pull request 'Design 47 H: lead answer H-A-73 (the hand-placement hold is the want's hold field)' (#746) from design/47-H1-answers-8 into main 2026-11-20 01:53:33 +00:00
jschoubben ce96602bd9 Design 47 H: H-A-73, the hand-placement hold is the want's hold field
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:53:10 +01:00
mesh-tools bdecac13d6 Merge pull request 'Design 47 H: lead answer H-A-72 (the runner's and mesh-wants' seat forms)' (#745) from design/47-H1-answers-7 into main 2026-11-20 01:52:59 +00:00
jschoubben 67f21a7b7e Design 47 H: H-A-72, the runner's and mesh-wants' seat forms
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:52:09 +01:00
mesh-tools d7a45777a4 Merge pull request 'Design 47 H: lead answer H-A-71 (owed stays; the kept member gains branch; steprun's asker judgement)' (#744) from design/47-H1-answers-6 into main 2026-11-20 01:50:53 +00:00
jschoubben b4dcdbc93d Design 47 H: H-A-71, owed stays, the kept member gains branch, steprun's asker judgement is mesh-delivery's
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:50:11 +01:00
mesh-tools 96bb5261f3 Merge pull request 'Design 47 H: lead answer H-A-70 (acts stays as the kind-5 declaration; the bound and the holds go)' (#743) from design/47-H1-answers-5 into main 2026-11-20 01:40:09 +00:00
jschoubben bd6c89af74 Design 47 H: H-A-70, acts stays as the kind-5 declaration; the bound and the holds go
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:39:34 +01:00
mesh-tools daa666d40a Merge pull request 'Design 47 H: lead answer H-A-69 (the gate knows a first-view tool by its name form)' (#742) from design/47-H1-answers-4 into main 2026-11-20 01:37:30 +00:00
jschoubben d6318938af Design 47 H: H-A-69, the gate knows a first-view tool by its name form
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:36:56 +01:00
mesh-tools 5ecc060ff1 Merge pull request 'Design 47 H: lead answers H-A-65 to H-A-68 to H1's developers' (#741) from design/47-H1-answers-3 into main 2026-11-20 01:35:31 +00:00
jschoubben b24a4dc6f6 Design 47 H: H-A-67 placing step, H-A-68 a revert is a placement's build without pipeline and rank
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:35:04 +01:00
jschoubben 56b4225bcc Design 47 H: H-A-66, the whole callers set goes from mesh-sdk in H1
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:34:41 +01:00
jschoubben 3ce936a16a Design 47 H: H-A-65, the tool runner's call-timeout setting waits for H4; 30 s constant in H1
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:34:15 +01:00
mesh-tools 386079389e Merge pull request 'Design 47 H: lead answers H-A-63 and H-A-64 to H1's developers' (#740) from design/47-H1-answers-2 into main 2026-11-20 01:30:44 +00:00
jschoubben b123fb5a28 Design 47 H: H-A-64, want:version is a generated constant; seatgen splits on ':'
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:30:23 +01:00
jschoubben 4d7921a087 Design 47 H: H-A-63, apply-order's waits is a list of module names
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:29:56 +01:00
mesh-tools 34ac44de75 Merge pull request 'Design 47 H: the lead's answers H-A-21 to H-A-62 to milestone H1's prepare' (#739) from design/47-H1-lead-answers into main 2026-11-20 01:26:33 +00:00
jschoubben 6a4fbb7892 Plan H: repository names in code spans
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:26:18 +01:00
jschoubben 66a8a7bc02 Plan H: section 9, H1's develop assignments and their reasons
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:25:54 +01:00
jschoubben 0732501b22 Design 47 H: the lead's answers H-A-21 to H-A-62 to H1's prepare; §760 and §762 say what H1 carries
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 02:24:51 +01:00
mesh-tools 4f3268f6bb Merge pull request 'Design 47, milestone H: the design lines of the one model (§734 to §774), reviewed by architectural domain' (#738) from design/47-H-lines into main 2026-11-20 00:59:30 +00:00
jschoubben 76ddab3ea7 Cite ADR 0682 by its final clauses in milestone H, with no branch protection read and every ask offered
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:58:40 +01:00
jschoubben 551da194b7 Settle milestone H's last small points: the placement read, a verb's timeout, the enrolment event, the counts
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:54:01 +01:00
jschoubben 166ad75748 Design 47 H: H-A-10's bound ends the record failed, the pipeline's outcome stays
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:53:34 +01:00
jschoubben f9d3ecf2c7 Write the review's findings and the lead's answers into milestone H's lines
The four reviewers by architectural domain asked changes on the first
version of design 47 H; the lead answered them in H-A-1 to H-A-20. The
lines now carry the merge call's kept key, its partial refusal and its
tools, the clean state's seed and drop order, one consumer per branch
for the take-in, the runner's queue and per-revision events, wants
written tier by tier with a rank the node-engine applies, the producer's
publish limit, the four owners' change events, the tests split per
module with their counts against the budget, and ADR 0682 cited by its
version 5 clause numbers.

Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:50:25 +01:00
jschoubben 5f49bfd818 Design 47 H: H-A-10 is the order inside the one pipeline, tier by tier on running
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:43:20 +01:00
jschoubben dae41a142f Design 47 H: the lead's answers H-A-9 to H-A-20 to the review by architectural domain
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:42:51 +01:00
jschoubben 689ab797a1 Design 47 H: the lead's answers H-A-1 to H-A-8 to the design writer's questions
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:38:18 +01:00
jschoubben 4baf34c7c3 Write milestone H's design lines so the one model is decided before code
Design 47 carried the commit-on-a-watched-branch model and the earlier
per-pull-request model side by side; the operator's answers of
2026-11-17 and 2026-11-20 (ADR 0682) lived only in lead answers. The
part H-1 to H-3 (design 47 H §734 to §765) states the merge call as the
start and group, the gate's watchdog, the take-in and validate step, one
pipeline per take-in with none in parallel, the want rule, the builder's
missing lines, reads made events, the missing read tools and what
retires; G's contradicting lines are corrected in place, the main file
gains the wake-up rule, tests 110 to 113 and the H row, and the
glossary gains merge call, untaken head, take-in and mover.

Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:37:02 +01:00
mesh-tools e6a73d8f59 Merge pull request 'Design 47, milestone H: the plan to deliver the build-and-delivery system properly' (#737) from plan/47-deliver-properly into main 2026-11-20 00:21:01 +00:00
jschoubben ee0b4a271f Plan H: the operator chose the merge through the mesh as the delivery's start (shape A, 01:20)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:20:13 +01:00
jschoubben 6b07fe2041 Plan H: the operator's answers of 01:10 (stale data goes, one take-in one pipeline and none in parallel, hand placement yes; groups still open)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 01:12:20 +01:00
jschoubben 44975de2e1 Plan H: every node's modules, the module graph, the hand placement in the graph's order, no pull request checked today
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 00:57:06 +01:00
jschoubben cb6cabdf5e Plan H: fold the checker's findings in (counts, guesses said as guesses, the medium gaps left out, H4b)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 00:35:54 +01:00
jschoubben bfa93e5f48 Plan design 47's milestone H: deliver the build-and-delivery system properly
The system cannot deliver itself: wants held, records never ended, two
models side by side. The plan decides one model, fix in place with the
indexer's take-in written again, one clean state by hand, milestones
H0 to H5 and the guards. Two agent definitions (spec-reader,
gap-surveyor) carried the readings it is written from.

Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-20 00:31:29 +01:00
mesh-tools 50c9123d9e Merge pull request 'Design 47 G: the lead's answer A-490m (a runner rereads its seat list for a defined seat's unnamed verb)' (#736) from design/47-G-lead-answers-144 into main 2026-11-19 22:05:48 +00:00
jschoubben f081c9a23b Answer A-490m: a runner rereads its seat list once for a defined seat's unnamed verb; undefined seats and module tools ask as before
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 23:05:32 +01:00
mesh-tools 1c75f01a1c Merge pull request 'Design 47 G: the lead's answer A-490l (mesh-config's wake-ups and its one timer)' (#735) from design/47-G-lead-answers-143 into main 2026-11-19 22:04:44 +00:00
jschoubben a09270d6b8 Answer A-490l: mesh-config wakes on its inputs alone; one timer for its real deadlines; a proposal naming a pull request is refused
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 23:04:25 +01:00
mesh-tools bbeb890399 Merge pull request 'Design 47 G: the lead's answer A-490k (the catch-up says take-in.taken; a runner behind asks again)' (#734) from design/47-G-lead-answers-142 into main 2026-11-19 22:03:56 +00:00
jschoubben ccacffdc90 Answer A-490k: the indexer's catch-up says take-in.taken with its entries; a runner behind on its view asks again on the next event
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 23:03:39 +01:00
mesh-tools 0928ed52be Merge pull request 'mesh-watcher: read the working program's journal each round; a stuck read is a finding' (#733) from fix/watcher-reads-the-journal into main 2026-11-19 20:33:09 +00:00
jschoubben a6f811fce0 mesh-watcher: read the working program's journal each round; a stuck read is a finding; never guess the time
A watcher reported no change for ten minutes while the sdk's build had failed at its start (2026-11-19 21:19): it read only states that do not show a failed build, and read around a verb that timed out on every call.

Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 21:32:59 +01:00
mesh-tools 39f7ecce8d Merge pull request 'Design 47 G: the lead's answer A-490j (a holder takes a publish while its build runs)' (#732) from design/47-G-lead-answers-141 into main 2026-11-19 20:32:37 +00:00
jschoubben 0b3974d773 Answer A-490j: a holder takes a publish while its build runs when the reservation names no digest yet
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 21:32:27 +01:00
mesh-tools 9c39e6be55 Merge pull request 'Design 47 G: the lead's answer A-490i (a floor read is a kept call; no timer in the indexer)' (#731) from design/47-G-lead-answers-140 into main 2026-11-19 16:34:03 +00:00
jschoubben bbc65d5e7a Answer A-490i: a floor read is a kept call of mesh-sdk's form, no timer in the indexer (corrects A-490h)
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 17:33:51 +01:00
mesh-tools 43dc902a0c Merge pull request 'Design 47 G: the lead's answers A-490g and A-490h (the floor reads every package's holder; an unanswered floor read is asked again)' (#730) from design/47-G-lead-answers-139 into main 2026-11-19 16:18:08 +00:00
jschoubben adbf113aa8 Answer A-490g and A-490h: the floor reads every package's holder too; an unanswered floor read is asked again and said
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 17:17:57 +01:00
mesh-tools 514eeb2a2e Merge pull request 'Design 47 G: the lead's answer A-490f (the tag floor at start and every revision meets A-490d)' (#729) from design/47-G-lead-answers-138 into main 2026-11-19 16:11:49 +00:00
jschoubben bdd86986de Answer A-490f: the tag floor filled at start and at every revision meets A-490d
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 17:11:38 +01:00
mesh-tools cff74e8419 Merge pull request 'Design 47 G: the lead's answers A-490c..A-490e (the catalog's tag floor)' (#728) from design/47-G-lead-answers-137 into main 2026-11-19 16:05:50 +00:00
jschoubben 4ac1cfb0b1 Answer A-490c..A-490e: the catalog's tag floor reads every tag of a module's form, per module with no version, in its own repository
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 17:05:38 +01:00
mesh-tools 9ae68e0dc0 Merge pull request 'Design 47 G: the lead's answers A-490a and A-490b (where the package holders' publish cases live; a rebuild reserves afresh)' (#727) from design/47-G-lead-answers-136 into main 2026-11-19 16:03:14 +00:00
jschoubben 1dd4daf184 Answer A-490a and A-490b: where the package holders' publish cases live; a rebuild reserves its version afresh
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 17:02:43 +01:00
mesh-tools 9bde22296e Merge pull request 'Design 47 G: the lead's answer A-490 (a package holder takes the no-parts publish as tag-or-refuse)' (#726) from design/47-G-lead-answers-135 into main 2026-11-19 15:44:47 +00:00
jschoubben 325bf58483 Answer A-490: a package holder takes the no-parts publish as tag-or-refuse and never reads the artifact store
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 16:44:24 +01:00
mesh-tools 9ad9511791 Merge pull request 'Design 47 G: the lead's answer A-489c (the build modules' kit also holds that a puller counts the runs that started)' (#725) from design/47-G-lead-answers-134 into main 2026-11-19 14:45:19 +00:00
jschoubben ef87d14e27 Answer A-489c: the build modules' kit also holds that a puller counts the runs that started
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 15:45:04 +01:00
mesh-tools d28cabd28e Merge pull request 'Design 47 G: the lead's answers A-489a and A-489b (every end of an ask is written, said, acknowledged; a build module that cannot say a step's end fails its own test)' (#724) from design/47-G-lead-answers-133 into main 2026-11-19 14:36:41 +00:00
jschoubben 7d2c80b9c0 Answer A-489a and A-489b: every end of an ask is written, said and acknowledged; a build module that cannot say a step's end fails its own test
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 15:36:25 +01:00
mesh-tools eb0e67753f Merge pull request 'Design 47 G: the lead's answer A-489 (an ask a build module ends itself is acknowledged once its end is written)' (#723) from design/47-G-lead-answers-132 into main 2026-11-19 14:22:08 +00:00
jschoubben 1e309472da Answer A-489: an ask a build module ends itself is acknowledged once its end is written, never terminated
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 15:21:32 +01:00
mesh-tools 420f2fd9c1 Merge pull request 'Design 47 G: the lead's answer A-488b (toolchain keyed by the artifact's language; every build module declares its seat-list reader's needs)' (#722) from design/47-G-lead-answers-131 into main 2026-11-19 12:47:56 +00:00
jschoubben 57c1404701 Answer A-488b: the toolchain stays keyed by the artifact's language; every build module declares what its seat-list reader uses
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 13:47:33 +01:00
mesh-tools 686bff9021 Merge pull request 'Design 47 G: the lead's answers A-488 and A-488a (a step ask names its seat's language; a seat-list follower consumes its events)' (#721) from design/47-G-lead-answers-130 into main 2026-11-19 12:40:26 +00:00
jschoubben 25df48ad70 Answer A-488 and A-488a: a step ask names its build seat's language; a module following the seat list consumes its events
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 13:40:10 +01:00
mesh-tools 96be83fcfc Merge pull request 'Design 47 G: the lead's answer A-487d (a read of the newest revision names the newest again while it moved on)' (#720) from design/47-G-lead-answers-129 into main 2026-11-19 11:16:58 +00:00
jschoubben 61bf1215e7 Answer A-487d: a read of the newest revision names the newest again while it moved on, bounded by progress
Claude-Session: https://claude.ai/code/session_01JG73KHUnBze48rdua8u2sy
2026-11-19 12:16:48 +01:00
mesh-tools 6938f6053d Merge pull request 'Design 47 G: the lead's answer A-487c (which catalog verbs answer from the view; the subject of catalog-revision)' (#719) from design/47-G-lead-answers-128 into main 2026-11-19 00:17:36 +00:00
jschoubben c1bd8d113f Answer A-487c: which catalog verbs answer from the view, and the subject a reader of catalog-revision subscribes to 2026-11-19 01:17:25 +01:00
mesh-tools 2378f3f6c8 Merge pull request 'Design 47 G: the lead's answer A-487b (the catalog half reads its newest revision from its own objects)' (#718) from design/47-G-lead-answers-127 into main 2026-11-19 00:03:35 +00:00
jschoubben 51e3c7b4d8 Answer A-487b: the catalog half reads its newest revision from its own objects, never by asking its own seat 2026-11-19 01:03:21 +01:00
mesh-tools b37b330221 Merge pull request 'Design 47 G: the lead's answers A-487 and A-487a (the catalog answers from memory; a node-engine asks a missing trunk build again on a backoff)' (#717) from design/47-G-lead-answers-126 into main 2026-11-18 23:54:00 +00:00
jschoubben 532aa1a379 Answer A-487: the catalog answers from its newest revision in memory; a node-engine asks a missing trunk build again on a backoff 2026-11-19 00:53:04 +01:00
mesh-tools 80d44d8547 Merge pull request 'Design 47 G: the lead's answer A-486 (a tool for every pipeline-runner verb no tool reaches)' (#716) from design/47-G-lead-answers-125 into main 2026-11-18 01:22:20 +00:00
jschoubben 3868280be2 Answer A-486: a tool for every pipeline-runner verb no tool reaches 2026-11-18 02:21:59 +01:00
mesh-tools 8f9954bc5d Merge pull request 'Design 47 G: the lead's answer A-485 (container module, a process naming no user, for ADR 0680 §3)' (#715) from design/47-G-lead-answers-124 into main 2026-11-18 00:21:45 +00:00
jschoubben ca8e13a306 Answer A-485: what a container module is and how a process naming no user runs, for ADR 0680 §3 2026-11-18 01:21:30 +01:00
mesh-tools 5d9961a789 Merge pull request 'Design 20: follow ADR 0680 (a process's own secrets belong to the account it runs as)' (#714) from design/20-adr-0680 into main 2026-11-18 00:09:45 +00:00
jschoubben 180a096a3e Follow ADR 0680 in design 20: a process's own secrets belong to the account it runs as
ADR 0436 §1's ban on root and the operator account for a bus process ended,
so design 20 now says whose own secrets are, the validator's two rules in
their place and how each is checked; zfs-owns-zfs A-37 notes its premise ended.
2026-11-18 01:09:07 +01:00
mesh-tools 9581cd5e26 Merge pull request 'Design 47 G: the lead's answer A-484a (the running statement's manifest is the second source)' (#713) from design/47-G-lead-answers-123 into main 2026-11-17 23:47:40 +00:00
jschoubben e69b81b8ea Answer A-484a: the running statement's manifest is the second source; a composition answers only kept data 2026-11-18 00:47:21 +01:00
mesh-tools 2c95a12b8f Merge pull request 'Design 47 G: the lead's answer A-484 (one place answers the manifest of the build a module runs)' (#712) from design/47-G-lead-answers-122 into main 2026-11-17 23:33:09 +00:00
jschoubben 4b8ef634ba Answer A-484: one place answers the manifest of the build a module runs 2026-11-18 00:32:49 +01:00
mesh-tools b5d19f3972 Merge pull request 'Design 47 G: the lead's answer A-483 (a running build with no entry keeps what its running composition keeps)' (#711) from design/47-G-lead-answers-121 into main 2026-11-17 23:22:51 +00:00
jschoubben 7225f752c6 Answer A-483: a running build with no entry keeps what its running composition keeps 2026-11-18 00:22:36 +01:00
mesh-tools 4eabfba370 Merge pull request 'Design 47 G: the lead's answer A-482 (a running build is one the node's own record says runs)' (#710) from design/47-G-lead-answers-120 into main 2026-11-17 22:42:54 +00:00
jschoubben bb04c5b089 Answer A-482: a running build is one the node's own record says runs 2026-11-17 23:42:44 +01:00
mesh-tools a3bb692725 Merge pull request 'Design 47 G: the lead's answer A-481a (the composition names its templates archives in a field of its own)' (#709) from design/47-G-lead-answers-119 into main 2026-11-17 22:40:08 +00:00
jschoubben 56b3cd6bc3 Answer A-481a: the composition names its templates archives in a field of its own 2026-11-17 23:39:54 +01:00
mesh-tools 7de86e7af9 Merge pull request 'Design 47 G: the lead's answer A-481 (a templates archive the composer read counts as a fetched part)' (#708) from design/47-G-lead-answers-118 into main 2026-11-17 22:38:06 +00:00
jschoubben 2359aea5e0 Answer A-481: a templates archive the composer read counts as a fetched part 2026-11-17 23:37:29 +01:00
mesh-tools 91d0ba10a2 Merge pull request 'Design 47 G: the lead's answer A-480a (a held module with no readable running manifest provides what its trunk manifest provides)' (#707) from design/47-G-lead-answers-117 into main 2026-11-17 22:06:42 +00:00
jschoubben 5502609780 Answer A-480a: a held module whose running manifest cannot be read provides what its trunk manifest provides 2026-11-17 23:06:28 +01:00
mesh-tools 89865f27d8 Merge pull request 'Design 47 G: the lead's answer A-480 (a held module still provides what its running build provides)' (#706) from design/47-G-lead-answers-116 into main 2026-11-17 21:53:43 +00:00
jschoubben ff43506156 Answer A-480: a held module still provides what its running build provides 2026-11-17 22:53:29 +01:00
mesh-tools a74591dc71 Merge pull request 'Design 47 G: the lead's answer A-478b (the order waits of one machine's apply stay)' (#705) from design/47-G-lead-answers-115 into main 2026-11-17 21:15:46 +00:00
jschoubben 1942c23c4a Answer A-478b: the order waits of one machine's apply stay beside the want rule 2026-11-17 22:15:30 +01:00
mesh-tools 580678e948 Merge pull request 'Design 47 G: the lead's answer A-478 (the assigner wants the newest released version that passed)' (#704) from design/47-G-lead-answers-114 into main 2026-11-17 21:05:09 +00:00
jschoubben b6b4e0232c Answer A-478: the assigner wants the newest released version that passed, on every placed machine
The operator's step 5 of the delivery walkthrough, 2026-11-17: the acting-part and
delivery-dependency holds held feh 0.1.8, a module nothing depends on.
2026-11-17 22:04:29 +01:00
mesh-tools dd89b437d2 Merge pull request 'Design 47 G: the lead's answer A-477a (the guard compares the recipe only when named)' (#703) from design/47-G-lead-answers-113 into main 2026-11-17 19:59:00 +00:00
jschoubben b27004f74d Answer A-477a: the build node's guard compares the recipe only when the ask names one 2026-11-17 20:58:49 +01:00
mesh-tools d757200010 Merge pull request 'Design 47 G: the lead's answer A-477 (a runner's step ask names the toolchain's recipe)' (#702) from design/47-G-lead-answers-112 into main 2026-11-17 19:57:58 +00:00
jschoubben b463a1518e Answer A-477: a runner's step ask names the agreed toolchain with its recipe 2026-11-17 20:57:42 +01:00
mesh-tools 21004f50e2 Merge pull request 'Design 47 G: the lead's answer A-476a (code dependents still move)' (#701) from design/47-G-lead-answers-111 into main 2026-11-17 19:21:40 +00:00
jschoubben 0ca76cf54f Answer A-476a: a commit still moves its moved modules' code dependents in any repository 2026-11-17 20:21:30 +01:00
mesh-tools 75857d9eb7 Merge pull request 'Design 47 G: the lead's answer A-476 (the take-in validates nothing, holds for no indexer build, moves only its repository's modules)' (#700) from design/47-G-lead-answers-110 into main 2026-11-17 19:13:31 +00:00
jschoubben d433411fcd Answer A-476: the take-in validates nothing, holds for no indexer build, and moves only its repository's modules 2026-11-17 20:13:20 +01:00
mesh-tools 8f08ef178f Merge pull request 'Design 47 G: the lead's answer A-475a' (#699) from design/47-G-lead-answers-109 into main 2026-11-17 15:17:50 +00:00
jschoubben 2eb4059897 Answer A-475a: the stale count, the catch-up's wait and validate-tries' bounds 2026-11-17 16:17:38 +01:00
mesh-tools fd3d7371bc Merge pull request 'Design 47 G: the lead's answer A-461e (no store address holds only the store's fetchers)' (#698) from design/47-G-lead-answers-108 into main 2026-11-17 15:11:23 +00:00
jschoubben 3982c57e98 Answer A-461e: no store address of any readable kind holds only the store's fetchers 2026-11-17 16:11:12 +01:00
mesh-tools 513920b66b Merge pull request 'Design 47 G: the lead's answer A-461d (a placed module not heard running is laid from its placement)' (#697) from design/47-G-lead-answers-107 into main 2026-11-17 15:03:50 +00:00
jschoubben 945a74b6f2 Answer A-461d: a placed module not heard running is laid from its placement or trunk build 2026-11-17 16:03:36 +01:00
mesh-tools ecf2a05179 Merge pull request 'Design 47 G: the lead's answer A-475 (a validation ask whose request moved is asked again)' (#696) from design/47-G-lead-answers-106 into main 2026-11-17 15:00:26 +00:00
jschoubben f906b7c8b3 Answer A-475: a validation ask whose request moved is asked again, not left could-not-validate 2026-11-17 16:00:12 +01:00
mesh-tools 0347a77bcb Merge pull request 'Design 47 G: the lead's answer A-474 (a module's tests read nothing outside their module)' (#695) from design/47-G-lead-answers-105 into main 2026-11-17 14:11:05 +00:00
jschoubben 16316da868 Answer A-474: a module's tests read nothing outside their module 2026-11-17 15:10:52 +01:00
mesh-tools 284c32062e Merge pull request 'Design 47 G: the lead's answer A-473 (the builder asks the tagged publish when its build ends)' (#694) from design/47-G-lead-answers-104 into main 2026-11-17 13:58:33 +00:00
jschoubben 96dadc6fde Answer A-473: the builder asks the tagged publish of a reserved version when its build ends 2026-11-17 14:58:19 +01:00
mesh-tools cc1c5cd5e1 Merge pull request 'Design 47 G: the lead's answer A-472a' (#693) from design/47-G-lead-answers-103 into main 2026-11-17 13:58:03 +00:00
jschoubben a622019bd5 Answer A-472a: the cancel's note stays; the read bound and revisions kept are settings 2026-11-17 14:57:52 +01:00
mesh-tools fa97929f21 Merge pull request 'Design 47 G: the lead's answer A-472 (a pipeline's end is said without waiting on another part)' (#692) from design/47-G-lead-answers-102 into main 2026-11-17 13:49:42 +00:00
jschoubben 76c92e4c1e Answer A-472: a pipeline's end is said without waiting on another part 2026-11-17 14:49:36 +01:00
mesh-tools bdb4efc160 Merge pull request 'Design 47 G: the lead's answer A-471 (a toolchain is agreed by its language version)' (#691) from design/47-G-lead-answers-101 into main 2026-11-17 13:48:18 +00:00
jschoubben 68148d886b Answer A-471: a toolchain is agreed by its language version, the patch not counted 2026-11-17 14:47:57 +01:00
mesh-tools 304ae5a5fb Merge pull request 'Design 47 G: the lead's answer A-470u (a reader that leaves has its wait ended)' (#690) from design/47-G-lead-answers-100 into main 2026-11-17 12:08:34 +00:00
jschoubben aa52fa5345 Answer A-470v: the search blames every refused module, not only the first 2026-11-17 13:08:28 +01:00
jschoubben e671d865d3 Answer A-470u: a reader that leaves or is replaced has its wait ended 2026-11-17 13:08:08 +01:00
mesh-tools 62b03ab4cd Merge pull request 'Design 47 G: the lead's answer A-470t (the runner's ends carry no message id)' (#689) from design/47-G-lead-answers-99 into main 2026-11-17 12:06:11 +00:00
jschoubben 5edbcdac6c Answer A-470t: the runner's ends carry no message id, so A-470q holds as built 2026-11-17 13:06:02 +01:00
mesh-tools 19e7da66d8 Merge pull request 'Design 47 G: the lead's answer A-470s (only build seats are judged for a holder)' (#688) from design/47-G-lead-answers-98 into main 2026-11-17 12:04:03 +00:00
jschoubben c1c7353191 Answer A-470s: only the build seats are judged for a holder at placing 2026-11-17 13:03:55 +01:00
mesh-tools 59cf3c25fd Merge pull request 'Design 47 G: the lead's answers A-470q, A-470r (an end said again has its own message id; a drop ends only owed ends)' (#687) from design/47-G-lead-answers-97 into main 2026-11-17 12:03:41 +00:00
jschoubben cd3428755a Answer A-470q and A-470r: an end said again has its own message id, and a drop ends only the ends owed 2026-11-17 13:03:28 +01:00
mesh-tools 7f846906ee Merge pull request 'Design 47 G: the lead's answer A-470p (a run's end before every other act)' (#686) from design/47-G-lead-answers-96 into main 2026-11-17 11:33:04 +00:00
jschoubben 42e9484261 Answer A-470p: a run's end comes before every other act of its step-ended 2026-11-17 12:32:55 +01:00
mesh-tools 83fab59cac Merge pull request 'Design 47 G: the lead's answer A-470o (an expiry judged by the key's absence)' (#685) from design/47-G-lead-answers-95 into main 2026-11-17 11:25:28 +00:00
jschoubben 384ceb714d Answer A-470o: an expiry is judged by the key's absence, and a run's end is never held behind a finding 2026-11-17 12:25:15 +01:00
mesh-tools 9ea4b009ed Merge pull request 'Design 47 G: the lead's answer A-470n (a run's bound names its ask)' (#684) from design/47-G-lead-answers-94 into main 2026-11-17 11:24:14 +00:00
jschoubben 72bc73df27 Answer A-470n: a run's bound names its ask, so a replaced run's late signal ends nothing 2026-11-17 12:24:05 +01:00
mesh-tools 60e23053c3 Merge pull request 'Design 47 G: the lead's answer A-470m (a reader's end is kept in the key)' (#683) from design/47-G-lead-answers-93 into main 2026-11-17 11:16:54 +00:00
jschoubben d71cbe8b49 Answer A-470m: a reader's end is kept in the key, so a take-up tells a passed reader from a late one 2026-11-17 12:16:40 +01:00
mesh-tools c79c69bfff Merge pull request 'Design 47 G: the lead's answer A-470l (a take-up says no end again for a reader that passed)' (#682) from design/47-G-lead-answers-92 into main 2026-11-17 11:14:54 +00:00
jschoubben 93a3e46923 Answer A-470l: a take-up says no end again for a reader that passed 2026-11-17 12:14:43 +01:00
mesh-tools b269b5f3fa Merge pull request 'Design 47 G-23: causes found on 2026-11-17' (#681) from design/47-G-after-delivery-notes-1117b into main 2026-11-17 11:12:14 +00:00
jschoubben c6943ee76f G-23: the causes found on 2026-11-17 for the assigner's long line, the toolchain disagreement and an untaken merge 2026-11-17 12:12:05 +01:00
mesh-tools 6e43789deb Merge pull request 'Design 47 G: the lead's answer A-470k (the case carries the revision's none)' (#680) from design/47-G-lead-answers-91 into main 2026-11-17 10:59:29 +00:00
jschoubben ff60461f5b Answer A-470k: the case carries the revision's none, so the composer can tell a build that can never be made 2026-11-17 11:58:59 +01:00
mesh-tools 3608c10b5c Merge pull request 'Design 47 G: the lead's answer A-470j (a module that can never be built is blamed)' (#679) from design/47-G-lead-answers-90 into main 2026-11-17 10:56:51 +00:00
jschoubben 10d6d75902 Answer A-470j: a module that can never be built is blamed, not a reason to judge no machine 2026-11-17 11:56:06 +01:00
mesh-tools 16c2f1339c Merge pull request 'Design 47 G: the lead's answer A-470i (a fill names the cap that refuses)' (#678) from design/47-G-lead-answers-89 into main 2026-11-17 10:55:33 +00:00
jschoubben e2e464a046 Answer A-470i: a fill names the cap that refuses, answered by the process that holds it 2026-11-17 11:55:24 +01:00
mesh-tools 8a45d966bd Merge pull request 'Design 47 G: the lead's answer A-470h (a holderless seat's pipeline asks none of its steps)' (#677) from design/47-G-lead-answers-88 into main 2026-11-17 10:54:25 +00:00
jschoubben 8231fd0acb Answer A-470h: a pipeline failed for a holderless seat asks none of its steps 2026-11-17 11:54:15 +01:00
mesh-tools 14a0c6c44f Merge pull request 'Design 47 G: the lead's answer A-470g (no holder fails at once; fill against the cap that refuses)' (#676) from design/47-G-lead-answers-87 into main 2026-11-17 10:51:01 +00:00
jschoubben bb99034295 Answer A-470g: no holder fails at once, and a fill is said against the cap that refuses 2026-11-17 11:50:50 +01:00
mesh-tools fd8c4aaeb7 Merge pull request 'Design 47 G: the lead's answer A-470f (the run drops refuse no caller)' (#675) from design/47-G-lead-answers-86 into main 2026-11-17 10:48:32 +00:00
jschoubben 12a3ef6960 Answer A-470f: the run drops refuse no caller, as every seat verb admits every caller 2026-11-17 11:48:23 +01:00
mesh-tools 79181abfa7 Merge pull request 'Design 47 G: the lead's answers A-470d, A-470e (a rerun carries every reader; a newer run is told by its ask)' (#674) from design/47-G-lead-answers-85 into main 2026-11-17 10:39:29 +00:00
jschoubben 520e95a10c Answer A-470d and A-470e: a rerun carries every reader with its mark, and a newer run is told by its ask 2026-11-17 11:39:20 +01:00
mesh-tools 12b7e246e4 Merge pull request 'Design 47 G: the lead's answers A-470a..A-470c (aging test runs; an untried module never fails; the composer's readings)' (#673) from design/47-G-lead-answers-84 into main 2026-11-17 10:24:26 +00:00
jschoubben 328e17112e Answer A-470a..A-470c: aging test runs, an untried module never fails, the composer's four readings 2026-11-17 11:24:13 +01:00
mesh-tools 36b66c0b54 Merge pull request 'Design 47 G: the lead's answers A-469w..A-470 (caps, settings, waits, bound keys)' (#672) from design/47-G-lead-answers-83 into main 2026-11-17 10:22:16 +00:00
jschoubben f62a73d757 Answer A-469w..A-470: the assigner developer's five questions on caps, settings, waits and bound keys 2026-11-17 11:22:06 +01:00
mesh-tools ff464265b9 Merge pull request 'Design 47 G-23: the faults met on 2026-11-17' (#671) from design/47-G-after-delivery-notes-1117 into main 2026-11-17 10:17:35 +00:00
jschoubben 8884fc9ee7 G-23: the faults met on 2026-11-17, so none is lost before G is delivered 2026-11-17 11:17:23 +01:00
mesh-tools c94cbff169 Merge pull request 'Design 47 G: the lead's answer A-469v (a drop's answer; named run definitions)' (#670) from design/47-G-lead-answers-82 into main 2026-11-17 10:11:37 +00:00
jschoubben 2ed3c5d368 Answer A-469v: a drop's answer names the run and the readers it ended apart, so C1 can go on 2026-11-17 11:11:27 +01:00
mesh-tools 5880f85e5b Merge pull request 'Design 47 G: the lead's answers A-469t, A-469u (one shared form; the four run verbs)' (#669) from design/47-G-lead-answers-81 into main 2026-11-17 10:10:14 +00:00
jschoubben 9bfb8b0211 Answer A-469t and A-469u: one shared form, and the four run verbs' names and answers, so C1 can go on 2026-11-17 11:10:03 +01:00
mesh-tools 9c23309b38 Merge pull request 'Design 47 G: the lead's answers A-468..A-469s (G-24's prepare: 14 design questions, 20 design-meets-code)' (#668) from design/47-G-lead-answers-80 into main 2026-11-17 10:04:16 +00:00
jschoubben e88cce45dd Answer A-468..A-469s: G-24's 14 design questions and 20 design-meets-code entries, so its develop can start 2026-11-17 11:04:04 +01:00
mesh-tools be550d1790 Merge pull request 'Design 47 G: the lead's answer A-467a (a composition step with no machine judged never passes)' (#667) from design/47-G-lead-answers-79 into main 2026-11-17 09:48:03 +00:00
jschoubben 92f967f542 Answer A-467a: a composition step with no machine judged never passes 2026-11-17 10:47:50 +01:00
mesh-tools c2aae05b59 Merge pull request 'Design 47 G: one composition run per candidate and machine, one test run per module's packages (G-24)' (#666) from design/47-G-shared-runs into main 2026-11-17 09:41:41 +00:00
jschoubben cbf12d3380 Merge main (A-467) into the shared-runs design 2026-11-17 10:41:23 +01:00
jschoubben 849a4012d6 Design 47 G: read "no holder" from the seat-holders view, not the seat list
The seat list answers definitions alone (§527); the holders are the seat mesh-module-assigner's
view seat-holders, which the runner now consumes (A-466z).
2026-11-17 10:39:47 +01:00
jschoubben 32302e0bd1 Design 47 G: test 109's line in G-24 names the queued run A-466y keeps 2026-11-17 10:38:51 +01:00
jschoubben 017e20504a Design 47 G: read "no holder" from the seat list, never from five minutes in the queue
A test run queued behind the build capacity with its holder present was ended failed for no
holder after five minutes, which is not true; the queue is §86's finding's and the untaken
bound's to time (A-466y).
2026-11-17 10:38:35 +01:00
jschoubben d7321df3e5 Design 47 G: write in round 2 of the review of the composition runs and test runs
A run's bound times owed progress from the ask's taking, a run never taken ends at a bound of
its own, blame-tries travels in the request, the hand order goes, a test run's output lives
with its key, stands names what a reader waits on, and G-24 reads in number order
(A-466p to A-466x).
2026-11-17 10:37:41 +01:00
mesh-tools e82887d67c Merge pull request 'Design 47 G: the lead's answer A-467 (no composition passes on a machine it did not judge)' (#665) from design/47-G-lead-answers-78 into main 2026-11-17 09:35:27 +00:00
jschoubben 605e0878a5 Answer A-467: a composition step never passes on a machine it did not judge; seats from the trunk while the revision names none 2026-11-17 10:35:05 +01:00
jschoubben cebac2e791 Design 47 G: count both run bounds from the ask's taking, 30 minutes
A bound counted from the asking would end a run that only waited in the queue as could not
run, which is not true; the queue is the build-capacity finding's to time (A-466n).
2026-11-17 10:33:58 +01:00
jschoubben 661fb8a742 Design 47 G: name the runs' asks where they are known, bound the runs, time every reader
The lead's answers A-466m to A-466o: the assigner names the composition runs' asks in its
step-progress, each run's bound is a setting, and since the watchman on main times only a
build seat's running ask, it now times every pipeline step that reads a run, so a reader no
part ends is said as pipeline-stands.
2026-11-17 10:33:22 +01:00
jschoubben 6b7adc5ec2 Design 47 G: write in the review of the composition runs and test runs
The four reviewers by architectural domain found results crossing candidates, a read of a
build seat's state, fixed durations and caps, unstated seat contract fields and a rerun rule
that differed between the two kinds; the lead's answers A-466 to A-466l settle each. The
lines move to a file of their own, G-24, so G-2 stays under its bound. The split of the main
file and G-22 waits until after G, as the operator decided at 10:31 CET.
2026-11-17 10:31:28 +01:00
jschoubben 77318ca60e Design 47 G: a test run asked again moves every step that reads it; two items after G
The coordinator's answers of 2026-11-17: a step left reading a failed result would say what
is no longer true, so a rerun of a test run moves all its readers; the node-engine's
differing input bytes and the split of design 47's main file wait until after G.
2026-11-17 10:25:15 +01:00
jschoubben 31cd62aeec Design 47 G: one composition run per candidate and machine, one test run per tested packages
G's live delivery of 2026-11-17 composed each machine once per pipeline (96 runs for one
revision, 80 more on rerun, the assigner's inputs state at its cap twice) and ran the same
tests once per artifact. Pipelines of one candidate now read one composition run per
machine, kept by the assigner, whose verdict names the modules it blames so one module's
fault fails only its own pipelines; test steps naming the same packages read one test run,
kept by the runner. The operator, 2026-11-17 about 10:20 CET: "In G, now".
2026-11-17 10:24:28 +01:00
mesh-tools 2407183c70 Merge pull request 'Design 47 G: the lead's answer A-465b (the inputs key form is the assigner's own)' (#664) from design/47-G-lead-answers-77 into main 2026-11-17 08:46:28 +00:00
jschoubben 399ac934f6 Answer A-465b: the inputs key form is the assigner's own; an ask's key holds its time; old keys age out 2026-11-17 09:46:15 +01:00
mesh-tools a550fb9256 Merge pull request 'Design 47 G: the lead's answer A-461c (the golden node grant checked against invokes); after G item 14' (#663) from design/47-G-lead-answers-76 into main 2026-11-17 08:40:06 +00:00
jschoubben ec6aae9997 Answer A-461c: the node-engine's golden grant is checked against its invokes; after G: no pass waits on a call 2026-11-17 09:39:53 +01:00
mesh-tools 1ac43e4ddf Merge pull request 'Design 47 G: the lead's answer A-465a (a composition input kept once per content)' (#662) from design/47-G-lead-answers-75 into main 2026-11-17 08:38:18 +00:00
jschoubben cb1ce7291f Answer A-465a: a composition input is kept once per content, shared by the asks that use it 2026-11-17 09:38:02 +01:00
mesh-tools 8fd85c6768 Merge pull request 'Design 47 G: after G, a full delivery cycle by hand (the operator, 2026-11-17)' (#661) from design/47-G-after-hand-cycle into main 2026-11-17 08:23:07 +00:00
jschoubben 0fc357c96b After G: say catalog, the glossary's word 2026-11-17 09:22:54 +01:00
jschoubben d84b7cbb57 After G: a full delivery cycle by hand, as the operator asked on 2026-11-17 2026-11-17 09:22:37 +01:00
mesh-tools 36d5b50d5b Merge pull request 'Design 47 G: the lead's answer A-465 (the assigner's inputs end with their objects)' (#660) from design/47-G-lead-answers-74 into main 2026-11-17 08:00:16 +00:00
jschoubben 64ff3fd4e9 Answer A-465: the assigner's composition inputs end with their objects, and a drop tool for those left 2026-11-17 09:00:02 +01:00
mesh-tools 2b59494c30 Merge pull request 'Design 47 G: the lead's answer A-464e (the gate's start comparison ends only when every head was compared)' (#659) from design/47-G-lead-answers-73 into main 2026-11-17 07:56:55 +00:00
jschoubben 782f90c003 Answer A-464e: the gate's start comparison ends only when every head was compared 2026-11-17 08:56:43 +01:00
mesh-tools f0a940463b Merge pull request 'Design 47 G: the lead's answers A-463a and A-464d (from the review)' (#658) from design/47-G-lead-answers-72 into main 2026-11-17 01:08:47 +00:00
jschoubben 5385c798c0 Answer A-463a and A-464d: a pending merge is not behind, and every source move is forward 2026-11-17 02:08:39 +01:00
mesh-tools cf08f6e69b Merge pull request 'Design 47 G: the lead's answer A-464c (the gate's source moves a head only forward)' (#657) from design/47-G-lead-answers-71 into main 2026-11-17 01:01:11 +00:00
jschoubben 163fe51f96 Answer A-464c: the gate's source moves a head only forward, and only when it changes 2026-11-17 02:00:59 +01:00
mesh-tools c81a186a3d Merge pull request 'Design 47 G: the lead's answer A-464b (a merge the gate did not make)' (#656) from design/47-G-lead-answers-70 into main 2026-11-17 00:57:23 +00:00
jschoubben acd7f1f588 Answer A-464b: what makes a merge one the gate did not make, and where that is said 2026-11-17 01:57:14 +01:00
mesh-tools 6664193aac Merge pull request 'Design 47 G: the lead's answer A-464a (a read tool for the indexer's takeins state)' (#655) from design/47-G-lead-answers-69 into main 2026-11-17 00:50:39 +00:00
jschoubben d47f617ad5 Answer A-464a: a read tool for the indexer's takeins state, so a hand take-in is seen 2026-11-17 01:50:25 +01:00
mesh-tools 84ef09d4b4 Merge pull request 'Design 47 G: the lead's answer A-464 (the gate's source follows every merge)' (#654) from design/47-G-lead-answers-68 into main 2026-11-17 00:43:10 +00:00
jschoubben 709ad9de59 A-464: name mesh-host in a code span, as the words test asks 2026-11-17 01:42:59 +01:00
jschoubben a4d7d44cd0 Answer A-464: the gate's source follows every merge into a watched branch 2026-11-17 01:42:43 +01:00
mesh-tools 47332934b8 Merge pull request 'Design 47 G: the lead's answer A-463 (the indexer takes in a merge it never took in)' (#653) from design/47-G-lead-answers-67 into main 2026-11-17 00:40:25 +00:00
jschoubben b6c99f171d Answer A-463: the indexer takes in a merge it never took in, and says so 2026-11-17 01:40:16 +01:00
mesh-tools e34c74c2f8 Merge pull request 'Design 47 G: A-459b withdraws A-459 and A-459a; A-462b' (#652) from design/47-G-lead-answers-66 into main 2026-11-17 00:32:53 +00:00
jschoubben f44b0fc90b Answer A-459b (withdraw A-459, A-459a: rule 7 reads the node's own composition) and A-462b 2026-11-17 01:32:44 +01:00
mesh-tools c91aafb6dd Merge pull request 'Design 47 G: the lead's answer A-457b (republish-max-versions goes)' (#651) from design/47-G-lead-answers-65 into main 2026-11-17 00:30:28 +00:00
jschoubben 1209008aba Answer A-457b: the builder's republish-max-versions setting goes with the path it bounded 2026-11-17 01:30:16 +01:00
mesh-tools 25449d1d4f Merge pull request 'Design 47 G: the lead's answer A-456b (the release record is bounded; a publish item ends with its release)' (#650) from design/47-G-lead-answers-64 into main 2026-11-17 00:24:02 +00:00
jschoubben dcaad0c6a4 Answer A-456b: the release record is bounded per module; a publish item ends with its release or its reservation 2026-11-17 01:23:38 +01:00
mesh-tools 66b3d0c7f6 Merge pull request 'Design 47 G: correct A-459a (composed is followed through the declared read)' (#649) from design/47-G-lead-answers-63 into main 2026-11-17 00:22:12 +00:00
jschoubben 92929d55ef Correct A-459a: rule 7 follows composed through the read the assigner's manifest declares 2026-11-17 01:22:00 +01:00
mesh-tools 0bb30e6bc9 Merge pull request 'Design 47 G: the lead's answer A-461b (an absent placed build is the trunk build)' (#648) from design/47-G-lead-answers-62 into main 2026-11-17 00:18:06 +00:00
jschoubben e69d2693de Answer A-461b: an absent placed build is the trunk build from mesh-catalog's module verb 2026-11-17 01:17:57 +01:00
mesh-tools b90489afc7 Merge pull request 'Design 47 G: the lead's answer A-460a (ace's refusal was a hand act; no retry added)' (#647) from design/47-G-lead-answers-61 into main 2026-11-17 00:11:52 +00:00
jschoubben bca45ea0df Answer A-460a: ace's refusal was a hand act; a failed install is tried again when its want changes 2026-11-17 01:11:39 +01:00
mesh-tools f48d5f5091 Merge pull request 'Design 47 G: the lead's answers A-456a, A-457a, A-461a' (#646) from design/47-G-lead-answers-60 into main 2026-11-17 00:11:21 +00:00
jschoubben 2ff96d1b7e Answer A-461a: another node's module is laid from the assignment's placed build 2026-11-17 01:11:09 +01:00
jschoubben e68d3a892e Answer A-456a and A-457a: a released build counts as reserved; the publish retries on the builder's own re-ask state 2026-11-17 01:10:59 +01:00
mesh-tools de9d3378af Merge pull request 'Design 47 G: the lead's answer A-462a (no entry in the followed catalog; the assigner asks the indexer nothing)' (#645) from design/47-G-lead-answers-59 into main 2026-11-17 00:10:20 +00:00
jschoubben 547006dd59 Answer A-462a: no entry means none in the followed catalog; the assigner asks the indexer nothing 2026-11-17 01:10:07 +01:00
mesh-tools 2945e57a4a Merge pull request 'Design 47 G: the lead's answer A-459a (rule 7 hears composed per module and fingerprint)' (#644) from design/47-G-lead-answers-58 into main 2026-11-17 00:09:49 +00:00
jschoubben fc5ec32b90 Answer A-459a: rule 7 hears composed per module and fingerprint from the node-engine's events 2026-11-17 01:09:38 +01:00
mesh-tools 242bae0d0c Merge pull request 'Design 47 G: the lead's answers A-454a, A-456..A-462 (the delivery path after a passed pipeline)' (#643) from design/47-G-lead-answers-57 into main 2026-11-17 00:05:31 +00:00
jschoubben 928aa7e5ba Answer A-454a, A-456..A-462: close the delivery path's six gaps after a passed pipeline 2026-11-17 01:05:14 +01:00
mesh-tools 4d4815c83d Merge pull request 'Design 47 G: the lead's answer A-455 (a stale reservation is taken over; builder_reserve_done)' (#642) from design/47-G-lead-answers-56 into main 2026-11-16 23:50:58 +00:00
jschoubben 975f83f045 Answer A-455: a newer trunk build takes over a stale reservation; builder_reserve_done for a done build 2026-11-17 00:50:50 +01:00
mesh-tools acc54c16b9 Merge pull request 'Design 47 G: correct A-453a (a held have has not moved past its hold)' (#641) from design/47-G-lead-answers-55 into main 2026-11-16 23:49:01 +00:00
jschoubben 77cf51b4c7 Correct A-453a: 'at another fingerprint' applies only to a have that is not held 2026-11-17 00:48:47 +01:00
mesh-tools 5f90d599c6 Merge pull request 'Design 47 G: the lead's answer A-454 (an upstream image is no part of the want)' (#640) from design/47-G-lead-answers-54 into main 2026-11-16 23:47:36 +00:00
jschoubben 987283dd30 Answer A-454: an upstream image a manifest pins is no part of the want 2026-11-17 00:47:28 +01:00
mesh-tools 0da67d6323 Merge pull request 'Design 47 G: the lead's answer A-453a (a have mid-move is not 'no longer held')' (#639) from design/47-G-lead-answers-53 into main 2026-11-16 23:46:13 +00:00
jschoubben 2ecb831355 Answer A-453a: a have mid-move is not 'no longer held' 2026-11-17 00:46:04 +01:00
mesh-tools 59d9f397e2 Merge pull request 'Design 47 G: the lead's answer A-453 (a backup ask only for a failed backup)' (#638) from design/47-G-lead-answers-52 into main 2026-11-16 23:37:41 +00:00
jschoubben 7b1607d933 Answer A-453: the operator is asked about a backup only when a backup failed 2026-11-17 00:37:32 +01:00
mesh-tools bfc7272b3c Merge pull request 'Design 47 G after delivery: retire held/kept for a module left on its build; hollow composition pass' (#637) from design/47-G-after-words into main 2026-11-16 23:17:48 +00:00
jschoubben 37fdc292f7 After G: retire held and kept for a module left on its build; a composition that judged nothing is no pass 2026-11-17 00:17:38 +01:00
mesh-tools 21ab7b1bdf Merge pull request 'Design 47 G: the lead's answer A-452 (an unanswered requirement costs its consumer, not the machine)' (#636) from design/47-G-lead-answers-51 into main 2026-11-16 23:15:03 +00:00
jschoubben e17cffa59e Answer A-452: an unanswered requirement costs its consumer, not the machine 2026-11-17 00:14:46 +01:00
mesh-tools dcdfbc12a6 Merge pull request 'Design 47 G: the lead's answer A-451 (composition with a node-engine not heard running)' (#635) from design/47-G-lead-answers-50 into main 2026-11-16 13:53:44 +00:00
jschoubben 7dd6563ab5 Answer A-451: a machine whose node-engine is not heard running composes with the revision's node-engine entry 2026-11-16 14:53:14 +01:00
mesh-tools eb5cc23a86 Merge pull request 'Design 47 G: the lead's answer A-237c (inline input stays for the checker until G)' (#634) from design/47-G-lead-answers-49 into main 2026-11-16 13:47:23 +00:00
jschoubben d07de2840d Answer A-237c: the token-less inline input stays for the checker until G removes it 2026-11-16 14:47:11 +01:00
mesh-tools 2ab0507440 Merge pull request 'Design 47 G: the lead's answer A-237b (compose-candidate sends its input in chunks)' (#633) from design/47-G-lead-answers-48 into main 2026-11-16 13:35:32 +00:00
jschoubben c47f9ea48e Answer A-237b: compose-candidate sends its input in chunks 2026-11-16 14:35:22 +01:00
mesh-tools dd462aa1ff Merge pull request 'Design 47 G: the lead's answer A-450 (the node-engine's account in the catalog's caller check)' (#632) from design/47-G-lead-answers-47 into main 2026-11-16 13:12:41 +00:00
jschoubben c0736710e8 Answer A-450: the node-engine's account counts as the module holding node-engine for the catalog's caller check 2026-11-16 14:12:28 +01:00
mesh-tools bedc92e251 Merge pull request 'Design 47 G: the lead's answer A-226c (a run's key reads candidates.run-<id>)' (#631) from design/47-G-lead-answers-46 into main 2026-11-16 12:59:58 +00:00
jschoubben d93dddf5b3 Answer A-226c: a run's key reads candidates.run-<id> 2026-11-16 13:59:47 +01:00
mesh-tools 2a52d91cc6 Merge pull request 'Design 47 G: the lead's answer A-226b (compose-candidate reads a revision key and a run key)' (#630) from design/47-G-lead-answers-45 into main 2026-11-16 12:55:03 +00:00
jschoubben ecaf070846 Answer A-226b: compose-candidate reads a branch revision key and a run key too 2026-11-16 13:54:52 +01:00
mesh-tools 2631a11447 Merge pull request 'Principles 11 and 12, and the reviewers ask them' (#629) from principles/configurable-and-own-files into main 2026-11-16 12:53:40 +00:00
jschoubben 39f7ae7203 Reviewers: a hard-coded value and a file written for another module are findings (principles 11 and 12) 2026-11-16 13:53:30 +01:00
jschoubben 6517bbebf0 Principles 11 and 12: configurable, never hard-coded; a module writes only its own 2026-11-16 13:51:28 +01:00
mesh-tools 4de493a784 Merge pull request 'Design 47 G: the lead's answer A-449a (the indexer pins a take-in's revision for its pipelines)' (#628) from design/47-G-lead-answers-44 into main 2026-11-16 12:44:33 +00:00
jschoubben 79214618f3 Answer A-449a: the indexer pins a take-in's revision for its pipelines when it keeps it 2026-11-16 13:44:23 +01:00
mesh-tools 8fa037074c Merge pull request 'Design 47 G: the lead's answers A-448 and A-449 (a take-in names its revision; a pipeline pins it)' (#627) from design/47-G-lead-answers-43 into main 2026-11-16 12:33:23 +00:00
jschoubben f07c337c22 Answer A-448 and A-449: a take-in names its revision; a pipeline pins it 2026-11-16 13:33:12 +01:00
mesh-tools d595d08042 Merge pull request 'After G: tests against the composed bus, kept events heard or loud, one way of asking work' (#626) from design/47-G-after-list-2 into main 2026-11-16 12:01:14 +00:00
jschoubben eaa6de7faf After G: tests against the composed bus, kept events heard or loud, one way of asking work 2026-11-16 13:01:03 +01:00
mesh-tools 2d4b34c51c Merge pull request 'Design 47 G: the lead's answer A-447e (the checker is removed at G, not rebuilt per module)' (#625) from design/47-G-lead-answers-42 into main 2026-11-16 10:39:32 +00:00
jschoubben d4c31d4ff0 Answer A-447e: the checker is removed at G, not rebuilt per module; the no-dir refusal lands with its removal 2026-11-16 11:39:23 +01:00
mesh-tools 71cb172953 Merge pull request 'Design 47 G: the lead's answer A-447d (the checker fetches per changed module)' (#624) from design/47-G-lead-answers-41 into main 2026-11-16 10:38:20 +00:00
jschoubben fa74c6d594 Answer A-447d: the checker fetches per changed module 2026-11-16 11:38:10 +01:00
mesh-tools 97c8c81e82 Merge pull request 'Principle 10: the module is the unit, never the repository' (#623) from principles/module-is-the-unit into main 2026-11-16 10:35:37 +00:00
jschoubben b5e5a35460 Principle 10: a merge moves only the modules it changes 2026-11-16 11:35:33 +01:00
jschoubben 3c0389f74e Principle 10: the module is the unit, never the repository 2026-11-16 11:32:32 +01:00
mesh-tools 99e2966bb7 Merge pull request 'Design 47 G: the lead's answer A-447c (nothing is ever repository-based)' (#622) from design/47-G-lead-answers-40 into main 2026-11-16 10:31:45 +00:00
jschoubben 42054c599c Answer A-447c: nothing in building, checking or delivering works on a repository 2026-11-16 11:31:35 +01:00
mesh-tools f887e777b2 Merge pull request 'Design 47 G: the lead's answer A-447b (a build is always for one module)' (#621) from design/47-G-lead-answers-39 into main 2026-11-16 10:30:14 +00:00
jschoubben bc9ba62959 Answer A-447b: a build, a step and a check are always for one module; a fetch with no dir is refused 2026-11-16 11:30:04 +01:00
mesh-tools 4353e7c018 Merge pull request 'Design 47 G: the lead's answer A-447a (a module whose directory holds no go.mod; an empty choice fails)' (#620) from design/47-G-lead-answers-38 into main 2026-11-16 10:15:43 +00:00
jschoubben ca8cda69d9 Answer A-447a: a module whose directory holds no go.mod takes the go.mod files under it; an empty choice fails 2026-11-16 11:15:34 +01:00
mesh-tools a7b72d49d2 Merge pull request 'Design 47 G: the lead's answer A-447 (a Go fetch asked with a module's directory resolves that module alone)' (#619) from design/47-G-lead-answers-37 into main 2026-11-16 10:00:19 +00:00
jschoubben 2c399ac708 Answer A-447: a Go fetch asked with a module's directory resolves that module and its local replaces alone 2026-11-16 11:00:04 +01:00
mesh-tools 8c307e0e7a Merge pull request 'Design 47 G: A-446a, what goes with the runner's set-aside' (#618) from design/47-G-lead-answers-36 into main 2026-11-16 09:35:06 +00:00
jschoubben 2d6da3f34d Answer A-446a: a placed pipeline says pipeline-started; what only the set-aside used goes 2026-11-16 10:34:59 +01:00
mesh-tools 45b43ce81c Merge pull request 'Design 47 G: A-446, the pipeline runner holds no pipeline for its own new version' (#617) from design/47-G-lead-answers-35 into main 2026-11-16 09:25:12 +00:00
jschoubben 13ad18268f Answer A-446: the pipeline runner holds no pipeline for its own new version (the operator) 2026-11-16 10:25:02 +01:00
mesh-tools 5be43e0193 Merge pull request 'Principles: no part without a need (7); depend on the contract, at the level it is interchangeable (9)' (#616) from principles/contract-and-no-part-without-need into main 2026-11-16 09:06:33 +00:00
jschoubben 4e0f340916 Principles: 7 becomes 'no part without a need', 9 'depend on the contract, at the level it is interchangeable' (the operator, 2026-11-16) 2026-11-16 10:06:25 +01:00
mesh-tools 12c0f66fab Merge pull request 'After G: the module graph's edges are dependencies, never verbs' (#615) from design/47-G-after-delivery-2 into main 2026-11-16 08:57:25 +00:00
jschoubben 6ccd7c3edf After G: the module graph's edges are dependencies on a seat or a module, never verbs or bus subjects (the operator) 2026-11-16 09:57:18 +01:00
mesh-tools ea02e537a3 Merge pull request 'Design 47 G: the lead's answer A-445 (the bus keeper's start path)' (#614) from design/47-G-lead-answers-34 into main 2026-11-16 08:52:04 +00:00
jschoubben 60872e67e9 Answer A-445: the bus keeper's A-376 path and its four manifest lines go 2026-11-16 09:51:57 +01:00
mesh-tools 9cf4d63c9c Merge pull request 'Design 47 G: what is owed once G is delivered' (#613) from design/47-G-after-delivery into main 2026-11-16 08:46:19 +00:00
jschoubben 1a2d31b82f Keep what is owed once G is delivered in hq, not in a session's memory 2026-11-16 09:46:13 +01:00
mesh-tools 267f1ef0de Merge pull request 'Design 47 G: the lead's answer A-443a (a given-up publish step)' (#612) from design/47-G-lead-answers-33 into main 2026-11-16 08:42:19 +00:00
jschoubben 23cba849e3 Answer A-443a: a given-up publish step's call ends at the SDK's own bound and is said 2026-11-16 09:42:12 +01:00
mesh-tools 3723b2ea41 Merge pull request 'Design 47 G: the lead's answers A-394 and A-434 to A-444 (the start fixes' review)' (#611) from design/47-G-lead-answers-32 into main 2026-11-16 08:38:26 +00:00
jschoubben 62e3715af4 Answer A-394 and A-434 to A-444: the start fixes' review questions, each against a principle or the operator's word 2026-11-16 09:38:16 +01:00
mesh-tools 18ce04f6e8 Merge pull request 'Design 47 G: the lead's answers A-363 to A-372 (G starts)' (#609) from design/47-G-lead-answers-31 into main 2026-11-16 08:24:38 +00:00
jschoubben 91f60c7075 Use the glossary's words in lead answers 31, so the words test passes 2026-11-16 09:24:19 +01:00
jschoubben 4169ded5cd Answer A-433: before the catalog half is read, a module account's call runs unmarked, seat verbs too 2026-11-16 02:34:22 +01:00
jschoubben d9dd5dab67 Answer A-432: a runtime of its own runs the tool runner's program, which §524's check of the module's parts leaves out 2026-11-16 01:00:35 +01:00
jschoubben 2829a47ea0 Answer A-431: a take-in names the source revision it makes, one above the newest seen or said 2026-11-15 17:51:01 +01:00
jschoubben 09cf7d609d Answer A-430b: a branch build is named by the version the builder reserved; build.ended carries it and the registry tags it unreleased 2026-11-15 16:48:29 +01:00
jschoubben 547eec3373 Answer A-430a: every wanted build has a version; the node-engine resolves artifacts by version through the registry's versions verb 2026-11-15 16:44:17 +01:00
jschoubben d2094f657e Answer A-430: a want names the module version; the node-engine fetches the artifacts by that version (the operator) 2026-11-15 16:43:23 +01:00
jschoubben ee80b4d84b Answer A-426a: every rule waiting on a serving part counts any healthy running build that serves it 2026-11-15 16:17:01 +01:00
jschoubben 14c3e90332 Answer A-429: a released build is done for rule 1 2026-11-15 16:13:29 +01:00
jschoubben cac334e749 Narrow A-428: only the pipeline runner's own want skips the passed-pipeline gate while no runner runs 2026-11-15 16:04:05 +01:00
jschoubben aad8f7c921 Answer A-428: with no pipeline runner running, wants do not wait for a passed pipeline 2026-11-15 16:03:56 +01:00
jschoubben b20b382cd9 Answer A-426 (an acting wait ends on any running build that serves the edge) and A-427 (a put may hold the running build) 2026-11-15 15:56:51 +01:00
jschoubben 8abbb539fd Answer A-425a: which proposals are asked is the assigner's config key ask-operator-for 2026-11-15 15:55:34 +01:00
jschoubben f988c23d1e Answer A-425: a passing placement proposal is applied without the operator's ask (the operator, 2026-11-15) 2026-11-15 15:54:46 +01:00
jschoubben fd45f01929 Answer A-424: the operator channel keeps no limit of open asks per asker (the operator, 2026-11-15) 2026-11-15 15:49:03 +01:00
jschoubben f0b9c73a9f Answer A-423: one reader per durable through the hub; a consumed kept event with no route is said 2026-11-15 03:52:41 +01:00
jschoubben 17ab2d028e Answer A-422: the SDK's event hub binds the kept seat streams in its own start 2026-11-15 03:48:47 +01:00
jschoubben 9a60a312a0 Answer A-421: a trunk entry with no fp-src is matched to its release by the digests its build made 2026-11-15 03:17:45 +01:00
jschoubben 1de67e76ed Answer A-419 (release matched by the fingerprint's source) and A-420 (a message given up is said) 2026-11-15 03:14:06 +01:00
jschoubben 1088d86e5c Answer A-418: a bundle's or archive's name for reserved is <module>/<artifact> from its entry's artifact reference 2026-11-15 02:59:19 +01:00
jschoubben 2ce9268504 Answer A-417: a registry holder tags a version on the artifact it holds by digest, else takes it from the artifact store 2026-11-15 02:37:16 +01:00
jschoubben 39ab7efb6b Answer A-416a: the builder reads a registry seat's holder from the catalog's seat holders, not FollowSeat 2026-11-15 02:17:33 +01:00
jschoubben 6c5f8aabb0 Answer A-416: the builder knows a registry seat's holder through FollowSeat, not from running events since its start 2026-11-15 02:16:18 +01:00
jschoubben 3d56f25942 Answer A-415: a published heard before the build's digests is kept on its reservation and marks it when the digests land 2026-11-15 02:05:05 +01:00
jschoubben 9da181e12d Answer A-414: the grants part composes a module whose running build has no entry from its newest entry, as A-407 does for the node 2026-11-15 00:23:25 +01:00
jschoubben 8b9b450855 Answer A-413: named asks are unused only while no account the part composes may publish one 2026-11-15 00:03:35 +01:00
jschoubben 16f4fd8696 Answer A-412a: the entry's digest readings live in the mesh-catalog contract 2026-11-14 23:44:50 +01:00
jschoubben ee543b8601 Answer A-412: the builder takes the entry's manifest digest bare, and a look that writes nothing says why 2026-11-14 23:44:11 +01:00
jschoubben 7981d72d9b Answer A-411: the builder reads the last versions-filled from the seat event's kept stream, not EVENTS 2026-11-14 23:37:07 +01:00
jschoubben 842071859a Answer A-410: a running module the node cannot fingerprint is held, never counted as removed 2026-11-14 23:14:41 +01:00
jschoubben 11054da240 Answer A-409: a consumed seat event matches by name and filters on the seat's subject; the revision's id names its branch 2026-11-14 22:51:41 +01:00
jschoubben 46b6b74148 Answer A-408: the builder reads the kept versions-filled at its start 2026-11-14 22:47:32 +01:00
jschoubben 706e76b566 Answer A-407: a running build with no entry is described for resolving by the entry of the build its want says is being built 2026-11-14 22:46:19 +01:00
jschoubben 4b2eb0bcf4 Answer A-405 and A-406: a node composes around a running build with no entry; the pipeline's manifest passes the judge and is built once by hand 2026-11-14 22:42:00 +01:00
jschoubben 634b4ea49e Answer A-404: a source the gate started from the forge names no group in its source-moved 2026-11-14 22:16:04 +01:00
jschoubben b6380d0958 Answer A-403: the gate adds a registered repository a held source key lacks, and changes no head it holds 2026-11-14 22:08:40 +01:00
jschoubben dbe9d711ea Answer A-402: a head is read whole on any failure to read its merged message 2026-11-14 22:08:11 +01:00
jschoubben de9f6d3353 Answer A-401: a newly watched branch starts from the forge at its head, as a clone does 2026-11-14 22:05:17 +01:00
jschoubben d5e7d42d17 Answer A-400: seat-holders names the assigned claimant of a seat no assignments holds 2026-11-14 20:21:29 +01:00
jschoubben 232454cf76 A-397: the rule is the developer's judgement (share a durable only for shared work), not an operator ruling 2026-11-14 18:51:53 +01:00
jschoubben f55a6e7927 Answer A-399: a seat pulled by kind's holder says the seat's own events 2026-11-14 18:39:47 +01:00
jschoubben f6f82721d5 Answer A-398: a seat-form read resolves to the assigned claimant when no holder is on record 2026-11-14 18:17:34 +01:00
jschoubben 98e8bc395b A-397: the operator's decision time read from the clock 2026-11-14 18:03:19 +01:00
jschoubben ed95207361 Answer A-397: every process that hears events has a durable consumer of its own, named constantly; a shared one split the first registrations between the assigner's program and its tool runner 2026-11-14 18:02:11 +01:00
jschoubben 525a9590cd Answer A-396: every module is composed as invoking the tool runner's own reads
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 12:09:52 +01:00
jschoubben 6bc258db0d Answer A-395: every trunk part and the tool runner may read the catalog half
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 12:00:53 +01:00
jschoubben 5f04a5e3af Answer A-393: a node seat's verb asked with no machine is answered by any one holder
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 11:31:09 +01:00
jschoubben 087e17c799 Answer A-392: a seed asks a silent build seat once, and says so
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 11:24:49 +01:00
jschoubben 19f5ad86d2 Answer A-391: the indexer's start replay writes the view once, and says its length
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 04:03:00 +01:00
jschoubben 5d7a96fe66 Answer A-390: the verb revision's callers count an invokes of a single *, as grants reads it
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 03:19:13 +01:00
jschoubben 5aa4510f0f Answer A-389: before the catalog half is read, a module's call of another module's tool runs unmarked, logged
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 02:36:55 +01:00
jschoubben c1a647e9bb Answer A-388: the start path takes only an explicit read class; the broker's four reads are classed
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 02:22:16 +01:00
jschoubben 8dffde7f33 Answer A-387: before the catalog half is read, a read verb of a held definition runs; git is bootstrapped
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 02:20:39 +01:00
jschoubben 240f6401ce Answer A-386: a runner kept as it runs still gains the runner grants of what it carries
mesh/delivery ready: it is merged on the Approve
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 01:32:07 +01:00
jschoubben 77a6a9aa68 Answer A-385: a kept account is the union of the list in force and this pass's composition
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 01:06:35 +01:00
jschoubben 9e619006c7 Answer A-384: a runner may make the ordered consumers its bundles hear new events on
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 00:43:18 +01:00
jschoubben 8af69b2082 Answer A-383: a right every account carries reaches a kept account too
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 00:22:48 +01:00
jschoubben 2c67b2e53b Answer A-382: every account may list stream names, which hearing an event starts with
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-14 00:06:50 +01:00
jschoubben be264026b5 Answer A-381: grants' own account may list stream names, which its stream assertion starts with
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 23:59:02 +01:00
jschoubben a95c0d6d98 Answer A-380: the runner's old-objects fallback uses the name grants creates
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 23:18:26 +01:00
jschoubben c1065d9868 Answer A-379: the runner reaches a module's old objects by the name its membership gives, beside bindings
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 23:15:17 +01:00
jschoubben 0e9b6e5286 Answer A-377 and A-378: first-view tools pass the start path; the keeper may read the indexer's views
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 23:12:47 +01:00
jschoubben 64b9ffe8c1 Answer A-376: the keeper makes the catalog holder's own bus state before its catalog half is in
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 23:01:04 +01:00
jschoubben c5adb7fe48 Answer A-375: the gate's start path also lets the bindings' first view through
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 22:46:43 +01:00
jschoubben 61149124f8 Answer A-374: the seat verb withheld reads a small key beside the list
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 22:10:26 +01:00
jschoubben 591297f02e Answer A-373: grants keeps its user list and pass in an object store, since they outgrow one bucket value
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 22:00:46 +01:00
jschoubben e9a2aaf253 Answer A-371 and A-372: grants counts a seat stream it creates itself as empty, and the assigner asks Ready
mesh/delivery closed: a newer head of the same pull request
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 20:54:05 +01:00
jschoubben 904f5bfac5 Answer A-370: the indexer answers the catalog's first views at once, never behind a running take-in
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 20:53:47 +01:00
jschoubben da161cdbf1 Answer A-369: grants' catalog wait asks the client's Ready, as every part does
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 20:45:09 +01:00
jschoubben 916be3a63c Answer A-363 to A-368: G's first start composes rights from the trunk, builds unpublished packages at the source commit, and waits for nothing only G answers
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
2026-11-13 20:34:47 +01:00
32 changed files with 4344 additions and 154 deletions
+54
View File
@@ -0,0 +1,54 @@
---
name: gap-surveyor
description: Holds the code on main of the Novox mesh against one part's specification (a spec-reader's file) and writes that part's gap table (hq playbook 07, before a prepare, when a design's parts are to be delivered again) - per design line what is built, built otherwise, contradicted or missing, where in code, how big the gap is, the leftovers of an earlier model, and which build runs live on each machine against main. Read-only on the mesh and every working copy; writes only the one file the caller names.
tools: Read, Grep, Glob, Bash, Write, mcp__mesh__mesh_call, mcp__mesh__mesh_describe, mcp__mesh__mesh_search
model: opus
---
You hold the code of the Novox mesh against the specification of one part of a design and write the gap table
the lead plans from. The caller names the part, its specification file, the repositories and modules that hold
it, the facts about what runs live on the machines that the caller already knows, and the file to write.
Read first: hq's glossary (`00-META/glossary.md`) and principles (`00-META/principles.md`), then the
specification file whole. Then the code.
How you read:
- **Code on main only, never a working copy's checkout.** Other sessions work in the working copies on this
machine; never check out, reset, stash, pull or commit there. Run `git fetch origin` in the repository and
read with `git show origin/main:<path>`, `git grep <words> origin/main`, `git ls-tree -r origin/main` and
`git log origin/main`. A branch the caller names (a held pull request) is read the same way as
`origin/<branch>`, and every fact from it is said to be from that branch.
- **Each fact with its source:** the repository, the short commit of `origin/main` you read, the module and the
function or type; never a line number.
- **What runs live** you read through the mesh MCP server with read tools only: `mesh_machine` for a machine's
modules, a module's own read tools and the seat verbs that read its state (find them with `mesh_search`).
Call no verb that writes, starts, cancels or restarts anything.
- **The tests count:** per module, how many unit tests and how many integration tests (`//go:build
integration`) main holds, and whether one integration test starts the module whole against a bus and a store.
Write the file the caller names, in this shape:
1. **The gap table**, one row per design line of the specification, in its order: the citation; its status, one
of *built* (the code does what the line says, by its contract), *built otherwise* (the outcome is there but by
another path: a poll, a timer, a direct call, a shared file, a read of another part's state, a comparison in
place of an event), *contradicts* (the code does something the line forbids or that a replaced line said),
*missing*, or *not judged* (say why); where it lives (repository, module, function or type); what the code
does instead, in one sentence, when it is not *built*; and the size of the gap, S (one function, under an
hour), M (one module's part, a day), L (the part's shape, more than a day).
2. **The leftovers**: every mechanism in these modules that no line of the specification asks for, or that a
replaced line asked for: what it is, where, what it does, and which line it contradicts or none. A sweep, a
poll, a retry loop, a stand-in, a hand-act path, a per-merge path count.
3. **Live against main**, a table per module of the part: the machine, the build that runs (its version or
fingerprint as the machine's tools say it), whether that build is main's or a hand-placed one, and what main
has that it lacks.
4. **The tests**: per module, the counts, whether an integration test starts the module whole, and what the
tests would not catch of the gaps above.
5. **Fix in place or write again**, one paragraph per module: your judgement from the table, with the reason:
the share of lines *built*, how many gaps are L, whether the module's shape is the specification's (its
events, verbs, kept data) or an earlier model's. A recommendation, not a decision.
6. **What you could not read**, and every fact you guessed, said as a guess with how to check it.
Report to the caller in under 400 words: the file's path, the counts per status, the three gaps that matter most,
your fix-or-rewrite recommendation per module in one line each, and what you could not read. Plain words in the
glossary's words.
+4
View File
@@ -26,6 +26,10 @@ questions are.
- an interaction the design does not place that is not a mesh path: one module naming another, reading another's
internals or being special-cased in another part; a module's name in shared code; state only in memory; a poll or
a timer where an event exists;
- a value written into the code that can differ between machines, modules or over time (a port, a path, a bound, a
cap, a name, a time) and is not a setting with a default (principle 11); name the setting it should be, eagerly;
- a module that writes another module's file or state, or anything in another module's place, instead of giving it
through a provision or a contribution that module declares (principle 12);
- an effect or a reader of the build map's `What exists today` whose fate the diff does not carry out (run its
searches again on the reviewed tree), or one the search finds and the answer omits;
- a test that does not fail without its change (check it by taking the change out in a copy), an acceptance
+6
View File
@@ -23,6 +23,12 @@ without that read on each machine the module is wanted on.
Loop: wait with `mesh_wait` on the event family (its answer only wakes you), then read the state from the verb
the handover names; never report a state you did not read in that round.
**A state that does not change is not "nothing happened".** A reservation, a queue or a key can stay as it was while
the work behind it already failed. Each round, also read the journal of the program that does the work
(`<node>/node-service-manager.journal` with its unit, the last lines since your previous read) and report any failure
line word for word. A verb that times out or refuses twice is a finding of its own: report it at once, never read
around it. You have no clock: take times from the tools' answers, never guess them.
**A report along the way does not end you.** Your final message ends you, and then nothing watches. So while you
keep watching, report with `SendMessage` to `main`: when five minutes passed with no change while something
should be moving (say what you last read and when), and at each change the handover asks to hear. Then go on
+3 -1
View File
@@ -25,7 +25,9 @@ write.
What you judge:
- Seams: where each piece of state lives on the bus (key-value bucket, stream, object store) and which module owns it; which seat, verb or event joins two parts; any link that is not NATS-native (a poll, a timer, HTTP between modules, a shared file, state only in memory); whether a form the manifest lacks is named with its cost (a mesh-sdk change rebuilds every module).
- Seams inside a module, at run time: for every event, queue and piece of state the subject adds or changes, which processes read it on each node (a module's program, its tool runner, any other process its manifest names, the same module on several nodes), and whether each gets every item it needs. A NATS durable consumer or a queue group hands each message to one reader only: two processes reading under one name split the messages silently, and the bus refuses nothing. A design line that names one reading place per module where more than one process reads is a **must**; only a work queue, where one reader per item is the point, is shared. (The operator, 2026-11-14: on ace the assigner's program and its tool runner read one durable consumer, and the runner took all 178 first registrations the program acts on; lead answer A-397.)
- Settings and ownership: every value that can differ between machines, modules or over time is a setting with a
default, never hard-coded (principle 11), and the design names it; no module writes another module's file or state:
what one module needs from another goes through a provision or a contribution that module declares (principle 12).
- Tools: for every new piece of state a tool that reads it, and for every new behaviour a tool that steers it (do by hand, ask again, drop, repair, re-check), as the instruction tools-are-how-a-module-is-spoken-to says.
Judge only whether the settled direction or design is complete and sound in your two architectural domains,
+43
View File
@@ -0,0 +1,43 @@
---
name: spec-reader
description: Reads one part of a to-be design of the Novox mesh with the records that decide it, and writes that part's specification as a list of its design lines with their checks (hq playbook 07, before a prepare, when a design's parts are to be delivered again) - what the part does, the seats, verbs, events and kept data it names, which later answers amend each line, and what the design leaves open. Reads no code; writes only the one file the caller names.
tools: Read, Grep, Glob, Bash, Write, mcp__mesh__mesh_call, mcp__mesh__mesh_describe, mcp__mesh__mesh_search
model: opus
---
You read one part of a to-be design of the Novox mesh and write its specification: everything the design and
its records say that part does, as a list a builder and a surveyor can hold code against. The caller names the
part, the design's files and sections to read, the records to read in the knowledge base, the other parts (so you
know where your part ends), and the file to write.
Read first: hq's glossary (`00-META/glossary.md`) and principles (`00-META/principles.md`), then the design's
main file sections the caller names, then the milestone part's files whole where the caller says whole, and the
answer files (the later design lines and the lead's answers) where they name your part: find them with `grep`
for the part's words (its seat, its module, its verbs, its events) and read each matching answer whole. Then the
records: `knowledge-base.get` with `reasoning: true` and `bound: 65536` for each record the caller names, and
`knowledge-base.ask` for the part's words to find records the caller did not name.
Never read code, a working copy or a repository's history: the specification says what should be, and the code
is the surveyor's job. Through the mesh MCP server only reads of the knowledge base.
Write the file the caller names, in this shape:
1. **What the part does**, in ten lines or fewer, in the glossary's words: its seat, the module that holds it,
its verbs, the events it publishes and consumes, the kept data it owns on the bus (state, streams, objects),
and the parts it talks to.
2. **The design lines**, a table, one row per line: the line's citation (`design 47 §n`, `design 47 G §n`,
`A-n`), one sentence of what it says, the record that decides it (`ADR nnnn §k`), how it is checked (the
acceptance test, the tool that reads it, the event that says it), and the later line or answer that amends or
replaces it, when one does. A line a later line replaced is kept in the table, marked replaced, so the
surveyor can tell a leftover from a build.
3. **The interactions**, one row per edge with another part: the event or verb, who publishes or serves, who
consumes or calls, and the line that decides it. A direct call, a shared file or a read of another part's state
is marked as such: it is a finding for the review, never a solution.
4. **What retires** at this part's delivery, with its line.
5. **What the design leaves open** for this part: a question no line answers, each in one sentence, with the
principle you think settles it or "no principle settles it". Decide nothing.
6. **What you could not read**: a record the knowledge base cut, a file you did not reach.
Report to the caller in under 400 words: the file's path, the counts (lines, interactions, open questions), the
three findings that matter most for delivering this part again, and what you could not read. Separate fact from
guess. Plain words in the glossary's words.
+61 -3
View File
@@ -600,7 +600,46 @@ controller (the planner, the rollout) until release 6 of ADR 0429; from each rel
registering and changing one are trusted acts (ADR 0666 §2). The mesh knows modules, and a repository only as a
module's git origin.
- **watched branch** — a branch a module's registration names, whose commits start that module's pipeline and which a
placement may choose for a node; registering one the forge does not protect is refused (ADR 0666 §3, §6).
placement may choose for a node; from milestone H's delivery it need not be protected, and nothing reads its
protection (ADR 0682 §13, which ends ADR 0666 §3's refusal and §11's judgement).
- **merge call** — one call of the seat `git`'s verb `merge-pull` naming one or more pull requests, each at the head
its caller names: from the delivery of design 47's milestone H the start of their delivery and, with two or more,
their group; the forge module judges every member before it merges any and says the call as its event
`merge-called`, from which the gate names the merged heads one source revision (ADR 0682). Not a merge made in the
forge's own screen, which delivers nothing and leaves an **untaken head**.
*Not:* ~~kick-off~~ (hq)
- **untaken head** — a watched branch's head that the forge bridge said moved (`head-moved`) and that no source revision
names within the gate's setting `untaken-bound`: a head moved outside the mesh, raised as the gate's finding
`head-untaken` and delivered by nothing but a later merge call or the gate's hand take (ADR 0682; design 47 H §740).
Not a pending revision, which is a source revision not yet taken in.
*Not:* ~~merge-untaken~~ (hq)
- **`merge-called`** — the seat `git`'s event saying one merge call: its members that landed, each at its merge
commit, and those refused; the gate's one input that names a source revision (ADR 0682 §1; design 47 H §736).
*Not:* ~~merge notice~~ (hq)
- **`head-moved`** — the forge bridge's one message for a move of a watched branch's head, however it moved, heard by
the gate's watchdog alone (ADR 0682 §3; design 47 H §739). It replaces the messages `merged` and `pushed`.
*Not:* ~~pushed message~~ (hq)
- **`head-untaken`** — the gate's finding for an **untaken head**, raised when `untaken-bound` passes (ADR 0682 §3).
*Not:* ~~untaken merge~~ (hq)
- **`untaken-bound`** — the gate's setting, default 10 minutes, after which a head moved and not named by a source
revision is `head-untaken` (ADR 0682 §3).
*Not:* ~~take-in bound~~ (hq)
- **`pipelines-at-once`** — the runner's setting, 1 to 8, default 1: how many pipelines have their steps asked at
once; the rest wait in order (ADR 0682 §7; design 47 H §748).
*Not:* ~~concurrent pipelines~~ (hq)
- **`tier-late`** — the assigner's finding when one tier of a pipeline's movers does not run on its present machines
within `tier-bound`: no later tier is written and the delivery record ends `failed` (ADR 0682 §8, §9; design 47 H
§751).
*Not:* ~~late tier~~ (hq)
- **take-in** — the indexer's act on one source revision of a watched branch: the commits it names, the
build-input fingerprints of the modules registered for those repositories, and one catalog revision of that branch,
said as `take-in.taken` naming the **movers**; it validates no manifest (that is the pipeline step `validate`) and
holds nothing for a named indexer (ADR 0682; A-476). Not a merge: what is taken in is a source revision the gate named.
*Not:* ~~merge take-in~~ (hq)
- **mover** — a module whose build-input fingerprint changed in a take-in, or a code dependent of one (a module, in any
repository, that imports a package of one): the modules a take-in's one pipeline runs and delivers (A-476a; ADR
0573 §3; ADR 0682). Not every module of the revision.
*Not:* ~~moving module~~ (hq)
- **run on request** — a pipeline run for a named commit or group that delivers nothing, asked by the runner's verb
`run`; the indexer builds its candidate from the runner's `run-started` (ADR 0666 §18).
- **artifact reference** — an artifact written `<module>:<artifact>`, such as `mesh-grants:program`: the end of a
@@ -620,6 +659,15 @@ controller (the planner, the rollout) until release 6 of ADR 0429; from each rel
- **composition step** — the pipeline step that composes each machine's declaration with a change, by the composer built
from its heads in the sandbox, refusing a change in which two modules own one path; served by the seat `mesh-wants`
(ADR 0666 §17). From milestone G's delivery it replaces the **composition check**.
- **composition run** — the one run of the composer for one candidate on one machine, started by the first
composition step of that candidate that needs the machine and kept by the assigner in its state `composition-runs`;
every composition step of the candidate whose module is placed on the machine reads its verdict (design 47 G §718).
Not a **composition**, one module's resources composed for one node; nor the **composition step**, the pipeline
step that reads one composition run per machine.
- **test run** — the one run of a lint, unit test or integration test pipeline step for one candidate, one level and
one set of tested packages at their digests, kept by the runner in its state `test-runs`; every artifact's pipeline step of that candidate
naming the same packages reads its result (design 47 G §724). Not a **pipeline step**, each artifact's own, nor a **run
on request**, a whole pipeline that delivers nothing.
- **merge check** — what a pull request carries before it may merge, judging the merge of its head into a
named **trunk commit**, made in a clone of its own on a build node and never pushed (ADR 0323). One status,
`mesh/check`, written by the checker: the change's tests, run by the language's build module, and the
@@ -807,7 +855,8 @@ controller (the planner, the rollout) until release 6 of ADR 0429; from each rel
level: a group holds deliveries, never groups. Its state is derived from its members, never set. From milestone G's
delivery: a shared branch name carrying its issue number, complete when each member's commit has landed on its
modules' watched branch, taken in as one source revision of that branch and run as one pipeline per watched branch
(ADR 0666 §13).
(ADR 0666 §13). From milestone H's delivery: the pull requests named in one **merge call**; no branch name and no
pull request read after the call forms one (ADR 0682).
- **hotfix** — one pull request in one repository, delivered at its merge (ADR 0401).
- **feature** — one branch name across its repositories: in two or more, its pull requests are one delivery group,
merged together and delivered member by member in the group's order; in one, a single pull request delivered as a
@@ -881,7 +930,9 @@ controller (the planner, the rollout) until release 6 of ADR 0429; from each rel
- **pending revision** — a source revision the running indexer has not taken in, because it names another build of
the indexer or follows one that does: only the named indexer's own keys are written, the rest kept in the
indexer's state `pending` until that indexer starts and takes it in (ADR 0633 §1 to §3). A group's members wait
there too until the group's end, as members still gathered (§4).
there too until the group's end, as members still gathered (§4). Ends at milestone H's delivery: a take-in holds
nothing for a named indexer, and one that cannot be written waits as its unacknowledged `source-moved` (ADR 0682;
design 47 H §744).
- **canary** — the first node of a module's nodes, ordered by ring then name: a new fingerprint reaches it
first and the module's other nodes only once it runs healthy there (ADR 0429 §6). Replaces sending to every
machine at once from release 6 (§15); a fingerprint that fails there is held and the canary set back, or for
@@ -1113,6 +1164,11 @@ it should not. **Recorded by** the controller. **Upstream of** Provisioning and
- **packet filter** — what the mesh enforces on a node about which packets pass, the
`node-packet-filter` seat and its verb `rules`. A **found firewall** is a program the mesh found on a
machine and keeps retired or in force; *firewall* says that, and a wrapped program's firewall is its own.
- **allowance** — a port of the packet filter opened by the operator, through the seat, beside the ports derived from
the modules' `listens`: set with the `node-packet-filter` seat's verb `allow`, taken back with `disallow`, read with `allowed`, each with
who set it, when, why and an optional `until`, at which it is removed, and rendered by the node-engine into a converged node's input chain
(design 08 §4, design 33). It is not an `opening`, the resource the node-engine converges through a found firewall.
*Not:* ~~pinhole~~ (hq), ~~hand rule~~ (hq)
- **intrusion prevention** and **ban** — refusing a source that misbehaved, and the refusal itself; the node seat
`node-intrusion-prevention` is of this domain.
- **reach** — how far an assignment's endpoint may be reached from: the node, the private network, or
@@ -1531,6 +1587,8 @@ renaming one sense, the old sense moves to a *Not:* line and becomes mechanical.
| | a human | **operator** or **person** | words every domain uses |
| firewall | what the mesh enforces | **packet filter** | Connectivity |
| | what was found on a machine | **found firewall** | Connectivity |
| allowance | a port of the packet filter opened by the operator, through the seat | **allowance** | Connectivity |
| | the share of model use every session draws on in a week (playbook 07's report) | **weekly allowance** | — |
| ticket | an answer's one-time token (ADR 0258) | **answer ticket** | Operator and conversation |
| | a question someone must answer | **issue** | The record |
| deploy | the per-node part of a delivery plan | **deploy plan** | Delivery |
+12 -3
View File
@@ -1,6 +1,6 @@
---
status: canonical
updated: 2026-11-11
updated: 2026-11-16
---
# The principles
@@ -22,9 +22,18 @@ add, sharpen or drop a principle in a pull request to this file.
6. **Nothing is exposed publicly.** No data leaves the mesh unasked: no secret in a record, a log or a public
repository, no service reachable from outside unless it is meant to be. Inside the mesh, every agent and module
is trusted; who may do what is not a question to ask.
7. **The simplest thing that works.** Prefer what already exists. No new mechanism unless it is needed, and fewer
parts over more.
7. **No part without a need.** Every mechanism, field, seat and step exists because something needs it; prefer what
exists over a new one, and remove what no longer serves. Never a shortcut instead of the right shape.
8. **One word per thing**, from the [glossary](glossary.md).
9. **Depend on the contract, at the level it is interchangeable.** A module depends on a seat where the mesh means
more than one module to be able to fill the need, and on a module where only that module does; never on another
module's internals.
10. **The module is the unit.** Everything is done per module, never per repository. A repository is only where a
module's files are kept; a merge moves only the modules it changes.
11. **Configurable, never hard-coded.** A value that can differ between machines, modules or over time (a port, a
path, a bound, a cap, a name, a time) is a setting with a default, never written into the code.
12. **A module writes only its own.** A module writes its own files and state alone; what another module needs from it
goes through a provision or a contribution that module declares, never a file written in its place.
How it is checked: the one code review of a milestone checks the code against these principles and the design's
shape; an answer to a develop question names the principle that settles it.
+1 -4
View File
@@ -112,10 +112,7 @@ the four reviewer agents in `.claude/agents/` (`review-security-and-upgrade`, `r
link that is not NATS-native. Each piece of state is kept in the bus storage form that fits it (a key-value
bucket, a stream, an object store); a form the manifest lacks is added to mesh-sdk, and the review names that change
and its cost (a mesh-sdk change rebuilds every module) to the operator, never folds it in silently (the
operator, 2026-11-02). Inside a module too: which of its processes (its program, its tool runner, another process
its manifest names) read each event, queue and piece of state at run time, and whether each gets every item it
needs: a durable consumer or a queue group hands each message to one reader only, so a reading place shared by two
processes splits the messages silently (the operator, 2026-11-14; lead answer A-397);
operator, 2026-11-02);
- **running qualities**: build capacity, speed, storage growth, and what happens while a part, the bus or the
control-node is down;
- **health**: what each part's health check says, which conditions are raised and cleared, and whether any
+41 -92
View File
@@ -22,7 +22,7 @@ code:
- mesh-host internal/apply (the service that reflects a rule set; a whole file handed to its new owner)
- mesh-controller cmd/mesh-controller/nodes.go (a token issued for a machine's tunnel key, ADR 0169)
- mesh-host cmd/mesh-host/join_tunnel.go (a machine joins through the tunnel, ADR 0169)
updated: 2026-10-14
updated: 2026-11-21
decisions:
- "ADR 0241"
- "ADR 0117"
@@ -36,6 +36,7 @@ decisions:
- "ADR 0194"
- "ADR 0191"
- "ADR 0180"
- "ADR 0683"
- "ADR 0170"
- "ADR 0169"
- "ADR 0168"
@@ -696,10 +697,14 @@ program that reads it, and another proxy may implement the same file.
## 4 — Filtering
**Derived from what is assigned here, and from the private network's shape** — a node's open ports are a
consequence of what runs on it and who must reach it, not an independent declaration to keep in
step by hand. That is true of a converged node; an adopted one keeps the firewall it was found
with until it converges (below).
**Derived from what is assigned here and from the private network's shape, with allowances set by hand beside
them.** The automatic part of a node's open ports is a consequence of what runs on it and who must reach it, not a
declaration to keep in step by hand. Beside it, the holder of `node-packet-filter` keeps **allowances**: ports opened
by the operator, through the seat's verbs, each with who set it, when, why and an optional expiry, rendered into the same input
chain and surviving every render and restart; an allowance is removed at its `until` and said (ADR 0683,
the firewall's control verbs, amending ADR 0170; the operator, 2026-11-21; design 47 H's lead
answers H-A-304, H-A-305, H-A-307, H-A-308, H-A-310, H-A-312 and H-A-314 to H-A-316). That is true of a converged node; an adopted one keeps the firewall it was found
with until it converges, and takes no allowance (design 08 adoption §2 to §6).
**A rule names its source** (ADR 0007).
A rule with no source is open, and must say so rather than appear to restrict something. `scope:`
@@ -790,57 +795,25 @@ undeclared one does not — then the module is removed and the port closes with
rule. *A rule set that is written but never loaded passes every check that reads the file, which
is why the check reads packets.*
### While a predecessor still serves the same thing
### By hand, beside the derived ports
*2026-09-22, ADR 0104.*
The mesh's proxy owns the two public ports and derives its routes from what modules contributed, so
it cannot stand beside a predecessor's proxy and cannot serve the predecessor's names. During a
migration a provision of this shape — one provider, a scarce machine-wide resource — may instead be
answered by an **adapter**: a module that provides `route` and writes each contribution into the
predecessor's own configuration, pointing at the machine port the contributing module publishes.
The predecessor's proxy goes on serving every name and renewing every certificate; the mesh's proxy
is taken last, when every route is one the mesh contributed. The adapter is refused on a converged
node, and removed when the predecessor's proxy retires. *How it is checked:* the adoption bed takes
a module behind the adapter and asserts the name answers and the predecessor's own names still do.
*2026-11-21, ADR 0683.* Deriving every open port from the modules' `listens` left no way to open a port no manifest
names (a program the operator runs by hand on the workstation, for a while). One renderer stays: the node-engine
composes the filter file, and nothing else writes the input chain.
### On an adopted node
```
caller ── allow / disallow ──▶ node-packet-filter (holder) state allowances, one key per allowance
│ event node-packet-filter.allowances-changed (node)
▼
node-engine ── allowed ──▶ holder composes the filter: derived ports + allowances, input chain only
│ writes the filter file; the holder's unit loads it (reload)
```
*2026-09-22, ADR 0100.* **The firewall found on the machine stays in force**, and the mesh
loads no table there that drops by default or holds an accept — neither genesis's base ruleset nor
the derived one. Every base chain at a hook runs and a drop in any is final, whatever the other
firewall is written in, so a second, stricter table would close every port the machine serves, and
an accept in one would open nothing the found firewall drops. What the mesh needs reachable it
declares as **openings**: a port, from where, on the incoming path or the forwarded path — a
published container port is forwarded, and a firewall that filters only incoming traffic never
sees it. The controller derives them from what the filter would be derived from, each from where
the filter would admit it: the assigned modules' `listens`, the hub's port, the bus and the registry
from anywhere; the store's port and the broker's management port from the private network. The node-engine converges each opening through
the found firewall in that firewall's own terms, marks it as the mesh's, removes only what it
marked, and re-checks every opening on each reconcile so a reload or a reboot does not lose it for
longer than one reconcile. An opening is state, not a command, so it travels over the link like any
other resource. **The mesh guards its own ports in a table of its own that only refuses** —
passing everything by default and holding nothing but refusals, and the found firewall's reload
does not touch it. Its ports are derived (ADR 0103): every machine port a
*taken* module publishes that the filter admits from the private network only, with the store's
port and the broker's management port. A port of a module not yet taken may still be the
predecessor's, so it is not guarded; and only packets addressed to this machine are matched, so
traffic it routes for others passes. An opening a found rule already answers is not added, so
removing it never removes the operator's rule. It refuses those ports except from
the private network and the machine itself — loopback and the container runtime's networks, known by the interface a packet arrives on, never by
its source address alone — at
the prerouting hook, before the container runtime redirects the packet, so it matches the port the
packet was sent to, for both address families; the
bus, the registry and the hub's port stay reachable from anywhere, as a node enrols and pulls
before it has a private-network address. One kind of found firewall is spoken; a machine with none
needs no openings, and a machine with another kind is refused adoption. Converging the node refuses
while a found container is still held; otherwise it previews what is reachable now — listening
sockets and published ports — what will close and which modules it will take, then loads the
derived filter in place of the refusal-only table and disables the found firewall without flushing
it. *How it is checked:* the adoption bed asserts the found firewall's rules differ only by the
mesh's marked rules, that the store is unreachable from off the private network before and after
the found firewall reloads and reachable from a container on the node, that a machine not yet enrolled reaches the bus, that a second machine
enrols through the openings before and after a reload and a reboot, and that after the flip the
declared port is open and the undeclared one closed.
The holder keeps each allowance in its own state on the bus and says every change as an event; the node-engine of
that node hears it, asks `allowed` and composes the filter again. An unanswered `allowed` refuses the compose and
keeps the loaded filter, said once, never read as "nothing opened" (the rule for a generator above). Allowances are
rendered only into a converged node's own table. The seat's verbs and event are design 33's (*The firewall seat's
verbs*); every line the milestone builds is in [design 08 control-verbs](08-milestones/control-verbs.md).
### It filters what arrives from outside, and not what the machine's own guests send
@@ -1245,27 +1218,15 @@ The list is worth having in one place, because it is most of the argument:
## Open
- **One resolver (ADR 0194).** *Correction of fact,
2026-10-05:* no node holds `node-dns-resolver` any more, and
ADR 0220 deletes the seat. What stood here before: *"Not built: every node still runs
`node-dns-resolver`. The migration's four steps are in the record, in order."* Nor did it
stay true that *"the workstation moves to the one resolver only once"* zones and the `hosts` file's
holder (ADR 0199)
exist: every node, the workstation included, asks the one resolver, and every node holds
`node-hosts-file`.
- **One resolver (ADR 0194), settled** (2026-10-05): ADR 0220 deletes the seat `node-dns-resolver`; every node, the
workstation included, asks the one resolver and holds `node-hosts-file` (ADR 0199).
- **When the hub is down nothing takes over** (ADR 0006): non-co-located paths stop, co-located direct peers and
every assigned workload keep running; recovery is restore, before certificate renewal.
- ~~**What happens when the hub is down.**~~ **Resolved** by
ADR 0006, together with `06`'s
matching question — they were one question. Nothing takes over. WireGuard has no failover, the
hub is declared rather than elected, and non-co-located paths stop while co-located direct peers
and every already-assigned workload keep running. The recovery path is restore, and its deadline
is certificate renewal.
- **Renumbering the private network.** Made *possible* by declaring the hub rather than inferring it from
an address, but no procedure exists, and a graph delivered node by node has an ordering problem
while it is half-applied.
- ~~**Revoking a route** when a module is unassigned.~~ **Resolved** 2026-08-31 — see §3. The file
a proxy is given is the whole truth about who has a route, so a route does not outlive the module
that asked for it.
- **IPv6.** ADR 0007 makes
it expressible; nothing here says the private network or the resolver handle it.
- **Reporting declared-versus-observed.** ADR 0007 makes the disagreement detectable and does not
@@ -1276,28 +1237,6 @@ The list is worth having in one place, because it is most of the argument:
the composition is a per-node override rather than the design. The interpolation that would let a
module carry a label and a node carry the domain, and the mesh join them, does not yet exist.
## The hub adopts the predecessor's tunnel
*2026-09-23, ADR 0105.*
On an adopted node that is the hub, the private network is not raised beside the tunnel it finds;
it **takes it over**: the found interface's private key, its port, its address and range, and every
peer it had, carried as peers not yet enrolled. The found interface is stopped, its configuration
kept on disk. A predecessor machine sees the same server key at the same endpoint and cannot tell
the tunnel changed hands; when it enrols, it keeps the address the tunnel already held for its key.
The mesh's own addresses are the adopted range's — every binding, hosts entry and endpoint the
controller composes follows it, as readers of a setting. ADR 0100's non-overlap rule applies only
where a found tunnel is left running beside the mesh's; where it is adopted there is one tunnel.
The guard admits the mesh's ports from that one interface, and the predecessor's peers arrive on
it.
*2026-09-27, ADR 0119.* The
found configuration is kept only until the take is proven — the found unit down, the mesh's
interface up and handshaking with a peer. Then it is removed from where the found unit reads it
(its original stays kept), the hold ends, and the predecessor's tunnel cannot be raised again by
anything but a person restoring it by hand. Undeclaring the private network does not bring it back.
## The bus is NATS
*2026-09-23, ADR 0106. Architecture to be written
@@ -1308,3 +1247,13 @@ NATS account with permissions derived from `emits`/`consumes`, declared as confi
writes and the server reloads. The sdk's contract is unchanged. The adopted AMQP broker stays as
the predecessor's compatibility broker until its last client is gone. Built in the lab beside the
migration; cut over in one switch-over after the core; the person's client is designed on it.
## Milestones
Each milestone's lines are in its part, cited as "design 08 <milestone> §n"; the rest of this file records what was
built before milestone parts existed.
| Milestone | Part | State |
|---|---|---|
| adoption: a node the mesh adopted, before it converges (ADR 0100, 0103, 0104, 0105, 0119) | [adoption](08-milestones/adoption.md) | delivered |
| control-verbs: allowances set by hand beside the derived ports, built in design 47's milestone H (ADR 0683) | [control-verbs](08-milestones/control-verbs.md) | to build |
@@ -0,0 +1,79 @@
---
layer: to-be
design: 08
milestone: adoption (delivered)
updated: 2026-11-21
decisions:
- "ADR 0100"
- "ADR 0103"
- "ADR 0104"
- "ADR 0105"
- "ADR 0119"
---
# Design 08, milestone adoption — a node the mesh adopted, before it converges
**What it built.** How the mesh stands beside what it found on an adopted node: a predecessor's proxy, the found
firewall and the predecessor's tunnel, each until the node converges. This part is delivered: it records what was
intended, moved whole from the main file when design 08 was put in the shape of playbook 02 (2026-11-21), and no
builder reads it again. Cited as "design 08 adoption §n".
## Lines
1. **While a predecessor still serves the same thing, an adapter answers the provision** (2026-09-22, ADR 0104).
The mesh's proxy owns the two public ports and derives its routes from what modules contributed, so it cannot
stand beside a predecessor's proxy and cannot serve the predecessor's names. During a migration a provision of
this shape (one provider, a scarce machine-wide resource) may instead be answered by an **adapter**: a module
that provides `route` and writes each contribution into the predecessor's own configuration, pointing at the
machine port the contributing module publishes. The predecessor's proxy goes on serving every name and renewing
every certificate; the mesh's proxy is taken last, when every route is one the mesh contributed. The adapter is
refused on a converged node, and removed when the predecessor's proxy retires. *How it is checked:* the adoption
bed takes a module behind the adapter and asserts the name answers and the predecessor's own names still do.
2. **On an adopted node the found firewall stays in force** (2026-09-22, ADR 0100), and the mesh loads no table
there that drops by default or holds an accept: neither genesis's base ruleset nor the derived one. Every base
chain at a hook runs and a drop in any is final, whatever the other firewall is written in, so a second, stricter
table would close every port the machine serves, and an accept in one would open nothing the found firewall
drops.
3. **What the mesh needs reachable it declares as openings**: a port, from where, on the incoming path or the
forwarded path (a published container port is forwarded, and a firewall that filters only incoming traffic never
sees it). The controller derives them from what the filter would be derived from, each from where the filter
would admit it: the assigned modules' `listens`, the hub's port, the bus and the registry from anywhere; the
store's port and the broker's management port from the private network. The node-engine converges each opening
through the found firewall in that firewall's own terms, marks it as the mesh's, removes only what it marked, and
re-checks every opening on each reconcile so a reload or a reboot does not lose it for longer than one
reconcile. An opening is state, not a command, so it travels over the link like any other resource. An opening a
found rule already answers is not added, so removing it never removes the operator's rule.
4. **The mesh guards its own ports in a table of its own that only refuses**: passing everything by default and
holding nothing but refusals, and the found firewall's reload does not touch it. Its ports are derived (ADR
0103): every machine port a *taken* module publishes that the filter admits from the private network only, with
the store's port and the broker's management port. A port of a module not yet taken may still be the
predecessor's, so it is not guarded; and only packets addressed to this machine are matched, so traffic it routes
for others passes. It refuses those ports except from the private network and the machine itself (loopback and
the container runtime's networks, known by the interface a packet arrives on, never by its source address alone)
at the prerouting hook, before the container runtime redirects the packet, so it matches the port the packet was
sent to, for both address families; the bus, the registry and the hub's port stay reachable from anywhere, as a
node enrols and pulls before it has a private-network address.
5. **One kind of found firewall is spoken**; a machine with none needs no openings, and a machine with another kind
is refused adoption.
6. **Converging the node refuses while a found container is still held**; otherwise it previews what is reachable
now (listening sockets and published ports), what will close and which modules it will take, then loads the
derived filter in place of the refusal-only table and disables the found firewall without flushing it. *How
lines 2 to 6 are checked:* the adoption bed asserts the found firewall's rules differ only by the mesh's marked
rules, that the store is unreachable from off the private network before and after the found firewall reloads
and reachable from a container on the node, that a machine not yet enrolled reaches the bus, that a second
machine enrols through the openings before and after a reload and a reboot, and that after the flip the declared
port is open and the undeclared one closed.
7. **The hub adopts the predecessor's tunnel** (2026-09-23, ADR 0105). On an adopted node that is the hub, the
private network is not raised beside the tunnel it finds; it **takes it over**: the found interface's private
key, its port, its address and range, and every peer it had, carried as peers not yet enrolled. The found
interface is stopped, its configuration kept on disk. A predecessor machine sees the same server key at the same
endpoint and cannot tell the tunnel changed hands; when it enrols, it keeps the address the tunnel already held
for its key.
8. **The mesh's own addresses are the adopted range's**: every binding, hosts entry and endpoint the controller
composes follows it, as readers of a setting. ADR 0100's non-overlap rule applies only where a found tunnel is
left running beside the mesh's; where it is adopted there is one tunnel. The guard admits the mesh's ports from
that one interface, and the predecessor's peers arrive on it.
9. **The found tunnel's configuration is kept only until the take is proven** (2026-09-27, ADR 0119): the found
unit down, the mesh's interface up and handshaking with a peer. Then it is removed from where the found unit
reads it (its original stays kept), the hold ends, and the predecessor's tunnel cannot be raised again by
anything but a person restoring it by hand. Undeclaring the private network does not bring it back.
@@ -0,0 +1,107 @@
---
layer: to-be
design: 08
milestone: control-verbs
updated: 2026-11-21
code: [mesh-host internal/compose (the filter's compose), mesh-catalog modules/nftables, mesh-networking modules/mesh-connectivity-contracts]
decisions:
- "ADR 0683"
- "ADR 0170"
- "ADR 0168"
- "ADR 0595"
- "ADR 0100"
---
# Design 08, milestone control-verbs — allowances set by hand, rendered beside the derived ports
**What it builds.** The packet filter of a converged node opens, beside the ports derived from the modules' `listens`,
every **allowance** the holder of `node-packet-filter` keeps: a port opened by the operator, through the seat, with
who set it, when, why and an optional `until`. The node-engine renders both into the mesh's own table, in the same
input chain, at every compose of the filter; an allowance survives every render and restart, and one with an `until`
is removed at its `until`, by the kernel in the loaded filter and by the holder in its state. This follows ADR 0683
(the firewall's control verbs, amending ADR 0170) and the operator's word of 2026-11-21 20:35 CET (the packet filter's
module has tools to control it beside the automatic provisioning from declarations), written from design 47 H's lead
answers H-A-304, H-A-305, H-A-307, H-A-308, H-A-312, H-A-314 to H-A-316, H-A-321, H-A-324 and H-A-325. The seat's verbs, its event and
its state are design 33 control-verbs; this part holds the render. Cited as "design 08 control-verbs §n".
**Where.** Built in design 47's milestone H, in one delivery group with design 33 control-verbs, three repositories:
the networking repository (the seat's contracts module), the catalog repository (the nftables module, the holder)
and the node-engine's repository (the compose of the filter, the verb `converged`). The ordering edge is the served
verb: the node-engine asks `allowed`, so the holder serving it comes first in the delivery plan. First use: tcp 8000
on the workstation's home-network interfaces.
## Lines
1. **The node-engine is the one renderer of the filter file** (H-A-305; P7). At every compose of the packet filter
(each compose it makes today, and each one line 3 starts) it asks `<node>/node-packet-filter.allowed` of its own
node, within its compose-ask bound, the node-engine's setting `compose-ask-bound`, default 10 s (H-A-314 (5)), and
renders the answer's allowances beside the derived ports. The node-engine's manifest invokes
`seat:node-packet-filter.allowed` and consumes `node-packet-filter.allowances-changed` (H-A-314 (3), H-A-315 (2)).
Nothing else writes the input chain: the holder writes no rule and loads the file as today (`reload`, ADR 0170
§3). The derived part is unchanged: the modules' `listens`, the hub's port, the foundation's ports and their
sources stay as §4 *What was built* says.
2. **Where an allowance is rendered.** In the mesh's own table (family `inet`, name `mesh`), its input chain, after
the derived ports' rules and before the chain's drop policy: one accept rule per allowance, matching its protocol
and destination port; its interface by name, or for the default set (resolved here at compose, H-A-308 (1)) every
input interface that is neither the private network's tunnel nor an outward link (a hub's internet-facing link
included; H-A-324), rendered as a negated match on the tunnel and on each outward link the node reports (§4, *It
filters what arrives from outside*), so an interface that comes up later is covered and neither the tunnel nor an
outward link ever is; the node-engine serves the same classification, the tunnel and the outward links, in its
verb `converged` (design 33 control-verbs §10; H-A-325), so one classifier of links serves the render and the holder; an outward link is opened only by an allowance that names it; its source when it has one, for the family of its CIDR, and no
source match when it has none (both families). An allowance with an `until` matches its port as the one element of
a set of the mesh's table with the `timeout` flag, the element's timeout the time left to its `until` at the
compose, at least 1 s, so the kernel closes the port at its `until` even while the holder is away (H-A-315 (4)); a re-render reloads the
filter and re-arms the time left, which is the right time (H-A-324).
Each rule carries the comment `allowance <key> <at>`, the allowance's key as its answers give it (without the node) and
the time it was set (design 33 control-verbs §7, §9). An allowance opens the input path only: a container's published port is
forwarded traffic and stays its module's `listens`' (§4, *It filters what arrives from outside*). An allowance and
a derived rule on the same port are both rendered; neither removes or narrows the other.
3. **The node-engine composes the filter again on every change of the allowances.** It hears the event
`node-packet-filter.allowances-changed` on the subject of its own node (design 33 control-verbs §8), asks `allowed`
and composes the filter; no timer and no poll. The event's body is not rendered: `allowed`'s answer is the one
input, so an event missed is made good at the next compose. The node-engine drops nothing past an `until` on its
own clock: the holder's answer is the truth (H-A-314 (6)).
4. **An unanswered `allowed` refuses the compose and keeps the loaded filter** (H-A-307 (1), H-A-314 (7), H-A-315
(6); the rule for a generator that cannot say what it opens, §4). When `allowed` does not answer within the
compose-ask bound or answers an error (the holder not running yet, its tool runner away, no holder on the node, a
rolled-back holder that does not serve the verb, a laptop asleep), the node-engine composes no filter file and
keeps the one loaded. The refusal is a withheld part in H-A-306's form: the node-engine seat's composition answer
gains the field `withheld-parts`, a list of `{part, why}`, set when a compose kept a part because its read did not
answer, here `part` `node-packet-filter.allowed`; it is added in the same contracts commit as the verb `converged`
(H-A-321). It is said once in the journal and in `<node>/node-engine.composition`, never "nothing opened". The
holder's module is not held for it, since an old holder that does not serve `allowed` must still be replaceable;
only the filter file resource is kept as it runs. The next `allowances-changed` (the holder says `started` when it starts serving, design 33 control-verbs
§8) or the next compose tries again; by hand it is recovered with the node-engine's compose verb, and the composed
file is loaded with `reload`.
5. **Allowances are rendered only into a converged node's own table** (H-A-307 (2), H-A-312 (1)). On an adopted node
the mesh loads no table that holds an accept (design 08 adoption §2) and speaks the found firewall's terms through
openings derived from what the filter would be derived from (adoption §3); an allowance is none of them. The holder
refuses `allow` and `disallow` there and `allowed` answers none (design 33 control-verbs §3, §6), so the
node-engine renders none. After the node converges, allowances are set with `allow` as on any converged node.
6. **An allowance is removed at its `until`** (H-A-308 (2), H-A-312 (2), H-A-315 (4)). In the loaded filter the
kernel removes its set's element at its `until` (line 2). In the holder's state the holder removes its key at its
`until`, says `allowances-changed` with change `expired` and writes a journal line (design 33 control-verbs §11);
the event composes the filter again (line 3), which leaves its rule out. A holder away at the `until` removes the
key at its next start, before it answers `allowed`; meanwhile the loaded filter is kept (line 4) and the port is
already closed by the kernel.
7. **An allowance survives every render and restart because it lives in one place.** The holder's state on the bus is
the only copy; the rendered file is output, rewritten whole at each compose and never read back. A restart of the
node-engine, of the holder or of the machine renders the allowances again from `allowed`; a delivery of either
module removes none. A running holder keeps the seat while the seat's definition grows (design 33 control-verbs
§13).
8. **How it is checked.** The node-engine's compose tests assert on the input chain's text: with no allowance the file
is byte for byte today's derived file; with one allowance of the default set, a rule with negated matches on the
private network's interface and on each outward link and no positive match on either; with three (one interface named, one with an IPv4 source,
one udp), three rules in that form after the derived ones, each with its comment `allowance <key> <at>`; one with an
`until` renders a timeout set whose element's timeout is the time left. A fake `allowed` that never answers within
the bound leaves the loaded file untouched and one entry in `withheld-parts` of the composition answer, `node-packet-filter.allowed` with its why, and the
holder's module still replaceable; an
`allowances-changed` for the node composes again and one for another node does not; an allowance the fake answers
with an `until` already past is rendered as answered, the node-engine dropping none on its own clock; an adopted
node renders none. Live, read in packets as §4 reads them: before `allow`, a machine on the home network is refused
on tcp 8000 at the workstation (this holds where the home-network links are outward in the node's render, as the
workstation reports them; H-A-314 (8), H-A-324); after `allow` of tcp 8000 with no interface the home-network
machine is still refused, since an outward link is not in the default set; after `allow` of tcp 8000 on the workstation's home-network interfaces it
reaches the port and a machine reaching the workstation over the private network does not; a node-engine restart
keeps it open; after `disallow` neither reaches it; an `allow` with an `until` a minute ahead closes at its `until`
with the holder stopped.
+36 -2
View File
@@ -8,8 +8,9 @@ code:
- mesh-sdk go/catalogue (the manifest's shapes)
- mesh-sdk go/declaration (the node-engine's resource vocabulary)
- mesh-contracts
updated: 2026-11-12
updated: 2026-11-18
decisions:
- "ADR 0680"
- "ADR 0578"
- "ADR 0595"
- "ADR 0619"
@@ -169,7 +170,7 @@ since a package is a build input and runs nowhere.
|---|---|---|
| **tools** | a `bundle` with `loads` naming what the runtime loads (for a Go bundle its `binary`), and the module's `tools` listing every tool it serves; `replaces` for the shell commands a tool replaces | the node's tool runner (ADR 0175); with `runs-as`, a **runtime of its own** under the module's account (below) |
| **a long-running service of its own code** | a `process` resource: `artifact` (a bundle), `name` (what the unit is called), `run` (the command, relative to the unpacked bundle), `user`, `env` and `env-file`, `restart-on`, `health`, `yields`, `replaces` (the ids it takes over from) | the node-engine, as a unit that stays up |
| **its own bus account for a process** | the process's `bus`: the `emits`, `state`, `queues`, `invokes` and `asks` it is granted, a subset of what the manifest declares; an own secret holding the account, owned by the process's `user` (ADR 0436 §1) | the grants part composes the account from it |
| **its own bus account for a process** | the process's `bus`: the `emits`, `state`, `queues`, `invokes` and `asks` it is granted, a subset of what the manifest declares; an own secret holding the account, owned by the account the process runs as, whatever that account (ADR 0680 §1; *Whose a process's own secrets are*, below) | the grants part composes the account from it |
| **a scheduled job** | a `process` or `container` resource with `schedule`, a five-field cron; never with `run-once`; a process with no `restart-on`, `replaces` or `witness`; a scheduled container may name `while-stopped` (below) | the node-engine, again on each fire (ADR 0053) |
| **a one-time step** (a seed, a first-start step, a validator of a file) | a `process` or `container` resource with `run-once` true: a **run-once step**, which gates every resource after it; `restart-on` makes it run again when what it names changes (ADR 0052, ADR 0099) | the node-engine, to completion |
| **a migration of its own state before a new version runs** | `prepares` true, and a `container` or `process` that stays up running an artifact the module built: the mesh runs that program in its preparation mode (ADR 0135); never a run-once step written by hand for it | the node-engine, before the new version starts |
@@ -178,6 +179,39 @@ since a package is a build input and runs nowhere.
| **a host program from the operating system** | a `package` resource (`package`, `absent`) and a `service` resource per unit it ships (`unit`, `state`, `boot`, `scope`, `restart-on`, `reload-on`, `health`); never an `action` | the node's package manager and service manager |
| **files** | `file` and `directory` resources (a directory by `place` or `path`), an `archive` resource unpacking an `archive` artifact, template files in the module's `templates` directory | the node-engine and the provisioner client |
**Whose a process's own secrets are** (ADR 0680). *Amended 2026-11-18.* Every own secret a process reads, its bus
credential first, is owned by the account that process runs as, whatever that account is: an account the module
makes (a `user` resource of that name), the operator account (`${machine:account}`), or root (ADR 0680 §1). The
manifest names it, as the secret's `owner` or the module's `secrets-owner`, and the process runs as it (its `user`)
or as root, as the second rule below holds; the node-engine writes the secret owned by it at mode 0600 (ADR 0680 §1). An account the module makes stays
the usual case, not a rule; a process that declares `bus` still names the account it runs as. ADR 0436 §1's "never
root and never the operator account" no longer holds for a process that declares `bus`, and the manifest validator's
two rules that refused them (as an own secret's owner and as a `bus` process's account) are gone (ADR 0680 §2). The
manifest validator holds two rules instead:
- **Every account named is one the mesh knows.** An own secret's `owner`, a `secrets-owner` and the account of a
process that declares `bus` are each the operator account, root or an account the module makes; any other is
refused, and so is a `bus` process naming no account (ADR 0680 §2).
- **A process can read every own secret it names.** A process resource that names an own secret (its path in the
process's `run` or `env`) and runs as neither that secret's resolved owner (its `owner`, else `secrets-owner`) nor
root is refused (ADR 0680 §3). A container module is excepted, since a container reads its secrets as its own
account, and a process account's secret needs no such rule, since it is delivered owned by the process's account
(ADR 0436 §2).
A secret owned by the operator account is as private as what runs as that account on its node: the node's tool
runner, and through it every agent session there (ADR 0436 §9). That exposure is accepted (ADR 0680 §1); what bounds
a process account's credential is its grants, which answer no tool and no seat's verb (ADR 0436 §3). A module's
`runs-as` is unchanged: it names an account the module makes (ADR 0259 §8, below).
*How it is checked:* the node-engine's tests write an own secret owned by the account its manifest names, at mode
0600, for an account the module makes, for the operator account and for root; the manifest validator's tests, run on
every change of the validator and over every manifest a change of mesh-catalog touches, accept an own secret owned by
`${machine:account}` or root, a `bus` process running as either, a process running as the resolved owner of the own
secret it names (also through `secrets-owner`), a process running as root, and a container module naming another
account's own secret, and refuse a `bus` process naming no account or one the module does not make, and a process
naming another account's own secret; the controller's grant test still finds no process account able to answer a
call (ADR 0680 §4).
**A runtime of its own.** A module that says `runs-as` names an account it makes (a `user` resource of that name,
never root nor the operator's); its tools bundle is then served by a runtime of its own, as that account, on a bus
account of its own (ADR 0259 §8), whose readiness is the manifest's `runtime-health` (ADR 0366). Its secrets belong to
@@ -0,0 +1,147 @@
---
layer: to-be
design: 33
milestone: control-verbs
updated: 2026-11-21
code: [mesh-networking modules/mesh-connectivity-contracts, mesh-catalog modules/nftables, mesh-host (the node-engine seat's verb converged)]
decisions:
- "ADR 0683"
- "ADR 0170"
- "ADR 0168"
- "ADR 0565"
- "ADR 0595"
- "ADR 0625"
- "ADR 0636"
- "ADR 0246"
---
# Design 33, milestone control-verbs — the packet filter seat's allow, disallow and allowed
**What it builds.** The node seat `node-packet-filter` serves, beside `rules`, `reload` and `remove` (ADR 0170), three
verbs that control the packet filter by hand: `allow`, `disallow` and `allowed`; `rules` lists the allowances beside
the ruleset; and the seat says every change of its allowances as the event `node-packet-filter.allowances-changed`.
The node seat `node-engine` gains the read verb `converged`. An **allowance** is a port opened by the operator, through
the seat, beside the ports derived from the modules' `listens`. This follows ADR 0683 (the firewall's control verbs,
amending ADR 0170) and the operator's word of 2026-11-21 20:35 CET, written from design 47 H's lead answers H-A-304,
H-A-305, H-A-307, H-A-308, H-A-310, H-A-312, H-A-314 to H-A-321, H-A-324 and H-A-325; P7, P11 and the instruction that tools are
how a module is spoken to (every act by hand has its verb, every kept thing its read). How the node-engine renders
allowances is design 08 control-verbs. Cited as "design 33 control-verbs §n".
**Where.** Built in design 47's milestone H, one delivery group with design 08 control-verbs, three repositories: the
networking repository (the seat's whole definition in its contracts module), the catalog repository (the nftables
module, the holder) and the node-engine's repository (the verb `converged`, and the render of design 08
control-verbs). mesh-sdk is not in the group: its TypeScript `state()` asks the runtime's state methods as a Go bundle
does (H-A-314 (4)).
## Lines
1. **The seat serves six verbs, and every holder serves all six** (§3). `rules`, `reload` and `remove` as ADR 0170
§1 to §7 say, with `rules`' answer widened by line 5; `allow`, `disallow` and `allowed` as lines 2 to 4 say. The
three are required from the start, in the one delivery group that adds them: §7's three steps (ADR 0246) end for a
seat defined in a contracts module (ADR 0683; H-A-310), since the group's check builds the definition and its
holder together. Each is a node verb, called `<node>/node-packet-filter.<verb>`. Their classes: `allow` and
`disallow` are class `act`, by-caller; `allowed` is class `read` (H-A-315 (2)).
2. **`allow` opens a port by hand.** Arguments: `port`, an integer 1 to 65535, required; `protocol`, `tcp` or `udp`,
default `tcp`; `interfaces`, a list of interface names, each 1 to 15 characters of letters, digits, `_`, `.` and
`-`, default none, which means the default set: every input interface that is neither the private network's tunnel nor an outward link (a hub's internet-facing link
included), so an outward link is opened only when named here (H-A-324; design 08 control-verbs §2);
`source`, one IPv4 or IPv6 CIDR, default none, which means any; `until`, an RFC 3339 time in the future, optional;
`why`, 1 to 500 characters, required. Any other argument is refused. Who set it is the caller as the bus names it on
the request, never an argument; for a session's call that is the tool runner's account (H-A-315 (3)). When is the
holder's clock at the write. It keeps one allowance per protocol, port and interface (H-A-308 (1)): one key per
named interface, each with the same source, who, when, why and `until`, or for none the one key of the default set
(line 7), resolved by the node-engine at compose. An `allow` for a key that exists replaces it whole and answers
what it replaced.
3. **What `allow` refuses**, naming the argument and writing nothing: a caller whose proof is `unproven` (H-A-315
(3)); a node that `converged` (line 10) says is adopted, naming the adoption, or a `converged` that does not answer
(line 10); a port out of range or not an integer, another protocol, a bad interface name, a source that is not a
CIDR, an `until` not in the future, an empty or longer `why`; unless a `source` is given, a port that the loaded
input chain accepts only with a source match (`ip saddr` or `ip6 saddr`) and never from any source, read from the
chain as `rules` shows it, which covers the store's port, the broker's management port and a port opened to the
local network only (H-A-315 (7), H-A-319, ADR 0683 §3; a comment naming each derived rule's scope, so the
holder reads scope and not shape, is owed to milestone H2 and not built here); and a write that would make more than `allowances-max` allowances on the node, the
holder's setting, default 100 (H-A-314 (5); the bound keeps the rendered chain readable and `allowed` in one
answer).
4. **`allow` and `disallow` answer what they did, and are written in the hand-act log.** When a named interface is
an outward link, as the `outward` list of the `converged` answer it asked for this call names them (line 10),
`allow`'s answer sets on that kept allowance the optional field `outward`, the outward link it opens (H-A-324,
H-A-325). The holder reads no shape of the loaded table to tell an outward link from the tunnel. `allow` answers each
allowance as kept, with its key and what it replaced, and that `allowances-changed` (line 6) was said; `disallow`
answers each allowance as removed and that the event was said (H-A-308 (3)). Whether the loaded chain carries an
allowance is read with `rules` (line 5). Each call of either is written in the hand-act log (ADR 0565 §10; H-A-315).
5. **`disallow` takes back an allowance.** Arguments: `port`, `protocol` and `interfaces` with `allow`'s forms and
defaults, which name the same keys (no `interfaces`: the default set's key), and `why`, required as for `allow`;
any other argument is refused, the source among them, since it is a property and not part of an allowance's
identity. It refuses as `allow` does an `unproven` caller, an adopted node and an unanswered `converged`. It deletes
every key named, or, when one does not exist, refuses naming it and deletes none, and says `allowances-changed` with
change `disallowed`, its who and its why.
6. **`allowed` reads every allowance.** No arguments. It answers every allowance kept for its own node only (line 7),
each with its key, `port`, `protocol`, `interface` (absent for the default set), `source` (absent for any), `until`
(absent for none), `why`, `by` (who) and `at` (when). The node-engine is its caller at every compose of the filter,
within its compose-ask bound (design 08 control-verbs §1). On an adopted node it answers none (H-A-312 (1)), so no
answer says a port is open that is not.
7. **`rules` lists the allowances beside the ruleset.** Its answer keeps the filter as the machine enforces it (ADR
0170 §2) and adds `allowances`: `allowed`'s answer, each marked `loaded` only when a rule of the mesh's table carries
its comment `allowance <key> <at>` with an equal `at` (design 08 control-verbs §2), and not loaded otherwise; and
every loaded allowance comment whose key is not kept, listed as "loaded, not kept" (H-A-315 (5)). So a kept
allowance the loaded filter does not carry, and a loaded one nobody keeps, are both visible in one call.
8. **The event `node-packet-filter.allowances-changed`.** The holder says it after every write or delete of its
state, in the seat's family `node-packet-filter.events` on the shared events stream, `max-age-days` 7, its subject
`allowances-changed.<node>` (as the node-engine seat's `composed.<node>`; H-A-314 (2)). Its body: `node`; `change`,
one of `allowed`, `disallowed`, `expired` and `started`; `allowances`, the ones concerned (all of them for
`started`); `by` and `why` for `allowed` and `disallowed`; `at`. The holder says `started` once each time it starts
serving the seat, so a node-engine that refused a compose for want of an answer composes again. The node-engine
of that node hears it (design 08 control-verbs §3); nothing else needs to.
9. **The holder keeps the allowances in its own state on the bus, one bucket per machine.** A state `allowances`
asked for in the holder's manifest's `requires` by the template `state` (ADR 0595 §1, §2) with the template's
option of one bucket per machine, `history` 1, reached only by its owner; a node's holder answers its own node's
allowances only (H-A-315 (1)). The per-machine option is one bucket whose keys carry the node, `<node>` and
`<node>.>` reached only by that machine's holder, so each holder reads and writes its own node's prefix only
(H-A-317). One key per allowance: `<node>.allow.<protocol>.<port>` for the default set (line 2), `<node>.allow.<protocol>.<port>.<interface>` for a named one (the interface is every token
after the fourth, since a name may hold a dot). The allowance's key in every answer, in the event and in the rule's
comment is the same key without the node, `allow.<protocol>.<port>[.<interface>]`. Its value: `port`, `protocol`, `interface`, `source`, `until`,
`why`, `by`, `at`; the source is a property of the value, never part of the key. Only `disallow` and the removal at
its `until` (line 11) delete a key; no render, restart or delivery does. The bus module's own tools read the state's
fill against its cap. The holder's manifest serves the six verbs and requires the state; it does not list the event in
`emits`, which the manifest check refuses for a claimed seat's event: the claim of `node-packet-filter` says the
event, and the holder says it with the seat's publish (H-A-314 (3), H-A-317).
10. **The node seat `node-engine` gains the read verb `converged`** (H-A-316). It answers the node's convergence
state, which the node-engine keeps (ADR 0168): `state`, `adopted` or `converged`; `since`, when it became so, optional and absent for a record written before this change (H-A-324); for an adopted node `found`, the found firewall's kind (H-A-321); and the node's links as the node-engine
classifies them for its render: `tunnel`, the private network's interface name, and `outward`, the list of the
node's outward links (H-A-325; design 08 control-verbs §2). Class `read`, built in the group's node-engine part, added
to the node-engine seat's whole definition. The holder asks it at its start and at every `allow` and `disallow`,
bound 10 s (the default of the node-engine's compose-ask bound), and refuses when it does not answer. When it
holds no answer (its start's ask did not answer and no `allow` or `disallow` asked since), `allowed` asks
`converged` again, bounded by the same 10 s, and refuses when that does not answer, so the node-engine keeps the
loaded filter (H-A-318). It never assumes either state and recovers without a restart.
11. **An allowance is removed at its `until`, by the holder.** The holder keeps an in-process timer per allowance with
an `until`, armed from the watch of its own state at every start (H-A-314 (6)). At an allowance's `until` it
deletes the key, says `allowances-changed` with change `expired` and writes a journal line naming the
allowance (H-A-308 (2)); no removed entry is kept, so `allowed` never answers one. At its start it first deletes
every key whose `until` has passed, with the same event and journal line, then says `started`, then answers
`allowed` (H-A-312 (2)). The kernel closes the port at its `until` even while the holder is away (design 08
control-verbs §2).
12. **The seat's definition becomes whole** (H-A-307 (3), H-A-314 (2); ADR 0625 §1, ADR 0636 §5, §6). It lives in the
Connectivity domain's contracts module, `mesh-connectivity-contracts`, in the networking repository: every one of
the six verbs with its class, its arguments and its answer form as lines 1 to 7 say, the family and event of line
8, the state's value of line 9 as a layout, and the generated Go and TypeScript packages. Its model is the
`node-engine` and `git` seats' whole definitions. The node-engine calls `allowed` and hears the event
through the Go package; the holder serves the verbs and says the event through the TypeScript one.
13. **A running holder keeps the seat while its definition grows** (H-A-315 (6)). A holder running before this group
is not withdrawn when the seat's definition gains the three verbs; it holds the seat until it is replaced. While a
node's holder does not serve `allowed` (a rollback, a laptop asleep), the node-engine refuses the filter's
composes and keeps the loaded filter (design 08 control-verbs §4).
14. **How it is checked.** The holder's tests, against a state in memory: `allow` then `disallow` writes the key, then
deletes it, with an `allowed` and a `disallowed` event, both answers saying the event was said, and two hand-act
log lines; port 0, 65536 and `80a` are refused, as are protocol `icmp`, an `until` in the past, an empty `why`, an
unknown argument, an `unproven` caller, the 101st allowance at the default `allowances-max`, and the store's port
with no source when the fixture chain accepts it only with a source match (accepted with one); `allow` with no `interfaces` stores no interface, so neither the private network's
interface nor an outward link is ever in the defaults; with a fake `converged` whose `outward` lists `eth1`, `allow` naming `eth1` answers that kept allowance with
`outward` `eth1`, and one naming another interface answers it without the field; `allow` with two interfaces keeps two keys; a second `allow` of one key answers
what it replaced; an allowance is removed at its `until`, says `expired`, writes its journal line, and
`allowed` then answers without it; a start with a key past its `until` deletes it and says so before `started` and
before its first `allowed`; with a fake `converged` saying adopted, `allow` and `disallow` are refused naming the
adoption and `allowed` answers none; with one that never answers, both are refused and `allowed`, holding no answer, asks again and refuses; an allowance kept on one node
is absent from another node's `allowed`, and its key in the bucket carries its node while its answer's key does not; `rules` over a fixture ruleset marks an allowance loaded only on an equal
`at` and lists a comment with no kept key as "loaded, not kept". The node-engine's test of `converged` answers both
states. The seat's contract test holds every holder to the six verbs, and the generated packages to the definition.
@@ -1,9 +1,10 @@
---
layer: to-be
status: implemented
status: in-progress
code: [mesh-controller, mesh-tools]
updated: 2026-10-07
updated: 2026-11-21
decisions:
- "ADR 0683"
- "ADR 0246"
- "ADR 0170"
- "ADR 0179"
@@ -214,6 +215,20 @@ container to declare a capability. Removing a predecessor's rule set is an opera
through the seat, recorded on the bus, instead of a shell on the machine. *How it is checked:* ADR
0169's table.
*2026-11-21, ADR 0683 (the firewall's control verbs, amending ADR 0170); the operator,
2026-11-21: the packet filter's module has tools to control it beside the automatic provisioning from declarations.*
The seat serves three more verbs: `allow` (a port 1 to 65535, protocol `tcp` by default or `udp`, interfaces by
default every input interface but the private network's tunnel and the outward links, a source CIDR by default any, an optional `until`, a
required `why`), `disallow` (the same keys) and `allowed` (every **allowance** with who set it, when, why and its
`until`, at which it is removed); `rules` lists the allowances beside the ruleset. The node seat `node-engine` gains
the read verb `converged` (adopted or converged, since when, the found firewall's kind), which the holder asks before
it keeps an allowance. The holder keeps them in its own state on the bus and says
every change as the event `node-packet-filter.allowances-changed`, which the node-engine hears to compose the filter
again; the node-engine stays the one renderer (design 08 §4, *By hand, beside the derived ports*). The seat's
definition becomes whole in its domain's contracts module. Every holder serves the six verbs. Every line is in
[design 33 control-verbs](33-milestones/control-verbs.md), from design 47 H's lead answers
H-A-304, H-A-305, H-A-307, H-A-308, H-A-310, H-A-312 and H-A-314 to H-A-316; *how it is checked* is its line 14.
## The intrusion seat's verbs, 2026-10-02
ADR 0179.
@@ -226,6 +241,15 @@ per-module container ADR 0175
retires; the verbs and the client are the same code once the node's own runtime loads them as a bundle.
The module's own tool beside them reads one jail's effective settings. *How it is checked:* ADR 0179's table.
## Milestones
Each milestone not yet delivered has its part, cited as "design 33 <milestone> §n"; what is above was built before
milestone parts existed.
| Milestone | Part | State |
|---|---|---|
| control-verbs: the packet filter seat's `allow`, `disallow` and `allowed`, built in design 47's milestone H (ADR 0683) | [control-verbs](33-milestones/control-verbs.md) | to build |
## What this does not settle
- Which verbs each seat should serve. That is a decision per seat, and the reason to do it slowly: a
@@ -3,9 +3,10 @@ layer: to-be
status: in-progress
design: 42
milestone: zfs-owns-zfs
updated: 2026-11-13
updated: 2026-11-18
decisions:
- "ADR 0679"
- "ADR 0680"
---
# 42, milestone zfs-owns-zfs — the lead's answers to the prepare's questions
@@ -136,6 +137,8 @@ settles it (`00-META/principles.md`), cited as "design 42 zfs-owns-zfs A-n". The
calls: emit an event, put or delete a key of those states, and invoke one of those three. The root processes say
everything they say on the bus through it; with the keeper not answering they act as with the bus down (A-8), and
warn. The module's tools keep their own account as today.
*Amended 2026-11-18:* ADR 0680 §1 ends ADR 0436 §1's "never root" for a process that declares `bus`, so this
answer's premise no longer holds; the keeper stands as the milestone builds it, a choice the records no longer force.
- **A-38** (A-37, §25, §26, met in develop; ADR 0436 §1, ADR 0595 §1, §2; the operator's "a case the delivery cannot
handle is a gap in the build and delivery system") A process's `bus` today is judged and granted only against the
module-level `emits` and `state`, which ADR 0595 replaced by what a module asks for in `requires`. In this
@@ -2,8 +2,10 @@
layer: to-be
status: in-progress
code: [mesh-controller, mesh-catalog modules/mesh-delivery-gate, mesh-catalog modules/gitea, mesh-delivery, mesh-host, mesh-sdk, mesh-vault, mesh-contracts, mesh-tools, mesh-networking, hq 00-META/checks]
updated: 2026-11-11
updated: 2026-11-20
decisions:
- "ADR 0682"
- "ADR 0681"
- "ADR 0673"
- "ADR 0672"
- "ADR 0668"
@@ -195,6 +197,16 @@ acting parts, the kept data on the bus restored with a delivery, and the verbs a
section this file names in italics and does not hold is there, under the same name. A sentence that holds a delivered
line and G's change together stays here.
**From milestone H's delivery, a delivery starts at the merge through the mesh** (ADR 0682; the operator,
2026-11-20). One **merge call** of the seat `git`'s verb `merge-pull`, naming one or more pull requests, is a delivery's
start and its group; the gate names the merged heads one source revision per watched branch and opens its delivery
record, the indexer takes it in and names its **movers**, one pipeline runs per take-in and none in parallel, and a
want names the newest released version whose pipeline passed. The forge bridge's one message per move of a watched
branch's head is the gate's watchdog: a head no merge call delivered is an **untaken head**, said and delivered by
nothing. Every line milestone H builds is in its part, opened by the plan
[47-milestones/H-deliver-the-delivery-system.md](47-milestones/H-deliver-the-delivery-system.md) and cited as "design
47 H §n" (§734 on); where an H line and a G line differ, the H line holds and the G line says so.
This is the delivery of [to-be 10](10-delivery.md) carried out. To-be 10 says what is built: the comparison
of source and artifacts decides, and nothing here changes that. This design says who answers for one
commit's journey, in what order the steps are allowed, and where each fact of it is kept. It was rewritten on
@@ -1508,6 +1520,23 @@ way is written down before the switch (ADR 0526, which ADR 0522 §6 asked of R4g
running versions are kept outside the artifact store on two nodes, each by digest with its module, artifact, commit, Go version and the
steps that put it back, because the store's nightly collection removes a blob the controller did not record (ADR 0526 §6).
## Every delivery part wakes on its inputs alone
*ADR 0682 §11; issue 1982; the plan of milestone H, section 3.6.* A delivery part (the forge module, the forge
bridge, the gate, the indexer, mesh-config, the runner, the builder, the build modules, the assigner, the node-engine,
the grants part, the watchman, the condition keeper, the status view, the teller and the tool runner) keeps a watched
copy of each of its own states, through one shape in mesh-sdk's state package (the shape the held pull request
mesh-catalog #708 built for mesh-config, moved into mesh-sdk), and runs a pass only when an input it consumes changed,
taking only the keys that input names. It keeps one timer, set to its earliest real deadline, a bound it states; it
polls nothing, reads nothing of its own whole on a wake, and asks no call again on a timer. Which part moves onto the
shape in which release, and its cost (the shape's release in mesh-sdk rebuilds nearly every module, so H4 alone
carries it), is design 47 H §762. How it is checked: the code review of every delivery module from milestone
H1 on refuses a timer that is no stated deadline, a poll and a whole re-read; mesh-sdk ships a test helper that counts
the requests a pass makes, and each delivery module's integration test that starts it whole against the bus (the mesh's
instruction *how-work-is-done*) asserts that a pass after one changed key requests that key alone; live, milestone H4 is proven by the bus's
slow consumers and consumers behind at 0 for an hour, read with `mesh-broker.rates` (H §760) and the bus module's
consumer reads.
## Acceptance tests
What the build proves before this design is implemented (research 055's direction):
@@ -1664,7 +1693,7 @@ list:
performance is fixed; no check starts real builds.
37. **Nothing is delivered ahead.** Research 085's milestone is delivered with no part delivered ahead of it, read
from its delivery record (ADR 0656, How it is checked).
38. **A newer indexer is reached.** A merged revision naming a newer indexer gives, from the running indexer, a catalog
38. **A newer indexer is reached.** *(Retires at milestone H, design 47 H §764.)* A merged revision naming a newer indexer gives, from the running indexer, a catalog
graph revision that is the previous one with only the indexer's entry replaced; no other want moves; the new
indexer is wanted, granted, started, and takes the revision in.
39. **Revisions are kept, and stores say their fill.** A revision pinned by two records stays until both free it, one
@@ -1706,7 +1735,7 @@ Each change ADR 0656 makes in an earlier record, and its test:
are read by the indexer's candidate pass and by no running part (whole).
48. **A process reports its handed build** (ADR 0639 §7). A part started while its node's want differs from what runs
reports the build it was started with, and the node-engine seat's verb `process-builds` reads it (whole, node-engine).
49. **The forge bridge reads a group from the gate's verb** (ADR 0633 §4, §5). The forge bridge sets the group on each
49. **The forge bridge reads a group from the gate's verb** *(Retires at milestone H, design 47 H §764.)* (ADR 0633 §4, §5). The forge bridge sets the group on each
merged message and publishes a group's end again from the seat `mesh-delivery`'s verb `groups`, and its account
holds no read of the gate's `merges` state, so a read of it is refused (whole, forge bridge).
@@ -1728,11 +1757,9 @@ What milestone G's prepare found, and its test:
(whole, gate, checker and forge module).
53. **What is owed is said, answered and kept.** An `owed` said while the forge bridge is stopped is performed once after
its restart; the bridge stopped between an act and its `owed.ended` repeats it and the forge holds it once; an
undecodable act is said `undecodable`; an act unanswered 15 minutes is listed by `stalled`; the switch with a merge
handed and not yet merged opens no `forge-made` record; a merge made at the forge opens one, said as `transition`
with why `forge-made`, and a later merge in its repository follows it; the checker's `mesh/check` is performed by
the bridge from the checker's `owed`, and a checker status the forge refuses is recorded in the checker's own
state and listed in its stalled read (whole, gate, checker and forge bridge).
undecodable act is said `undecodable`; an act unanswered 15 minutes is listed by `stalled` (whole, gate and forge bridge). The rows of a `forge-made` record
and of the checker's `mesh/check` went: no record is marked `forge-made` (design 47 G §151), a head moved outside
the mesh is an untaken head (design 47 H §740), and the checker goes (corrected in place by design 47 H §763).
54. **One reader of the seat list per part, and one view machinery.** Every part that reads the seat list has the two
tools of ADR 0636 §7 from the runtime and no wiring of its own; every part that takes a first view answers its
held and awaited revision in `<module>_first_views`. With the forge on its old build, serving no verb of the seat
@@ -1865,9 +1892,9 @@ No module ends the milestone with more tests than it starts with, counted per ca
The failure paths:
81. **A group that never completes** raises `take-in-waits` at 30 minutes, and none of its members is delivered (whole,
81. **A group that never completes** *(Retires at milestone H, design 47 H §764.)* raises `take-in-waits` at 30 minutes, and none of its members is delivered (whole,
indexer and watchman).
82. **An indexer whose manifest its validator refuses** leaves its revision pending, holds only its own watched branch,
82. **An indexer whose manifest its validator refuses** *(Retires at milestone H, design 47 H §764.)* leaves its revision pending, holds only its own watched branch,
raises `take-in-waits` at 60 minutes, a later merge on that branch waits behind it and is not taken in past it, and
a revert taken in moves the branch on (whole, indexer).
83. **A declined registration change, and an unprotected branch,** each write nothing: the first after its Decline, the
@@ -1953,13 +1980,13 @@ ADR 0667's checks, of milestone G itself:
package rebuilds it, started by the runner after its `outdated` ask, one that leaves its digest as it was rebuilds nothing, and a change of the Go build module's
own code rebuilds no module; a `beta` build's entries compare by digest (whole, indexer, builder and runner,
mesh-delivery).
105. **One query finds what is behind** (design 47 G §375). With the runner stopped while mesh-sdk `0.1.16` is
105. **One query finds what is behind** *(Retires at milestone H, design 47 H §764.)* (design 47 G §375). With the runner stopped while mesh-sdk `0.1.16` is
released, `outdated` names the module's kind-1 edge at `0.1.15` against `0.1.16`, and at its restart the runner
starts that module's pipeline once, keyed by mesh-sdk's source revision; `runner_catch_up` does it by hand and
writes the hand-act log
(whole, indexer and runner, mesh-delivery). After the first fill (§377), a mesh running its newest builds is
answered with nothing behind.
106. **A delivery is waited for mesh-wide** (design 47 G §376, §379, §380). A module ordered after a provider's
106. **A delivery is waited for mesh-wide** *(Retires at milestone H, design 47 H §764.)* (design 47 G §376, §379, §380). A module ordered after a provider's
new version is wanted only once `version-delivered` says every present machine placing the provider runs it; with
one machine away the module's status reads "waiting for delivery of" the provider, naming the machine, and it
catches up when that machine is back; with every placing machine away the provider reads "waiting for <machine>
@@ -1972,9 +1999,85 @@ ADR 0667's checks, of milestone G itself:
`condition-raised` (whole, condition keeper and a fixture observer, mesh-catalog and mesh-sdk). Live: after G's
delivery, a node-engine restarted with the full seat list clears its `seat-list-unread`, read with
`mesh-conditions.conditions`.
108. **One composition run per candidate and machine** (design 47 G §718 to §723, §727 to §733). Two composition
steps of one candidate started at once make one key, one request and one run ask; three steps of one revision on
two machines start two runs, each step reading both verdicts. A machine whose verdict blames one module fails that
module's composition step alone and says it as a warning in the others; a verdict with no `blamed` is judged as
today, naming no culprits. A run with no `step-ended` within its bound ends `could not run` with its readers; steps
asked again together start one run per machine that did not pass and move every former reader, a reopened
pipeline saying `pipeline-ended` again; a run that keeps saying its progress past 30 minutes is not ended, and a
taken ask silent for 30 minutes ends `could not run`. Both drop tools end
the readers, and a reader-only account calling one is refused (whole, assigner and the node-engine's composer,
mesh-delivery and `mesh-host`). Live: in the first publish after G's delivery, the composition run asks of a
revision number at most its machines, read with `assigner_compositions` and `assigner_inputs`.
109. **One test run per level and tested packages** (design 47 G §724 to §733). A module whose program and tools
bundle test one directory runs its unit and integration tests once per level, both artifacts' pipeline steps
ending with that result and naming it in `shared`, a failing test failing once, and `pipeline-ended` counting the
module's tests once. A rerun that passes moves both artifacts' steps and the pipeline ends again; a runner
restarted during a run takes it up again; a replaced reader leaves the ask to the others; at the cap a step ends
`could not run` naming the fill; a second revision with the same digests runs a failed test again; a test run queued past five minutes with a
holder present is not failed (design 47 G A-466y) (whole, runner
and Go build module, mesh-delivery and mesh-catalog). Live: each module's test steps run once per level, read with
`runner_test_runs`. A reader no part ends raises `pipeline-stands` within five minutes of its last sign of life
(whole, watchman; design 47 G §732).
110. **One merge call is delivered end to end with no hand act** (design 47 H §734 to §738, §743, §746, §750, §751,
§771). This is the live proof of the chain (the plan's section 5, step 7), read with the tools and quoted with its
times, not one test: with `pipelines-at-once` at 1 and one placing machine away, a merge call naming one pull
request that changes one module says one `merge-called`; the gate writes one source revision and opens one record;
the indexer says one `take-in.taken` naming that module alone; the runner runs one pipeline; once it passes and the
version is released, the module is wanted and runs on every present placing machine, the record ends naming the
machine away, and that machine applies its want when back. Each part's half is a whole test in its own repository
(H-A-17): the forge module's (§734 to §736, §766), the gate's (§737, §738), the indexer's (§743), the runner's
(§746), the assigner's (§750, §751) and the node-engine's (§771).
111. **A head moved outside the mesh delivers nothing and is said** (design 47 H §739 to §742, §766). A merge made in
the forge's screen gives one `head-moved`, no source revision and no pipeline; `head-untaken` is raised within
`untaken-bound`, naming the repository, branch and head; a later merge call in that repository clears it;
`gate_head_take` delivers such a head by hand, written in the hand-act log; and a forge module stopped between its
last merge and the write of that member ends as an untaken head recovered so (whole, forge bridge and gate in
mesh-delivery and mesh-catalog; the stop case in the forge module's whole test).
112. **One take-in is one pipeline, and pipelines run one at a time** (design 47 H §743, §746 to §748). Split per
module (H-A-17): the forge module's whole test that a call naming a mesh-sdk change and an importer merges both or,
with one refused, neither; the indexer's that its `source-moved` names mesh-sdk, that importer and mesh-sdk's other
importers as movers; the runner's that they are one pipeline in which every importer builds after mesh-sdk's
`package-published`, that a second take-in waits and is placed on the first's end, on a cancel and on a set-aside,
that a pipeline waiting on its `approval` step does not count and is placed first when answered, that a newer
take-in moving the same module supersedes it in the waiting pipeline, that `pipelines-stalled` is raised when
pipelines wait and none runs, and that a restart starts no pipeline of its own.
113. **A want is the newest released version whose pipeline passed, written tier by tier** (design 47 H §750, §751,
§771, §772). The assigner's whole test: a module released with a passed pipeline is wanted on every placing
machine whatever build an acting part runs; a version whose pipeline failed is not wanted; an older end heard after
a newer one writes nothing; a placement `build` names the want's version; a consumer's want is written only after its
provider's tier runs, and a tier that does not run within `tier-bound` raises `tier-late`; a node-engine without the
want-by-version capability is said held. The node-engine's whole test: the bus module's and the backup holder's
changed wants apply first, then by `rank`, its own last.
114. **A head whose `required` step failed is refused** (design 47 H §735). The forge module's whole test: a runner's
`step-ended` naming a failed `required` step at a head's fingerprint makes `merge-pull` refuse that head, naming the
step; a later passed run asked again by the operator lets it merge.
115. **A member refused mid-call is said and raised** (design 47 H §767). The forge module's whole test against a
fixture forge that refuses the second merge: one `merge-called` with the first member and `refused`, `merge-partial`
raised by the call, cleared by a later call landing the refused member and by `forge_partial_clear`.
116. **A take-in held on a full store** (design 47 H §744). The indexer's whole test: with the catalog store full a
`source-moved` is held unacknowledged, said once as `take-in.pending`, not redelivered on a timer, and taken in on
`room`; another branch's take-in goes on meanwhile; `indexer_take_in_skip` ends it in the hand-act log.
117. **`validate` refuses a manifest problem and a cycle** (design 47 H §745). The assigner's and the runner's whole
tests: a take-in whose manifest names an unknown field fails `validate` with `manifest-refused` naming it and asks
no composition; one that makes a cycle fails it with `cycle` naming the cycle; the indexer's manifest runs no
validator.
118. **Every account calls every verb; a module publishes only what it declares** (design 47 H §752). The grants
part's whole test: a module calls a verb its manifest does not invoke and is answered, publishes an event its
manifest does not declare and is refused, and an enrolling node's account is held to its own grant.
119. **The builder's lines** (design 47 H §753 to §758). The Go build module's whole test takes the delivery queue
first; the builder's whole tests: a done, unreserved fingerprint asked by a build step ends carried with a version,
a failed 0.1.19 built again is 0.1.19, a catalog entry heard with no step ask builds nothing, `build.ended` names its
version; the runner's: a step with nothing to run ends `nothing to run` and its pipeline passes; the indexer's: a
registration with every step `bypass` is refused, naming them.
## What retires
- From milestone H's delivery: everything design 47 H §764 lists (the per-pull-request paths of the gate and the forge
bridge, the catch-up by comparison, the runner's sweep, the start-time `outdated`, the checker, the groups by branch
name, the want holds A-478 ends), and the stale state H §765 drops once.
- The "release branches" and the controller's check of them (ADR 0358, superseded by ADR 0401).
- The toolchain image and its module (ADR 0405).
- The window cutter, the merge catch-up timer, the rollout ticker, mesh-delivery's tick, reconcile and
@@ -2068,6 +2171,7 @@ release reports the controller's code lines against research 063's baseline (acc
| L7 | History and records leave the controller (ADR 0571): the builder keeps its build log, one entry at a build's start and one at its end, from which build history and durations are read and the module mesh-catalog asks the builds it missed; the artifact sweep is the seat `mesh-artifact-sweep`, held by artifact-store-tools beside the artifact store, its keep set from the store's own listing, the wants and the build log, sweeping at each build's end from the lease write and recording each deletion in the hand-act log; disk readings are disk-load's own state on every node; who holds which seat is the state `seat-holders` of the seat `mesh-placement` (now `mesh-module-assigner`), and tools are found by the tool runners' announcements; what a node runs is its want and its have; the grants part writes `grants.<module>.<fingerprint>` and keeps memberships in its own state; root-free is a probe of the self-check; no manifest but the controller's reads, invokes or consumes `mesh-controller`, but a named list kept to the lease write, so the lease write's check passes (ADR 0189, ADR 0547 §11, ADR 0549 §6) | — | mesh-sdk, mesh-controller (its pin only), mesh-delivery (builder, indexer, assigner, grants part, view), mesh-catalog (artifact-store-tools, disk-load, the backup seat's holder, mounts, nfs-server, minio, mesh-watcher, the messenger, the self-check, the module mesh-catalog, docker), mesh-tools (the tool runner) | 1955 |
| B0 | The bootstrap delivery: every check, a lone pull request's as a group of one and a verdict carried after a trunk move included, is judged by a checking program built from the change's own heads in the build seat's sandbox, which also makes the heads' one seat list that the check hands the manifest validator and the composer; it brings no seat definition and no manifest field, so the running checker checks it (ADR 0644; *The bootstrap delivery, and the one milestone after it*) | 26 needs it | mesh-sdk (`go/delivery`: the seat `node-build-go`'s ask `run` and the checker's record of which checking program judged; `go/composer`: the request's seat list), mesh-catalog (mesh-build-go: the run step, the refused answer, the kept programs' tools and sweep), mesh-delivery (checker: the checking program and its command), hq (this design, the glossary) | 1921 |
| G | Two graphs and one revision view: the catalog graph and the assignment graph kept per revision, a change delivered as its graph difference, the acting parts' rules on their seats as bounded waits that hold their modules when the new version fails, the check on the change's own acting parts, the revision view and each process's own build, the gate's delivery record, and the trace's links (ADR 0656); no part reads a pull request's description (ADR 0659); every delivery part reaches the forge through the seat `git` and names no other module; nothing holds an approved merge for another check; and, one milestone with them, research 086's pipelines: a change of any module goes from a commit on a watched branch to the machines through its module's pipeline, its policy in its registration, ordered by edges between artifacts, with no Approve per merge and no hand act, the checker gone (ADR 0666, ADR 0667). Test counts at its start (G's prepare, 2026-11-09): mesh-delivery 665 unit and 68 integration, the node-engine (`mesh-host`) 1,742 unit; each other repository's and each catalog module's from the last check's verdict, read in the joint prepare | 26 to 62, 63 to 106 | mesh-sdk (the runtime: the fingerprint, the view, the definition reader, the seat list's one wiring), mesh-contracts and the domains' contracts modules (`acts`, the new verbs and events, the seat `mesh-grants`, the seat `git`'s verbs), mesh-delivery (indexer, assigner, checker until its removal, builder, grants part, gate's view, forge bridge, watchman, the new pipeline runner `mesh-delivery-pipeline`, `contracts` and `sdk` moved under `modules/`), mesh-catalog (the gate, the build modules, the forge module, mesh-issues, mesh-knowledge, the `node-backup` holder, the router, nats: a module's kept data on the bus as a backed-up item and the object store beside each state), mesh-sdk (also its small object store package, for values of any size), the node-engine (`mesh-host`), hq | 1966, 1961 (retargeted to this row's shape), 1660, 1972, 1721, 1783, 1785, 1794, 1791, 1373, 595; 1962 goes back with 1958, whose assignment entry it needs (ADR 0641 §3) |
| H | The build-and-delivery system delivers itself by its one model, in six releases (H0 to H5, the plan's section 6): one merge call of the seat `git`'s `merge-pull` naming one or more pull requests is a delivery's start and its group; the gate names the merged heads one source revision and opens its record; the indexer takes it in (a commit, the fingerprints of its repositories' modules, one catalog revision) and names the movers; one pipeline per take-in, none in parallel (`pipelines-at-once`, default 1); a want is the newest released version whose pipeline passed; the forge's event is the watchdog (`head-untaken`); every delivery part wakes on its inputs alone and reads no other part's state; the per-pull-request paths, the catch-up by comparison, the sweep and the checker go; one clean state placed by hand once, after which every release is delivered by its own pipeline (ADR 0682, ADR 0681, ADR 0666, ADR 0667; design 47 H §734 to §774) | 110 to 119 | mesh-delivery, mesh-catalog, `mesh-host`, mesh-tools, mesh-sdk, mesh-contracts, hq | opened by each release's prepare |
An issue is in one row only (ADR 0493 §2): where a row named an issue another row names too, the last row keeps it and the
earlier row says so in words (the operator, 2026-10-26). Milestone M1 makes each row a record of the issue tracker
@@ -84,9 +84,10 @@ check and deliver milestone G itself (ADR 0667 §2); each of them says so in a l
what moved); the verb `registrations` answers every registration, or one with its history. The indexer derives the
modules it keeps from them, and the forge bridge its watched branches, each by a first view of `registrations` and
then that event, never by reading the ledger (ADR 0666 §3, ADR 0596).
- §5 **An unprotected branch is refused.** `register` and `watch` read a branch's protection through the seat `git`'s read
verb `branch` (its head, and whether a rule refuses a direct push and admits only merges), and refuse a watched branch
the forge does not protect, naming it (ADR 0666 §3).
- §5 **An unprotected branch is not refused at registration** (corrected in place by design 47 H §763). `register`
and `watch` read no branch's protection and refuse no watched branch for it: the operator approved removing the
check on 2026-11-14 (G-23, *Met during G's start*), and ADR 0682 §13 ends ADR 0666 §3's refusal, `merge-pull`'s
protection judgement and §17's `branch-unprotected` (design 47 H §735, §739).
- §6 **The foundation parts are the manifest validator's list.** The manifest validator's list (ADR 0619 §3) holds
exactly the foundation parts, ADR 0619 §1's, the runner, which joins them (ADR 0666 §4), and the forge and image
registry modules (§584); the checker leaves the
@@ -317,8 +318,10 @@ check and deliver milestone G itself (ADR 0667 §2); each of them says so in a l
removed with the checker's buckets and bus account, after the first pipeline after G's delivery has delivered a change
end to end and on the operator's word (ADR 0667 §12; *How milestone G itself is checked and delivered*); the forge bridge hears the landed commit as it hears every commit of a watched
branch.
- §40 **The operator-channel hides no ask.** It keeps design 46's three open asks per asker, never holds the newest back,
and replaces an open ask by the same asker's ask with the same `about` (ADR 0666 §12; design 46 §6).
- §40 **The operator-channel hides no ask.** It keeps no count of open asks per asker and holds none back: every ask
is offered when it is raised (A-424, the operator, 2026-11-15 about 15:50 CET), and it replaces an open ask by the
same asker's ask with the same `about` (ADR 0682 §14, which ends ADR 0666 §12's three open asks per asker; design 46 §6;
corrected in place by design 47 H §763).
- §41 **How it is checked.** Whole (forge module, forge bridge): a session's `merge-pull` at its head merges with no run yet,
and with a passed one, and asks no Approve; with the runner stopped a head merges; a head at whose fingerprint any
run failed a `required` step, a head that moved and a branch whose protection was removed since its last move are
@@ -242,8 +242,9 @@ each the design's choice with its reason, §468 to §479 those of the runner's,
whole (forge module), a head that passed, then another commit at its fingerprint failed a `required` step, is
refused; its `seen` key goes when the pull request closes.
- §476 **`take-in.taken`'s body and its type.** (§493) The indexer's own `take-in.taken` carries, beside what it carries on
main, `branch` (the watched branch), `source` (the source revision taken in) and `entries`, one `{module, commit,
fingerprint}` per module of that revision. Its body is defined in the seat `mesh-catalog`'s definition, in its
main, `branch` (the watched branch), `source` (the source revision taken in) and `entries`, one
`{module, commit, fingerprint, repository}` per mover of that take-in (design 47 H §743; corrected in place by H §763, where this line
said one entry per module of the revision). Its body is defined in the seat `mesh-catalog`'s definition, in its
section `bodies` (§439, §454), so the generated contract carries the type the indexer writes and the runner reads,
and neither imports the other (§333). Why: a body said on a module's own stream and read by others is a `bodies`
entry (§439), never a client's hand-written type (develop question Q-RN4). How it is checked: unit (seat generator),
@@ -384,12 +384,15 @@ that round are written into the lines they name, here. Questions that an earlier
- §535 **A missing build seat and a seat with no holder, each read where it shows.** The runner reads that no build
seat serves an artifact's language from the catalog half of its revision view: no entry whose `acts`, the acting
rules of the seats it claims, hold a rule `build` with `on` naming that language (§323); the pipeline then ends
failed at once, naming the language and the missing build seat. It reads that a seat has no holder from the step
itself: an ask on that seat's work queue whose `step-started` is not heard within five minutes ends the pipeline
failed, naming the seat (§32). It reads neither from the seat list, which answers definitions alone (§527), nor
from any holding. Why: §32 already ends a pipeline whose step nothing starts, and the catalog half is the runner's
one view (develop question Q-RN7). How it is checked: whole (runner), a fixture language with no build seat ends at
once, naming it; a seat whose holder is stopped ends within five minutes, naming the seat.
failed at once, naming the language and the missing build seat. *Amended by A-466y and A-466z (2026-11-17):* it reads that
a seat has no holder only when the seat `mesh-module-assigner`'s view `seat-holders` names none for it (G-21 §717's
follower; never the seat list, which answers definitions alone, §527), and a seat with none ends the pipeline failed
at once, naming it; an ask waiting for an existing holder is timed by §86's `build-capacity-waits` and A-466s's
`run-untaken-bound` (G-24 §720, §725), never by five minutes from the ask. Why: §32 already ends a pipeline whose
step nothing starts, and the catalog half is the runner's one view (develop question Q-RN7); a queued step ended
failed for "no holder" while its holder is present would say what is not true (principle 3). How it is checked:
whole (runner), a fixture language with no build seat ends at once, naming it; a seat the seat-holders view names
no holder for ends at once, naming the seat; a step queued past five minutes with a holder present is not failed.
- §539 **The runner's tools are named `runner_`.** The runner's module, `mesh-delivery-pipeline`, names its tools with
the prefix `runner_`, so the design's `runner_catch_up`, `runner_waits`, `runner_release` and `runner_health`
stand, and its first-view tool is `runner_first_views`; `<module>_first_views` (§205) carries the prefix the
@@ -378,12 +378,21 @@ for several build maps is answered once, and the maps' names for it are given in
- §646 **The runner's bodies name `fingerprint`.** (§585) The runner's `step-ended`, pipeline and wait bodies name the
build-input fingerprint `fingerprint`, not `fp`, as §585 reads it. [mesh-delivery DQ-2.] How it is checked: unit
(runner's contract).
- §647 **The runner times an ask from `ask-taken`.** (§535, §602) §535's five minutes run from the ask until the build
seat holder's `ask-taken` (§602), which the runner consumes; `step-started` is the runner's own. [mesh-delivery
DQ-3.] How it is checked: whole (runner), an ask no holder takes ends its pipeline after the setting.
- §648 **A `take-in.taken` entry carries the pull request.** (§476, §542) An entry is `{module, commit, fingerprint,
repository, pull}`, the last two from a `merged` message and absent for a `pushed` one. [mesh-delivery DQ-5.] How it
is checked: unit (indexer's contract).
- §647 **The runner times an ask from `ask-taken`.** (§535, §602; A-466y) The runner consumes the build seat holder's
`ask-taken` (§602); `step-started` is the runner's own. *Amended by A-466y and A-466z (2026-11-17):* §535's five
minutes from the ask until `ask-taken` go. A seat for which the seat `mesh-module-assigner`'s view `seat-holders`
names no holder ends the pipeline failed at once, naming it, judged only once that view's first view is read, never
from the not-ready picture `FollowSeat` answers before it (principle 3). The runner's manifest gains the `consumes`
entry `seat:mesh-module-assigner.views.seat-holders` (on main it has none: its `invokes` names only
`seat:mesh-module-assigner.assignments`); an
ask waiting for an existing holder is timed by §86's `build-capacity-waits` and by A-466s's `run-untaken-bound`
(G-24 §720, §725), and from its `ask-taken` by its step's bound (A-466p). [mesh-delivery DQ-3.] How it is checked:
whole (runner), a seat with no holder ends the pipeline at once, naming it; an ask a present holder has not taken
past five minutes is not failed (test 109).
- §648 **A `take-in.taken` entry names its repository, never a pull request.** (§476) An entry is `{module, commit,
fingerprint, repository}`, the repository from the gate's `source-moved` (design 47 H §743); the field `pull` goes,
since nothing reads a pull request after the merge call (ADR 0682 §1; corrected in place by H §763). [mesh-delivery
DQ-5.] How it is checked: unit (indexer's contract).
- §649 **`ParseStepKey` answers the asker.** (§484) `buildstep.ParseStepKey(key)` answers `(asker, node, id, err)`,
`asker` empty for main's form. [mesh-delivery DQ-8.] How it is checked: unit, both forms.
- §650 **The `mesh-builds` client's name.** (§586) It is `contracts/go/clients/meshbuildsclient`, with `Reserved`.
@@ -400,9 +409,8 @@ for several build maps is answered once, and the maps' names for it are given in
- §654 **The readers of the forge module's old events and tools move to the seat `git`.** (§310, §311, §313, §431;
ADR 0659 §1) [mesh-catalog Q-F1.]
- **mesh-config** consumes `git`'s `pull-merged` and `pull-closed` in place of `gitea.pull.merged`,
`gitea.pull.closed` and `gitea.forge.pull-request.closed`, and invokes `seat:git.pull` in place of
`gitea.gitea_get_pull_request`; **the records module** consumes `git.pull-merged`.
- **mesh-config** consumes no pull request event and invokes no `seat:git.pull` (ADR 0681 §1, which replaces this
bullet's first form; corrected in place by design 47 H §763); **the records module** consumes `git.pull-merged`.
- **The order.** Main's forge module says its own `pull.merged`, not the seat's events, so each consumer of
`git.pull-merged` or `pull-closed` is ordered after the forge module's version that says them, by the edge of kind
6 its `consumes` gives (§558); its durable consumer is made from the stream's first message, so no merge said
@@ -3,7 +3,7 @@ layer: to-be
status: in-progress
design: 47
milestone: G
updated: 2026-11-11
updated: 2026-11-17
decisions:
- "ADR 0667"
- "ADR 0666"
@@ -149,7 +149,7 @@ One file of milestone G's part of [design 47](../47-delivery-from-commit-to-deli
| `pipeline-started`, `pipeline-ended` | a pipeline's start, and its end with every step's result, its test counts, the source revision and `widens` |
| `group-ended` | a group's one pipeline's end, on which alone its wants are written |
| `run-started` | a run on request's start, with its members at their commits |
| `step-started`, `step-ended` | each step's start and end: the step, its status in the policy, its outcome, its build-input fingerprint, its test counts, for a carried result what it was carried from, and for a rerun that it is one and the caller the runtime proved; a bypassed step too |
| `step-started`, `step-ended` | each step's start and end: the step, its status in the policy, its outcome, its build-input fingerprint, its test counts, for a carried result what it was carried from, and for a rerun that it is one and the caller the runtime proved, for a step that read a test run or a composition run another step asked its `shared` (G-24 §729); a bypassed step too |
| `wait-held`, `wait-ended` | a step's wait: on a person (approval, code review), on a build node, on the revision view |
- §74 **What a session hears after its merge.** Between its `merge-pull` and the delivery, a session waits with `mesh_wait`
@@ -195,14 +195,14 @@ four lines a module's tools answer; every read tool of a store also answers its
| Part (seat) | Read everything it holds | Do by hand what it does itself | Recover without help |
|---|---|---|---|
| the runner (`mesh-pipeline-runner`) | `show` (one pipeline whole, its steps, step log and outputs), `list` (the pipelines of a module, a branch or a commit), `stands` (what stands still, since when), `why-carried` (where a step's result was carried from), `kept` (the kept results by digest) | `start` (a delivering pipeline for a watched branch's head), `run` (the run on request), `rerun` (one step again) | `accept` (a failed `optional` step, with why), `cancel` (a pipeline, dropping its asks), `ask-again` (a declined `approval`) |
| the runner (`mesh-pipeline-runner`) | `show` (one pipeline whole, its steps, step log and outputs), `list` (the pipelines of a module, a branch or a commit), `stands` (what stands still, since when), `why-carried` (where a step's result was carried from), `kept` (the kept results by digest), `runner_test_runs` (each test run, G-24 §728) | `start` (a delivering pipeline for a watched branch's head), `run` (the run on request), `rerun` (one step again) | `accept` (a failed `optional` step, with why), `cancel` (a pipeline, dropping its asks), `ask-again` (a declined `approval`), `runner_test_runs_drop` |
| the indexer (`mesh-catalog`) | `registrations` (with each one's history); `indexer_edges` (an artifact's edges per watched branch), `indexer_order` (the order of a change, its cycles named), `indexer_catalog_difference` (two revisions); `indexer_pending` (a group with its missing members) | `register`, `policy`, `watch`, `unwatch` (each on the operator's Approve) | `indexer_seed_again` (the first registrations' ask raised again), `indexer_take_in_pending` (a group taken in as gathered), `indexer_drop_pending` (dropped) |
| the gate (`mesh-delivery`) | `source` with a branch (`gate_source`: every branch's source revision) | — | `gate_source_again` (say `source-moved` again for a branch) |
| the forge module (`git`) | `git_failed_steps` (the failed `required` steps it holds, by module and fingerprint, with the run that failed each) | — | `git_failed_steps_drop` (drop one, with why) |
| the grants part (`mesh-grants`) | `grants_callers` (the callers each verb admits, read from the seat definitions) | — | `grants_why_refused` (for an account and a verb, the grant it lacks and the definition's line that leaves it out) |
| the build seats | `steps` (the steps detected for an artifact) | `steps` with `again` (detect again) | `drop` (ADR 0664 §4) |
| the builder (`mesh-builds`) | `builder_reserved` (each module's reserved version, §369); the released versions are the catalog's `versions` | `builder_publish_again` (publish the missing artifacts under the reserved version) | — |
| the assigner (`mesh-wants`) | `assigner_why_not` (why a module is not on a machine yet: the pipeline it waits for, its hold, the placement, the node's last `running`) | — | `assigner_release` |
| the assigner (`mesh-wants`) | `assigner_why_not` (why a module is not on a machine yet: the pipeline it waits for, its hold, the placement, the node's last `running`), `assigner_compositions` (each composition run, G-24 §728) | — | `assigner_release`, `assigner_compositions_drop` |
| the watchman | `watchman_views` (how far behind each part's view is, from each part's `view-waits` and first-view tools) | — | each part's `<module>_first_view_again` |
| the operator-channel | `asks` (with each asker's asks held back) | — | `ask-again` (offer a held-back ask now) |
| every module that imports mesh-sdk's small object store package | `<module>_state_objects`, `<module>_sent_objects` | — | `<module>_state_objects_drop`, `<module>_sent_objects_drop` |
@@ -230,6 +230,8 @@ and an estimated 2 mesh-sdk publishes a day, each rebuilding about 120 artifacts
| the builder | `reserved`: one key per module and `MAJOR.MINOR` (§369) | old `state` section | until its `version-released`, a release of another `MAJOR.MINOR` of the module, or `builder_reserved_drop` | a few KiB; at the cap the builder reserves nothing and says why |
| the assigner | `written`: the last source revision and end count per module and watched branch (§352) | old `state` section | until the branch is unwatched | a few KiB |
| the indexer and the assigner | `inputs`: a `run` ask's input by write time and sha256 (§354) | old `state` section | until the ask's `step-ended`, at most a day | 64 MiB; an estimated 300 asks a day of under 100 KiB each is under 30 MiB; at the cap the asker refuses the ask loudly, naming the store |
| the assigner | `composition-runs`: one key per candidate and machine, its program, input, ask, readers and verdict (G-24 §719) | old `state` section, `max-mib` the setting `composition-runs-cap-mib` | the setting `composition-runs-kept` (default 24 hours) after its end; deleted by the assigner (§719) | default 16 MiB; an estimated 30 candidates a day on four machines, 120 keys of at most 8 KiB, is under 1 MiB a day |
| the runner | `test-runs`: one key per candidate, level and tested packages, its ask, readers, outcome and counts (G-24 §725) | old `state` section, `max-mib` the setting `test-runs-cap-mib` | the setting `test-runs-kept` (default 7 days) after its end, deleted by the runner in the step log's order with its output object in `test-runs-objects` (§330) | default 16 MiB; an estimated 1,000 test runs a day of about 1 KiB is about 7 MiB at 7 days |
| the runner and the assigner | `findings` (§355), read by their own read tools alone; the keeper hears the findings as events (§383) | old `state` section | an entry deleted when its finding clears | a few KiB |
| every observer (§387) | the stream `<module>-findings`: `finding-written`, `finding-cleared`, `findings-stated` (§383) | old `streams` section for a foundation part, the module's `streams` otherwise | 7 days, its oldest dropped when full; the next statement covers a lost one (§384) | 16 MiB; an estimate of a few hundred findings a day of about 1 KiB is under 2 MiB a week |
| the condition keeper | `findings-views`: one key per observer and node, its last stated set, pending keys and open asks (§388) | the form of its `conditions` | deleted when the module leaves the catalog or the node (§365) | under 1 MiB; about 30 observers on four machines |
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,209 @@
---
layer: to-be
status: in-progress
design: 47
milestone: G
updated: 2026-11-17
decisions:
- "ADR 0667"
---
# 47, milestone G — what is owed once G is delivered
The list of what waits for G's delivery: the operator's remarks deferred while G was brought up, the stand-ins the
lead's answers mark to go after G, the hand state of G's live start, and the faults met on the way. It is checked when
G is delivered: each item is then removed, taken into a design or a record through its playbook, or noted on the issue
that covers it (asking the operator before a new issue). An item is struck from this list in the same pull request that
deals with it, which is how the list is kept true.
## The operator's remarks (2026-11-14, deferred until G is live)
1. **Two graphs only:** the catalog graph and the assignment graph. Each node's modules publish the state they own and
the version they run on the bus; a gap between the assignments and what a node says is closed by the node-engine.
2. **What a node can run is checked only on assignment and un-assignment** (for the modules that depend on it), never
as an input to every computation. The glossary's *assignment graph* and ADR 0656 say the assigner computes from the
catalog graph, the placements and the nodes' capabilities: check the record and the code against this.
3. **Assignments enter as commands** (the seat `mesh-module-assigner`'s `propose`: place, take off, ...); the assigner
computes no placement, it only derives each placed module's version from the catalog graph into the wants. Check the
glossary's wording against this.
4. **A pin** ("use this provider for this provision") may not belong in the assignments at all: question it, and what
the provision wiring should be instead.
5. **Who says a node's running versions:** the node-engine (its haves) or each module itself. The operator's picture
is each module; check which the design means.
6. **One address for a seat's view.** A reader's manifest names the assigner's view by its seat, the reader's code asks
the holder module's tool by its module name, and the grants part translates one into the other (and gave only the
events, not the right to ask: the indexer's first view was refused at G's live start). One address, the same in
manifest, code and rights.
7. **A registration is part of the module graph, not a list beside it.** After G's start the module graph held 178
modules while the registrations held none, with a seed ask to copy them from the old registry. Check the design
against one graph whose entries are the registrations and what each registered repository says.
8. **One bookmark (durable consumer) per process, never shared** (A-397). A module's program and its tool runner read
events under one durable name; the runner took all 178 first registrations and the program heard none, silently.
Share a durable name only for shared work (a queue). Owed: amend design 19's "one durable consumer per module"
(written when a module was one process) and the lines of designs 25, 32 §3, 57, G-5 and G-8 that follow it; build
per-process names in the grants part, mesh-sdk's process hub and the tool runner, with a start refusal and a log
line; the test of two processes of one module both hearing one event on a real bus; a check that a durable has one
reader; and a design-review playbook question: "what if two processes of one module do this?"
9. **The same mistake elsewhere.** A survey at G's start found it live in three modules (the assigner, the indexer and
the grants part: the runner acks and drops what no tool view wants) and possible in every module whose tool runner
arrives later (builder, checker, forge bridge, pipeline runner, teller, view, watchman, distribution, recovery).
Already avoided: the bus and zfs keepers (A-340), the node-engine, work queues. Also: a seat verb served by both a
program and its runner in one queue group splits the answers between them; durables nothing reads keep pending
messages.
10. **`registration-changed` is said for a first registration too:** check the event's name against what it says.
11. **Silent waits met during G's start:** the indexer's first versions fill returns without a word while another fill
runs, so a hung fill held every later one silently for two hours; `grants_membership` and `grants_assert` time out
after 30 seconds with no log line; the tool runner says a dropped event once per key.
12. **The module graph's edges are dependencies, never verbs or bus subjects** (2026-11-16; principle 9): a manifest
says what its module depends on, a seat where the need is meant to be interchangeable and otherwise a module (a
build dependency on a package module, a program needing another module), and the graph's edges come from that
alone. The verb edges of ADR 0614 §2 and §3, and the cycle exceptions they needed (A-361, A-428, A-437), go; whether
`invokes`, `uses` and the `views` forms stay in a manifest at all is asked with the research on the mesh's
accounts, and "each fact is declared in one place".
13. **A full delivery cycle by hand** (the operator, 2026-11-17 about 09:00: "make sure that every part of the
build-and-delivery system is also doable by hand via mcp tools", "so you can fully take control of a full cycle";
asked when, "after G"; principle 4). A survey of 2026-11-17 found, along the cycle merge, the gate's source,
take-in, entries, pipeline, steps, release, wants, grants, put and running: about 52 acts a module does by itself
with no hand tool, about 110 bus states with no per-key clear and 26 with no read tool, all but one of about 25 capped
states with no fill against their cap, and 14 recoveries with no tool. One milestone: the SDK's state tools for
every declared state (read, fill against cap, clear a key, drop key-gone objects), the missing hand acts per module,
a summary of the pipeline runner's pipelines by outcome, `source-moved` and the runner's `ended` declared so they
can be waited on, and the catalog check that refuses a module whose state or act has no tool. Also a consumer
composed from the union of the want's and the trunk's manifests, as grants already are (a bundle's new `consumes`
is unbound until a want names its build).
14. **No pass waits on a call** (principle 1; from the node-engine's #233 (`mesh-host`), A-461b): the node-engine's composer asks the seat
mesh-catalog's `module` for a trunk build at the top of every pass and waits up to the call's bound, so a seat that
answers slowly stalls every pass. The ask goes beside the pass, which composes with what it holds and takes the
answer at the next pass.
## Met during G's start (2026-11-14)
- **The seed of registrations** left out modules whose manifest lives in the application's own repository on its
watched branch, so they were assigned but unheld and the assigner refused every proposal (ADR 0567 §2). The seed
should register every assigned module from where its manifest is.
- **The indexer's catch-up and take-in** say nothing in the indexer program's journal.
- **The provisioner** drops a published composition with neither plan nor refusal silently and then removes the
node's plans; no read tool shows a node's composition.
- **The node-engine** is refused the builder's `build.ended` by the bus every five minutes: the grants part does not
give it; to fix before builds matter.
- **The register verb refuses an unprotected watched branch:** the operator approved removing the check (from the
seat `mesh-catalog`'s definition and the indexer that serves `register`).
## Stand-ins the lead's answers mark to go after G (2026-11-16)
- **A-436 with A-430b:** a want carries parts beside its version for a node-engine that reads parts, and a module
wanted with no passed pipeline names the parts its `build.ended` made until its release is heard.
- **A-438:** for an acting wait, a running have whose node-engine states no health counts as healthy, and an online
node runs its node-engine; goes when every node-engine states health.
- **A-439:** `assigner_written` with `why` answering for `assigner_why`.
- **A-434:** every account may do everything, and every seat verb admits every caller, until the research on the
mesh's accounts replaces it (the operator, 2026-11-15: three accounts, mesh, agent and operator; approval only as a
per-tool mark for agent calls).
## Hand state of G's live start (2026-11-16)
- **Runtime drop-ins on the control-node** that pin a hand-made fingerprint (and the tool runner's environment) for
the grants part, the teller, the forge bridge and the tool runner. A delivery does not remove them: each goes before
its module restarts on the delivered build, or the module believes it runs the old one.
- **Hand-placed builds** (the indexer and its tools, the tool runner, the control-node's node-engine, the assigner) are
replaced only by their deliveries; check each is.
- **About 80 hand-built modules on the control-node have no catalog entry:** their first move is held for a backup
(the kept-form hold).
## Faults met on the way (2026-11-16; note each on the issue that covers it)
- `node-files.delete` cannot move a root-owned file to the trash; `node-service-manager` has no daemon-reload.
- The indexer does not stop on SIGTERM: it keeps draining build-log entries until systemd kills it.
- The backup ask's headline says "backup failed" for a kept-form hold.
- The assigner logs a line naming about 140 modules every one to three seconds, and did not raise the backup ask
again after a restart.
- `mesh_wait` refuses the operator channel's events: `operator-channel` is neither a seat nor a module of the catalog
half.
- Four tests of mesh-delivery fail on main, reading files that do not exist there: `sdk`'s
`TestThePackageArtifactNamesNoPathForTheWholeDirectory`, `sdk/buildstep`'s two fixture tests and
`meshdeliveryclient`'s `TestTheRemovedVerbsAreTheDefinitionsRemovedRows`.
- The node-engine's repository holds about 1721 unit tests against the budget of 500.
- The live indexer was seen refusing the assigner's manifest for claiming `mesh-placement`, which neither main nor
G's branch claims: confirm the verdict after G's take-in.
## Tests run against what runs (the operator, 2026-11-16 about 13:00 CET)
- **A test runs against the bus as the mesh composes it.** The pipeline runner's integration test made its own
shared event stream catching every seat event, so a step's end reached the runner there; on the live bus a step's
end is a kept event on a stream of its own, which the runner never bound, and no pipeline could pass (fixed in the
runner at G's start, 2026-11-16). Wanted: a principle ("a test runs against what the module runs against: the bus as
the mesh composes it from the manifest, never a layout written in the test"); the shared library gives an
integration test the streams, kept streams and consumers the composition makes for the module's manifest; every
integration test that creates its own streams moves onto it (count them first); the runner's test publishes a step's
end on the kept stream.
- **A handler of a kept event is heard or fails loudly.** Starting a program's events binds only the shared stream;
kept seat events need a second call, and a program that registers a handler for one and never makes it hears
nothing and says nothing. The shared library binds the kept streams of every registered handler itself, or refuses
to start.
## One way of asking work (2026-11-16)
- The builder, the pipeline runner, the indexer (its validation before a take-in) and the checker each ask the build
seats in a shape of their own; the validation rides the check queue behind pipeline steps (a 6-second validation
waited over 20 minutes). Parked by the operator for research right after G: one generic pipeline for work and gates.
- An ask given again after a fix keeps its old form (a step asked with no module directory came back without one).
- The builder asks a failing build again every two to three minutes, reserving a new version each time.
- The builder's queue tool lists only the oldest 50 waiting entries (left over from 2026-11-14), so it cannot show
what waits now.
- The pipeline runner serves no tools on the control-node and logs nothing per step; the build seat's step events say
no pipeline before a step ends; `mesh_wait` refuses the build seats' and the runner's events; the bus's stream reader
cuts a body at 4 KiB; a running test step logs nothing.
- `TestTheBuildModulesKeepNoCopyOfTheDeliverySDKsStepCode` fails on mesh-catalog's main (the empty
`modules/mesh-build-go/buildstep` directory); the seven `checks/` tests, nats-tools and distribution fail in a run over
the whole of mesh-catalog.
- The glossary's **pipeline** and **pipeline step** entries leave out the build, which the runner asks as the step
`build` of the seat `mesh-builds`: correct them.
- A pipeline the runner set aside before A-446 with steps already written (`main.940.mesh-delivery-sdk`) is started
again by neither of the runner's start paths.
- The node-engine's journal and the have's phases say a module is "held" and that the node "keeps running what it
runs" for one thing: a module left on the build it runs, its new build not applied. The operator, 2026-11-17, wants
those words gone: retire "held" and "kept" in that sense from the node-engine's lines and the glossary, with one
glossary word in their place.
- A composition that passes on every machine as "not judged: the composer was handed no seat list" is reported as
passed: a composition step that judged nothing must fail or say so (principle 3).
## From the composition runs of 2026-11-17
- **The node-engine answers `compose-candidate` with different bytes for one candidate.** Within 30 seconds the
assigner kept twelve inputs of exactly 1,491,623 bytes with different digests for the same revision. The bytes differ
in the node-engine's answer, not in the assigner's request; a guess, not checked: a slice whose order changes between
asks, perhaps `ComposedHow.Running`. Under design 47 G §718 one composition run per candidate and machine asks each
node once, so this no longer multiplies anything and stays out of G (principle 7; the coordinator, 2026-11-17). After
G: find it by comparing two kept inputs of one node, read by digest through `mesh-wants`' `input`, and decide whether
the answer must be the same bytes while the node is unchanged.
- **Design 47's main file and G-22 are split** (playbook 02, *The shape of a design*). The main file is about 250 KB
against its 80 KB bound, and G-22 about 165 KB against a milestone file's 60 KB. The operator decided at 10:31 CET on
2026-11-17 that the split comes after G, as a separate amendment.
## Faults met on the way (2026-11-17; note each on the issue that covers it)
- **The assigner's long line, met a second time.** On 2026-11-17 from 11:02 CET the assigner on the stable node logs,
several times a minute, that "the manifest of the running build of" about 100 modules "cannot be read, so what the
module graph says they serve is counted as served". Cause not yet known: nearly every module is named, so the
assigner reads the running builds' manifests where none is kept. Being the second meeting, its fix is folded into
G-24's assigner change once A1 to A7 are built. Cause found the same day: the assigner reads a running build's
manifest only from the builder's end in its stream, which the builder has said only since 2026-11-11, so no build
made before then has one. G-24 says the list once per change of the set; after G, the builder says the end of each
build it keeps into its stream once (principle 1; the alternative, reading the catalog entry at that fingerprint,
reads another module's data).
- **The Node build modules disagree on their toolchain.** On 2026-11-17 shanks's recipe differed from the other three
nodes', and g14 ran Node v26.13.1 against v26.13.0, so no version and recipe held a majority, and lidarr, radarr and
gitea had no fingerprint at revision 449 (A-470j keeps that from leaving whole machines not judged). After G: one
version and one recipe declared through the mesh on every build node.
- **The catalog had not taken in the node-engine's merge #234** by revision 450 (2026-11-17 about 11:50 CET), and the
indexer's module reads answered "no responders": read why after G-24's delivery.
- **The builder's and teller's integration tests fail only the 30 s setup bound under load**, and pass idle: the bound's
kind (a setup bound judged on a busy build node) is decided after G.
- The runner's journal floods with ends that name `pipeline ""`: the build seats' run asks are heard on the runner's
subjects.
- `TestAProposalIsAnsweredApprovedAndWritten` (assigner) fails on main; `TestTheNodeComposesItsWantsFromTheBus`
(node-engine) fails now and then.
- `mesh-enrolment` is placed on no node, so its composition step can never run and it never releases.
- `claude_code_wait`'s `said` refused a wait past its bound; `mesh_wait` cannot wait on the gate's `source-moved` nor on
the runner's events.
@@ -0,0 +1,290 @@
---
layer: to-be
status: in-progress
design: 47
milestone: G
updated: 2026-11-17
decisions:
- "ADR 0667"
- "ADR 0666"
- "ADR 0596"
- "ADR 0564"
- "ADR 0267"
---
# 47, milestone G: one composition run per candidate and machine, one test run per level and tested packages
One file of milestone G's part of [design 47](../47-delivery-from-commit-to-delivered.md); the part opens at
[G.md](G.md). Its design lines are numbered across every file of the part and cited as "design 47 G §n".
*§718 to §733 come from milestone G's live delivery of 2026-11-17. The 24 pipelines of one catalog revision asked 96
composition runs, one per pipeline and machine, and 80 more when they were asked again. Each run kept its own copy of
a node's input of 1.1 to 1.5 MB in the assigner's state `inputs`, which reached its 64 MiB cap twice. The indexer's
and the teller's same failing tests failed once per artifact. The operator, 2026-11-17 about 10:20 CET: "In G, now".
Each line is a choice within ADR 0666 §7, §8 and §17, settled by the principle it names. The review by architectural
domain of 2026-11-17 was answered by the lead in A-466 to A-466l (G-22), each cited at the line it settles.*
### Composition runs
- §718 **One composition run per candidate and machine.** (§354, A-226, A-243, A-252, A-466l; ADR 0666 §17; principle
7) The assigner, holding the seat `mesh-wants`, composes each machine once per candidate, whatever number of
composition steps name it. The candidate is named by its key (A-226): `<branch>.<revision>` for a branch pipeline,
`run-<id>` for a run on request. The first composition step of a candidate that needs a machine starts that machine's
composition run: one `compose-candidate` ask of the node, its request kept once in `inputs`, one `run` ask on
`node-build-go`. Every other composition step of that candidate whose module is placed on that machine (A-252) reads
that run's verdict, under §720's rules. The request carries the candidate's manifests, verdicts and seat
definitions, depth compose, the one machine and the case `change`, and the node's own `compose-candidate` answer at
the moment the run started; it carries nothing of the module whose pipeline asked it. A step of the same candidate
that comes later so reuses a verdict composed from the node's answer of that earlier moment, knowingly: within one
candidate, that answer is the machine as the candidate is judged on it (A-466l). The composer program is chosen once,
when the run starts, by A-252's and A-451's rule, and recorded in the run. This replaces "one run ask per machine" for
each composition step (A-243, A-252) with one per candidate and machine.
- §719 **Kept in the assigner's state `composition-runs`.** (§354, §85; ADR 0666 §4, §7; A-466e; principles 1, 2, 7 and
11) Each composition run is one key of the assigner's old `state` section `composition-runs`, whose `max-mib` is the
assigner's setting `composition-runs-cap-mib`, default 16. The key is `<candidate key>.<machine>`, each part escaped as
a key's name (`keys.EscapeName`). The step that starts the run writes the key create-only, so two steps of one
candidate starting at once start one run and the second reads it. The value holds:
- `candidate` and `machine`;
- `program`: repository, commit, directory, and the node-engine fingerprint it was chosen from;
- `input`: the digest of the kept request;
- `ask`: the id of the build seat's run ask;
- `started`;
- `readers`: the step keys of the composition steps reading the run, each added and removed by compare-and-set;
- at the run's end, `ended`, `outcome` (`passed`, `failed` or `could not run`, with its words) and `verdict` (that
machine's composer verdict, §721).
A step waiting on a running composition run hears its end through the assigner's own watch of that key (a module
watches its own state, principle 1). At its start the assigner takes up again each key with no `ended`, by its
`ask`'s `step-ended` (A-240) and the bound of §720. The assigner deletes a key once the time its setting
`composition-runs-kept`, default 24 hours, names has passed since its `ended`; both this and the bound are judged as
the state `inputs` is (A-465b). At the cap no run is started, and its steps end `could not run`, naming the fill (as
A-465a).
- §720 **What a run's result serves, and its bound.** (§75, §86, §718; ADR 0666 §8, §17; ADR 0596; ADR 0564 §2;
A-466, A-466b, A-466n, A-466p, A-466q, A-466s; principles 1 and 3)
- **Within its candidate only.** An ended run's verdict is read only by composition steps of the same candidate. A
step of another candidate starts a run of its own; nothing composed for one candidate crosses to another.
- **`could not run` is never read.** A step that finds an ended run whose outcome is `could not run` starts a new
run, replacing the key by compare-and-set; its former readers have already ended `could not run` with it.
- **The bound times owed progress, not run length** (§75; ADR 0564 §2; A-466n, A-466p). A run has one clock: a bound
key `bound.<run key>` in `composition-runs`, with its own time to live (A-77, A-101). The assigner writes it at
the build seat's `ask-taken` of the run's ask, and writes it again at each `step-progress` of that ask, its time
to live the assigner's setting `composition-step-bound`, default 30 minutes. An ask still waiting in the queue has
no clock: it is timed by §86's build-capacity finding, since a `could not run` for a run that only waited for a
build node would say what is not true. A run that runs past 30 minutes and keeps saying its progress is never
ended by it.
- **When the clock ends.** Its expiry, heard as the bus's event, ends the run: the assigner drops its ask (§119,
A-182), writes `ended` with `could not run` naming the ask and the bound, and ends every reader `could not run`
with it. A `step-ended` of that ask heard later is ignored, said in its journal with the ask and the run's key.
- **A run never taken** (A-466s). A run whose ask has no `ask-taken` within the assigner's setting
`run-untaken-bound` after its `started`, default 24 hours, ends `could not run` the same way, naming the bound;
§86's `build-capacity-waits` says it meanwhile.
- **What the assigner hears** (A-240, A-466q). Its manifest gains `consumes` entries for the build seats'
`ask-taken` and `step-progress`, heard on its durable consumer as A-240 hears `step-ended`. The assigner never
reads the build seat's state (ADR 0596).
- §721 **A machine's verdict names the modules it blames.** (ADR 0666 §17; A-452, A-466g, A-466h, A-466r, A-466u; principles 10, 3, 7
and 11) The composer's verdict of a machine (mesh-delivery-sdk's `composer.Machine`, answered by the node-engine's
`mesh-compose`) gains `blamed`, a list of `{module, problem}`.
- **What it names.** Each module that the candidate's composition of the machine holds or refuses and the
composition without the candidate does not (the modules in its input's `HeldBefore` are left out). Two modules
owning one path are each blamed, with that path. `composes` stays false when the machine does not compose with
every module of the candidate. A machine that does not compose and blames no module fails on the node's own
account (its own composition, or no module composed at all, as A-452 has it).
- **How culprits are found, and its bound.** Only the candidate's changed modules placed on the machine (those whose
entry in the candidate differs from the composition without it) are held, one at a time, each try composing once
more; the search ends at the first module whose hold ends the problem, as the indexer's validator names a manifest
whose removal ends a problem. It composes at most once per such module plus once, never more than `blame-tries`, the
assigner's setting, default 20, carried in the run's request to the composer (the request's field
`blame-tries`), since the composer program has no settings of its own (A-466r); past that bound, the modules not yet tried are named in `blamed` with the words
"not tried: the search reached its bound".
- **An old composer.** A verdict with no `blamed` field (a node-engine's composer from before this line) is judged as
today: a machine that does not compose fails every composition step placing a module on it, with the warning "the
composer names no culprits".
- **No order between them.** mesh-delivery-sdk's `composer` package is published before the node-engine and the
assigner build (an edge of kind 1, as for every package); nothing orders the node-engine's composer before the
assigner, since an assigner meeting an old composer judges as today (A-466u).
- §722 **One module's fault fails only its own pipelines.** (§721; ADR 0666 §9, §17; A-452; principles 10 and 3) A
composition step judges each of its machines from that machine's verdict:
- `failed` when the verdict blames the step's module, or when the machine does not compose and blames no module;
- `could not run` when the composition run could not run;
- `passed` otherwise, with a warning for each module blamed there that is not the step's own (`<machine> holds
<module>: <problem>`). Another module's fault is so said in every pipeline it touches and fails none of them.
The step's outcome then follows from its machines, as today: failed before could not run before passed (§354). A
machine that is away passes with a warning and is not composed (§64).
- §723 **How a composition run ends and is said.** (§73, §116, §119, §496, A-182; ADR 0666 §7, §21; A-466j;
principles 1, 7 and 8) A composition run says no event of its own. Its end is the build seat's `step-ended` of its
`ask`, which the assigner alone hears (A-240). Each composition step reading the run says its own `step-ended` on
`mesh-wants`, as today, with its own build-input fingerprint and `shared` (§729), naming per machine the run it read,
that run's `ask` and its verdict: the result was taken from the build seat's holder once, at that ask, and the step
names it, as §25's carry names where a carried result came from (ADR 0666 §21). Each reading step says its
`step-progress` once a minute while it waits (A-262), naming the runs' asks in `shared`. A step that is superseded or
cancelled leaves `readers`. The run's `ask` is dropped (§119, A-182) only when no reader is left, so a superseded
pipeline never takes away another pipeline's composition run.
### Test runs
- §724 **One test run per level and tested packages.** (§22, §25, §27, §416; ADR 0666 §8, §17; ADR 0267 §1; A-466,
A-466d; principles 3 and 7) The runner asks a lint, unit test or integration test step of a build seat once per
candidate, level and tested packages, whatever number of artifacts' steps name them. A test step's candidate is its
pipeline's in A-226's form: `<branch>.<revision>`, or `run-<id>` for a run on request. A module's program and its
tools bundle are built from different import closures (ADR 0267 §1), but their test steps run the packages of one
directory (the build seat's `steps`, §22), so one set of tests is run. The share digest is the sha256 of:
- the step's build-input fingerprint with the artifact's own part left out (its program's package and recipe), so two
artifacts of one directory meet;
- the step's status in the policy (§23), so the key is §25's carry key: two artifacts whose policies differ do not
share;
- the sorted lines `<package>@<digest>`, one for each package the step would run after its carry (§25: those with no
kept result for their digest), the digests being those the runner already reads for its carry (§416).
The first step of a key puts the build seat's `step` ask; every other step of the same candidate with that key reads
its result. An ended run's result is read only within its candidate; a step of another candidate asks a new run, and
a passed result crosses candidates only as §25's carry, through `kept`. A `could not run` result is never read: a
step finding one asks a new run, replacing the key by compare-and-set. So within a candidate the same failing tests
fail once, not once per artifact, and a second revision with the same digests runs a failed test again.
- §725 **Kept in the runner's state `test-runs`.** (§27, §85, §330; ADR 0666 §4; A-466e, A-466f, A-466p, A-466v; principles 1, 2, 3, 7
and 11) Each test run is one key of the runner's old `state` section `test-runs`, whose `max-mib` is the runner's setting
`test-runs-cap-mib`, default 16. The key is `<candidate key>.<level>.<digest>`, each part escaped as a key's name,
written create-only by the step that asks the run. The value holds:
- `level` and `digest`; the list of packages with their digests is in the run's output object, never in the key or
the value;
- `seat` and `ask`, and `node` once a holder has taken the ask;
- `started`;
- `readers`: each reading step's pipeline key and artifact, added and removed by compare-and-set;
- at the run's end, `ended`, `outcome`, its test counts, and the name of its output object.
A test run's output object is kept in the state's own object store `test-runs-objects`, an old `objects` section
with no age of its own, so it lives exactly as long as its key, whatever `step-outputs`' 7 days (A-466v); a step's
own output stays in `step-outputs` as today. A reader whose run's object is gone (a drop, a fault) answers its run
without the packages, saying "the run's output is gone".
The runner hears the run's end as the build seat's `step-ended` of its `ask`, on its durable consumer, as it hears
every step (§24). A test run has §720's one clock, a bound key `bound.<run key>` in `test-runs`, written at the
build seat's `ask-taken` of its ask and again at each `step-progress` of it, its time to live the runner's setting
`test-step-bound`, default 30 minutes; an ask still in the queue is timed by §86's build-capacity finding alone. Its
expiry ends the run as §720's does: the ask dropped (§119), every reader ended `could not run`, a later
`step-ended` of the ask ignored and said in the journal. A run whose ask is never taken ends `could not run` at the
runner's setting `run-untaken-bound`, default 24 hours after its `started` (A-466n, A-466p, A-466s). The runner
consumes `ask-taken` already (§647); its manifest gains `step-progress` where it does not consume it. At its start the runner takes up again each key with no `ended`, by its
`ask` and that bound. It deletes a key once the time its setting `test-runs-kept`, default 7 days, names has passed
since its `ended`, in the step log's order as §330 ages its `pipelines`, and deletes the run's output object from
`test-runs-objects` with the key, after the last reader's end. At the cap no run is asked, and the step ends `could not run`, naming the fill, as
§719 has it, with no condition of its own. A passed run's results go into `kept` per package, as today (§27), so a
step asked after the key is deleted carries them.
- §726 **How a test run ends and is said.** (§24, §73, §119, §419, §496; ADR 0666 §8, §21; A-466i, A-466j; principles
3, 7 and 1)
- **Each reading step is said.** Each step reading a test run says its own `step-started` and `step-ended`, with its
own build-input fingerprint and `shared` (§729). Its `step-ended` carries the run's outcome and counts. Its
`step-started` names the run's `ask`, so the watchman times it from that ask's `step-progress` (§496). The result
was taken from the build seat's holder once, at that ask, which the step names, as §25's carry does (ADR 0666 §21).
- **Counted once.** `pipeline-ended` counts a module's tests once per pipeline and test run, never once per step
reading it (§24, §419).
- **Dropped only with its last reader.** A step that is superseded or cancelled leaves `readers`, and the run's ask
is dropped (§119) only when no reader is left.
### For both kinds
- §727 **Asking a run again moves every reader.** (§30, §73, §83 `rerun`, §29; ADR 0666 §9, §11, §24; A-466a;
principle 3) One rule for composition runs and test runs. A `rerun` of a composition step (for each machine whose
verdict in it was not `passed`) or of a test step joins the run of its key while that run is running, reads one that
started after the step's last end, or otherwise starts a new run, replacing the ended key by compare-and-set and
carrying the former run's `readers` into the new value. The new run's end moves every step that read the former run
and did not pass: each says `step-ended` again with the new outcome, marked as a rerun with the caller of the step
asked again (§73). A moved step reopens its pipeline, which says `pipeline-ended` again with a higher end count; the
assigner writes wants only from a pipeline's latest end (§29). A step a person `accepted` (§30) does not move: a
person's act stands. A machine that passed in a composition step keeps its verdict. A step left reading a failed
result would say what is no longer true. So twenty composition steps of one revision asked again together start one
run per machine.
- §728 **The tools.** (§83, §391; ADR 0666 §25; A-466c; principle 4) Two for each new state, in §83's table:
- `assigner_compositions` (class `read`) takes a candidate, a machine, or neither, and answers each composition run
with its program, input digest, ask, readers, outcome and verdict, and the state's fill against its cap.
- `assigner_compositions_drop` (class `act`, §391's default callers) takes `{candidate, machine, why}`. It writes the
hand-act log first, drops the run's ask, ends each reader `could not run` naming the drop, and deletes the key.
- `runner_test_runs` (class `read`) takes a candidate, a level, a digest, a module, or none of them, and answers each
test run with its packages (read from its output object), ask, node, readers, outcome and counts, and the state's
fill against its cap.
- `runner_test_runs_drop` (class `act`, §391's default callers) takes `{key, why}` and does the same as
`assigner_compositions_drop`, deleting the run's output object with the key.
- §729 **`shared` in the seat contracts.** (§73, §116, §496; ADR 0666 §21, §24; A-466j; principles 1 and 8) The seat
definitions in mesh-delivery's `mesh-delivery-contracts` change, and the contracts' code is generated from them by
`seatgen`, never edited by hand:
- **`mesh-wants`' `step-ended`** gains `shared`: a list of `{machine, run, ask}`, one per machine composed, `run` the
`composition-runs` key and `ask` its run ask's id, all strings.
- **The composer's request** (mesh-delivery-sdk's `composer.Request`) gains `blame-tries` (a number, §721).
- **`mesh-wants`' `step-progress`** gains the same `shared` list, naming the runs the step waits on.
- **`mesh-pipeline-runner`'s `step-started`** gains `shared`: `{run, ask}` for a test step, `run` the `test-runs` key
and `ask` its build seat ask's id, strings.
- **`mesh-pipeline-runner`'s `step-ended`** gains `shared`: `{run, ask, asked-by}`, `asked-by` being `{pipeline,
artifact}` of the step that asked the run (strings), left out for that step itself.
- **`mesh-pipeline-runner`'s `stands`** answer: each entry gains `ask` (string) and `shared` (`{run, ask}`, strings,
where the runner knows it), as §731 says.
A step that names `shared` names its own build-input fingerprint beside it, as every `step-ended` does (§73). A
composition step's `step-started` is said by the runner when it asks the assigner, before any run is chosen, so it
cannot name the runs' asks; the assigner, the part that knows them, names them in its `step-progress`, and the
watchman times a composition step from that (§496, A-262, A-466m).
- §730 **What runs in flight, and going back.** (§65, §78; ADR 0666 §19, §24; ADR 0667 §9; A-466k; principle 3) A
composition step or test step asked before the new assigner or the new runner runs ends by the path it was asked
under, one run per step; composition runs and test runs are made only for steps asked after the new part runs. A
return to an earlier build (§78) leaves `composition-runs` and `test-runs` unread; their readers are ended `could not
run` by the bound of §720 and §725 once the new build runs again. Until then the earlier build applies no such bound,
and a reader standing still is said by the watchman as `pipeline-stands`, by §732 (A-466o).
- §731 **The runner's `stands` names what a reading step waits on.** (§75, §83; A-466t; principles 3 and 4) `stands`
already answers every pipeline step started and not ended whose outcome is running or waiting (mesh-delivery's
runner on main, its `Stands` read), so a composition step waiting on the assigner is in it. Each of its entries
gains `ask`, the step's own ask (§496), and `shared` where the runner knows it (§729, a test step's run key and
ask). At its start the watchman seeds §732's open steps from it, each clock begun at that moment, and times each by
those asks from then on.
- §732 **The watchman times every step that reads a run.** (§75, §496, §602; ADR 0564 §2, §4; ADR 0666 §24; A-466o;
principles 3 and 1) §75 and §496 time each pipeline step from its `step-started` or its latest `step-progress`, but
the watchman on main times only a build seat's running ask (its `ask-taken` and `step-progress`) and a step queued
for a build node; a step whose run has ended, or a composition step waiting on the assigner, is timed by nothing.
From G the watchman also:
- **hears** the runner's `step-started` and `step-ended` of every pipeline step (it consumes them already, for the
build capacity), and `mesh-wants`' `step-progress`, a `consumes` entry its manifest gains;
- **keeps** each started step with no `step-ended` in memory under `<pipeline>/<step>`, its clock begun at its
`step-started` and begun again at each `step-progress` whose `ask` is the step's own ask (§496) or an ask its
`shared` names (§729), and at the `step-ended` of the run's ask a test step names;
- **raises** `pipeline-stands` for that step when its clock passes the watchman's existing five-minute progress bound
(§75), naming the pipeline, the step and the ask it last heard of, and clears it at the step's `step-ended`. A
step said waiting on a person, or queued for a build node (§86), is owed no progress, as today;
- **seeds** its open steps at its start from the runner's `stands`, as §75 has it.
So a reader that neither the assigner nor the runner ends, after a return to an earlier build or any fault, is said
within five minutes of its last sign of life.
- §733 **How it is checked.** Tests 108 and 109.
- **Whole (assigner, mesh-delivery), test 108.** Two composition steps of one candidate started at once make one key,
one request and one run ask. Three composition steps of one revision on two machines start two runs, and each step
reads both verdicts. A machine whose verdict blames module A fails A's composition step and passes B's, with a
warning naming A; a machine that does not compose and blames no module fails both; a verdict with no `blamed` is
judged as today and says "the composer names no culprits". A step of another candidate starts its own run, and a
`could not run` result starts a new one. A taken ask with no `step-ended` and no `step-progress` for 30 minutes
ends `could not run`, with its readers and its ask dropped, and a `step-ended` of that ask heard later is ignored
and said in the journal; a run past 30 minutes that keeps saying its progress is not ended; an ask never taken
ends `could not run` at `run-untaken-bound`. Steps asked again together start one run per machine that did not
pass, and move every former reader; a reopened pipeline says `pipeline-ended` again. A superseded step leaves the run running for its other readers. An assigner restarted during a run takes it
up again by its ask. At the cap a step ends `could not run`, naming the fill. `assigner_compositions_drop` ends the
readers and deletes the key, and a reader-only account calling it is refused.
- **Whole (the node-engine's composer, `mesh-host`).** A fixture module whose manifest is refused on the machine is
blamed by name, two modules owning one path are both blamed, and the search past the request's `blame-tries` names
the untried.
- **Whole (runner and Go build module, mesh-delivery and mesh-catalog), test 109.** A module whose program and bundle
test one directory asks one unit step and one integration step, and both artifacts' steps end with that result and
`shared`. A failing test fails once, and `pipeline-ended` counts the module's tests once. A rerun of one artifact's
failed step that passes moves the other artifact's step to passed too, and the pipeline says `pipeline-ended` again;
an `accepted` step does not move. A runner restarted during a run takes it up again by its ask. A replaced reader
leaves the ask to the other readers. At the cap a step ends `could not run`, naming the fill. A run past 30
minutes that keeps saying its progress is not ended, and a taken ask silent for 30 minutes ends `could not run`
with its readers. A test run queued past five minutes with a holder present is not failed (A-466y). A second
revision with the same digests runs a failed test again. `runner_test_runs_drop` deletes the key and its output object, and a
reader-only account calling it is refused.
- **Unit (runner).** A test run whose output object is gone is answered by `runner_test_runs` and its readers
without its packages, saying "the run's output is gone"; the object outlives `step-outputs`' 7 days while its
key lives.
- **Whole (watchman), tests 108 and 109.** A composition step whose `mesh-wants` `step-progress` stops raises
`pipeline-stands` within five minutes; a test step whose run's ask ended with no `step-ended` of the step raises it
five minutes later; each clears at the step's `step-ended`; a step waiting on a person raises nothing; a watchman
restarted while a composition step waits on the assigner seeds it from `stands` and raises it if no progress
follows.
- **Live.** In the first publish after G's delivery, the composition run asks of each revision number at most its
machines, read with `assigner_compositions` and `assigner_inputs`; each module's test steps run once per level, read
with `runner_test_runs`.
@@ -103,7 +103,9 @@ record says so in its Why. Round 3 of the review by architectural domain (2026-1
`*.findings-stated` are the first. The manifest validator takes `*` only as the whole emitter, and only before an
event that a seat's definition or a contract in mesh-sdk defines, with its fields (for the finding events, the seat `mesh-conditions`' definition, §439); it refuses `*.>` and `*` in part of an emitter's name, and a `*` entry before any other event keeps its meaning on main (§474). The grants part,
from the catalog graph as it derives every grant (§59, §442), makes on each declaring module's stream one durable consumer
for the consumer, named `<machine>_<module>` as ADR 0363 §5 names one, with its grants to read and acknowledge it and
for each process of the consumer that hears it, named as A-397 names a process's durable (`<machine>_<module>` for
its program, `<machine>_<module>_tools` for its tool runner, `<machine>_<module>_<process>` for another; corrected in
place by design 47 H §763, where one per module stood), with its grants to read and acknowledge it and
its membership's word to bind it; it drops that consumer when the module no longer declares the event, and makes one
for a module that declares it later, with no change to the consumer's manifest. A `*` entry never reaches a
restricted stream (one with a `grant`, ADR 0363 §4). Why: the keeper hears modules it finds in the catalog and its
+4 -2
View File
@@ -3,7 +3,7 @@ layer: to-be
status: in-progress
design: 47
milestone: G
updated: 2026-11-12
updated: 2026-11-17
decisions:
- "ADR 0500"
- "ADR 0673"
@@ -88,7 +88,7 @@ file on 2026-11-10, and is cited as "design 47 G §n". A sentence that holds a d
stayed in the main file, and G builds what it says from G's delivery. A section named in italics is in this part, or,
when it is not here, in the main file.
The part is twenty-two files, each read whole, its design lines numbered across them:
The part is twenty-four files, each read whole, its design lines numbered across them:
- [G-1-pipelines-registration-and-steps](G-1-pipelines-registration-and-steps.md): pipelines per module: registration, watched branches, steps and policy, what holds a delivery, approval, groups, dependencies, a publish.
- [G-2-pipelines-sizes-checks-and-tools](G-2-pipelines-sizes-checks-and-tools.md): pipelines per module: any size, the pipeline makes the check, who may say what, health, recovery, tools, kept data, how G itself is delivered, the choices under ADR 0666 and ADR 0667.
@@ -112,5 +112,7 @@ The part is twenty-two files, each read whole, its design lines numbered across
- [G-20-lead-answers](G-20-lead-answers.md): the lead's one-line answers to develop's questions, each tested against the principles (A-1 on).
- [G-21-provider-and-revert-ask](G-21-provider-and-revert-ask.md): the provider a requirement points to, the named seat and the contribution edge (§708, §709, completing §357), and the assigner's ask of a held revert (§710, completing §300), the edge an acting rule gives (§711, completing §47's kind 5), the validation before take-in (§712, §713, completing §16), the take-in and groups of every watched branch (§714, §715), where the gate keeps a group's members (§716) and the seats of the status view's and the assigner's views, followed before their holder is known (§717).
- [G-22-lead-answers-2](G-22-lead-answers-2.md): the lead's answers continued, from A-182.
- [G-23-after-delivery](G-23-after-delivery.md): what is owed once G is delivered.
- [G-24-composition-runs-and-test-runs](G-24-composition-runs-and-test-runs.md): one composition run per candidate and machine, one test run per level and tested packages, from G's live delivery of 2026-11-17 (§718 to §733).
Design 34's milestone [every-event](../34-milestones/every-event.md) and design 46's milestone [7-approval-mark](../46-milestones/7-approval-mark.md) ship in milestone G's release, built in its repositories beside G's lines (lead-47, 2026-11-11 15:30 CET); where one of their lines and a line of G differ, G's later line holds and the other is corrected in place (§624 to §626).
@@ -0,0 +1,297 @@
---
layer: to-be
status: in-progress
design: 47
milestone: H
updated: 2026-11-20
decisions:
- "ADR 0682"
- "ADR 0681"
- "ADR 0667"
- "ADR 0666"
- "ADR 0656"
- "ADR 0633"
- "ADR 0596"
- "ADR 0573"
---
# 47, milestone H: the merge call, the watchdog, the take-in and one pipeline at a time
One file of milestone H's part of [design 47](../47-delivery-from-commit-to-delivered.md); the plan that opens the part
is [H-deliver-the-delivery-system](H-deliver-the-delivery-system.md). Its design lines continue the numbering of
milestone G's part after §733 and are cited as "design 47 H §n". Each line says which release of the plan's section 6
builds it (H1 to H5), how it is checked, and which line of G or lead's answer it replaces; where the code it lands in
is named, it is named by module and function, from the gap tables of 2026-11-20, for the builder to find it.
*The operator, 2026-11-19 about 23:30 CET: "Nobody should 'take in a merge', only commits to the configured branch of
registered modules." 2026-11-20 about 01:10 CET: "parallel pipelines sounds like a very bad idea". 01:15 to 01:20 CET:
a delivery is started by hand by the merge it already needs, the lead named principle 3 (a start a person forgets is
a silent non-delivery), and the operator chose "A": the merge through the mesh is the start and the group, and the
forge's event is the watchdog. ADR 0682 records it (cited by the clause numbers of its version 8, fourteen clauses). The chain these lines build, in one sentence: a **merge call**
(the seat `git`'s `merge-pull` naming one or more pull requests) merges the heads, the gate names them one source
revision per watched branch and opens its delivery record, the indexer takes that revision in and names the movers,
the runner runs one pipeline for it when no other runs, and the assigner wants the newest passed release; the forge
bridge's one message per head move is heard only by the gate's watchdog, which raises an **untaken head** for a head
no merge call delivered.*
```mermaid
flowchart LR
S[session or operator] -->|git.merge-pull pulls| F[forge module, seat git]
F -->|merge-called| G[gate, seat mesh-delivery]
G -->|source-moved| I[indexer, seat mesh-catalog]
I -->|take-in.taken: movers| R[runner, seat mesh-pipeline-runner]
R -->|pipeline-ended| A[assigner, seat mesh-wants]
F -->|branch-moved| B[forge bridge]
B -->|head-moved| G
G -.->|head-untaken| K[condition keeper]
```
### The merge call
- §734 **One merge call is a delivery's start and its group.** (ADR 0682 §1, §2; ADR 0666 §11; the operator, 2026-11-20
01:20 CET; principle 3) The seat `git`'s verb `merge-pull` takes `pulls`, a list of one or more members
`{repository, number, head}` (the head the caller names, as ADR 0666 §11 has it), and `by`, the caller as who@node;
the forge module records `by` beside the bus account the tool runner proves for the call. The members named in one
call are one delivery and, with two or more, its group; nothing else makes a group. The verb's single-pull form goes:
a call naming one pull request is a list of one, and its callers change with it in H1 (the tool runner's
`merge-pull` tool and its description, the claude-code module's instructions and playbook 07's merge step, which name
one pull request). The verb answers `call` (§766) at once and the members' outcome when the call ends. Built in H1 in
the forge module's `merge-pull` handler and the seat `git`'s contract in mesh-delivery-contracts. How it is checked:
whole (forge module): a call naming two pull requests in two repositories merges both and says one `merge-called`
(§736). Replaces G §38's one-pull form.
- §735 **Every member is judged before any is merged.** (ADR 0682 §1, §13; ADR 0666 §11; principle 3) The verb judges
each member as ADR 0666 §11 does, but for the branch protection (still the pull request's head; no run at the head's
build-input fingerprint failed a `required` step), all of them first. It judges no branch protection: a watched
branch need not be protected (ADR 0682 §13), and a head moved outside the mesh, a push included, is the watchdog's
untaken head (§740). When one member is
refused, none is merged, and the answer names every refused member and why. Then it merges the members in the order
the call names them. The forge module's `failed-steps` that this judges is written from the runner's `step-ended`,
whose body names the mover, its commit and its build-input fingerprint (H-A-18). Why judging first: a group merged in
part lands half a delivery. Built in H1 (forge module). How it is checked: whole (forge module): a call whose second
member's head moved merges neither and names it; a head at whose fingerprint a `required` step failed is refused,
naming the step (acceptance test 114).
- §766 **The call is written before its first merge and said again at the forge module's start.** (H-A-18; principles
3 and 4) Before it merges anything, the forge module writes the call's key `call.<call>` in its own template state
`calls` (value: `by`, the bus account, `at`, the members, each with `state` `judged`, `merged` with its commit, or
`refused` with why), and updates each member as it merges. At its start it reads every call whose event it has not
said (`said` false) and says `merge-called` for what landed, so a stop between the last merge and the event loses
nothing it recorded. A call is kept `call-retention` (a setting, default 30 days), then deleted by the forge module.
Its tools: `forge_calls` (the calls, newest first, with each member's state) and `forge_call_say_again` (with `call`
and `why`, says its `merge-called` again, written in the hand-act log). A stop between a merge and the write of that
member's `merged` leaves a head the call never names: it is an untaken head (§740), recovered with `gate_head_take`
(§742). Built in H1 (forge module). How it is checked: whole (forge module): a restart after the second of three
merges says one `merge-called` naming the two that landed; acceptance test 111's stop case.
- §767 **A member refused after every member was judged is said and raised.** (H-A-1; principle 3) When the forge
refuses a member after all were judged (a conflict that appears between the judging and its merge, the forge down
mid-call), the forge module stops merging, says `merge-called` with the members that landed and `refused`, one
`{repository, number, why}` per member not merged, and raises the finding `merge-partial`, scope `delivery`, id the
call, naming each refused member. The landed members are delivered as any call's. It clears the finding when a later
call lands every refused member (read from its own `calls`), or by its tool `forge_partial_clear` with `call` and
`why`, written in the hand-act log. A call that landed no member says no `merge-called`. Built in H1 (forge module).
How it is checked: whole (forge module): a fixture forge that refuses the second merge gives one `merge-called` with
the first member and `refused`, and `merge-partial` raised and cleared by a later call (acceptance test 115).
- §736 **The merge call says one event.** (ADR 0682 §1; principles 1 and 11) Once its members are merged or refused,
the forge module says the seat `git`'s event `merge-called` on `mesh.seat.git.event.merge-called.<call>`, kept on
the stream `SEAT_git_merge-called` for `merge-called-retention` (a setting, default 30 days) and up to
`merge-called-cap` (a setting, default 64 MiB), its body in the seat's `bodies`: `call` (16 lowercase hex, made once
per call), `by`, `account`, `at` (the time of the last merge), `members`, one `{repository, number, branch, head,
commit}` per landed member, where `branch` is the pull request's base (a watched branch) and `commit` the merge
commit the forge records for that pull request, and `refused` (§767). Only the forge module says it. Its one consumer
is the gate (§737), on a durable consumer of its program. Why 30 days: the gate takes in a call heard after a long
stop, as the forge module's `seen` keys are kept (G §585). Built in H1 (forge module, the seat `git`'s definition and
contract). How it is checked: unit (the seat definition: the event and its body); whole (forge module) as §734.
- §737 **The gate names the merged heads one source revision per call and opens the record.** (ADR 0682 §1, §2; H-A-13,
H-A-19; principles 1, 3 and 12) The gate consumes `merge-called`. For each watched branch its members name, it writes
`source.<branch>` once, by compare-and-set keyed by the call (the revision carries `call`, and a revision of that
branch already carrying that call writes nothing): the branch's previous heads with each member's repository at its
`commit`, where two members of one repository on one branch name the last member's commit in call order, numbered one
above the branch's newest. It opens one delivery record per such revision, keyed `<branch>.<source>` (A-33 stands),
its members the call's members on that branch, `call`, `by` and `account` in its trace; says `record-opened` on
`mesh.seat.mesh-delivery.event.record-opened.<call>` with `branch` and `source`, so the session that called can wait
on its own call; and says `source-moved` on `mesh.seat.mesh-delivery.event.source-moved.<branch>.<source>` with
`branch`, `source`, `call`, `start` (false; true only for a branch's first revision, §768) and `heads` (the
repositories it moved, each `{repository, commit}`). A call whose members name two branches is two source revisions
and two records, each naming the call. The gate numbers every source revision; no other part keeps a number of its
own. `source-moved` is kept for `source-moved-retention` (a setting, default 30 days). Built in H1 in the gate: its
`moveSource` moves from the `take-in.taken` handler to a `merge-called` handler; `ForgeMerged`, `moveHead` and
`gate_source_follow` go (§764). How it is checked: whole (gate): a `merge-called` with two members on one branch gives
one revision, one record and one `source-moved`; heard twice it gives one; two members of one repository give its
last member's commit. Replaces A-34, G §291, A-431's numbering by the indexer, and A-464 to A-464e.
- §738 **A delivery record follows its pipeline to the machines.** (ADR 0666 §9; G §141 to §147) The record opened by
§737 is the record of G §141 to §147, one per source revision: it consumes the runner's `pipeline-started` and
`pipeline-ended` (§746) of its source revision, and of the pipelines `package-published` starts from it (§747), the
parts' `wait-held` and `wait-ended`, the assigner's `wanted` and the node-engine seat's `running`, on the gate's own
durable consumer; writes its trace (the assignment graph revisions it moved between, from `wanted`); and ends as G §147
says, or `failed` when the assigner says a tier late (§751). A take-in with no movers ends its record at once as
`nothing to deliver`. Built in H3 in the gate's consumer (`stepFamilies`) and record (`SetTrace`, `EndIfRun`). How it
is checked: whole (gate): a record of a one-module delivery ends when the module runs on every present placing
machine.
### The forge bridge's one message, and the watchdog
- §739 **The forge bridge says one message per move of a watched branch's head.** (ADR 0682 §3; G §11, A-236) On every
`git.branch-moved` of a watched branch, and at its start for every watched branch's head, the bridge reads the head
through `git.branch` (never the event's body) and writes one message `head-moved` on its stream `forge`, subject
`forge.<repository>.<branch>.head-moved`, kept `head-moved-retention` (a setting, default 7 days): `repository`,
`branch`, `head`, `before` (the head it held for that branch in its state `heads`, empty the first time) and `start`
(true when said at its start). It writes it whether the head moved by a merge call, a merge in the forge's screen or
an admitted push: it reads no pull request and no tree. A head equal to the one it holds writes nothing. It
reads no branch protection: an unprotected watched branch moves like any other, and `branch-unprotected` (G §17)
ends (ADR 0682 §13). Its one consumer is the gate (§740). Built in H1 in the bridge; the messages `merged` and `pushed`, their `base`, their trees in its state
`inputs`, and its call of `git.pull` for a merged commit go (§764). How it is checked: whole (bridge): a merge, an
admitted push and a restart each give one `head-moved`; a head heard twice gives one. Replaces G §509 and §714's
message form.
- §740 **A head no merge call delivered is an untaken head.** (ADR 0682 §3; principles 1, 3 and 11) The gate consumes
`head-moved` and keeps, in its template state `heads` (one key `<repository>.<branch>`, value `{head, since}`, deleted
by the gate when `source.<branch>` names that head), each head that `source.<branch>` does not name for that
repository. When `untaken-bound` (a setting of the gate, default 10 minutes, at least 1) passes and `source.<branch>`
still does not name that head, the gate raises the finding `head-untaken`, scope `delivery`, id
`<repository>.<branch>`, naming the repository, the branch, the head, since when, and "moved outside the mesh:
delivered by nothing". It clears it when `source.<branch>` names that repository at the newest head `head-moved`
said (a later merge call or §742's hand take). The gate keeps one deadline timer, set to the earliest `since` plus the
bound, and no other clock. Built in H1 in the gate. How it is checked: whole (gate, bridge): a head moved by a merge
call raises nothing; a head pushed outside the mesh raises `head-untaken` within the bound, starts no pipeline, and
clears when a later merge call takes its repository in (acceptance test 111). Replaces A-463's `merge-untaken` and
the head half of G §12. An acknowledge tool that silences a known `head-untaken` without delivering it was proposed
by the review and is not built: no record or answer asks for it (noted for H4's tools table).
- §741 **`pipeline-stands` times a pipeline's steps, not a head.** (ADR 0682 §7; ADR 0666 §24; principle 7) The
watchman's finding `pipeline-stands` (G §12, §75) is raised for a pipeline step standing still five minutes; a watched
branch's head that starts no pipeline is the gate's `head-untaken` (§740) alone, so the watchman learns no head. A
pipeline waiting its turn (§748) is progress not owed and raises neither. Built in H1 in the watchman's `takein` and
`signals` parts. How it is checked: whole (watchman): a pipeline waiting behind another raises nothing.
- §742 **Taking an untaken head in by hand.** (ADR 0682 §3; principle 4) The gate's tool `gate_heads` reads the state
`heads` with each head's since and finding; `gate_head_take`, with `repository`, `branch` and `why`, names that head
a source revision as §737 does, its `call` `hand.<by>.<time>`, written in the hand-act log, and the chain follows.
Nothing else delivers an untaken head. Built in H1 (gate tools). How it is checked: unit (gate tools); whole (gate):
a hand take writes one revision and clears the finding.
- §768 **The clean state seeds every watched branch's first source revision.** (H-A-15; ADR 0682 §12; principles 2 and
3) Before the runner starts at milestone H1's clean state (the plan's section 5), the gate's tool `gate_source_seed`
(with `why`, written in the hand-act log) writes, for every watched branch the registrations name, one source
revision at each repository's current head read through `git.branch`, with `start` true and `call`
`seed.<by>.<time>`, and says `source-moved`; the indexer takes each in with no mover (§743), so the newest catalog
revision of every branch names those heads, and the forge bridge's `head-moved` at its start then matches
`source.<branch>` and no `head-untaken` is raised mesh-wide. The tool answers the branches and repositories it wrote,
counted. A branch newly watched later starts the same way, by A-401's start pass, which stays and writes through this
path. Built in H1 (gate tools; indexer). How it is checked: whole (gate, indexer, bridge): after the seed a bridge
restart raises no `head-untaken`; live, the counts read at plan step 5.
- §769 **`predefined` is read from the take-in of the catalog's commit.** (ADR 0681 §2; ADR 0682 §1) The indexer's state
`predefined`, which ADR 0681 §2 has it read from the forge bridge's merged message of the catalog repository, is read
at the take-in of a source revision that moves the catalog repository, from that commit's trees (§743), the merged
message being gone (§739). Built in H1 (indexer). How it is checked: whole (indexer): a merge call landing a catalog
commit that changes a predefined key rewrites `predefined`. ADR 0681 §2's words "the forge bridge's merged message"
are owed an insight in that record.
### The take-in
- §743 **The indexer takes in what the gate names.** (ADR 0682 §4; A-476, A-476a; principles 7 and 10) The indexer
consumes the gate's `source-moved` on one durable consumer per watched branch, named `<node>_<indexer>_<branch>` (the
branch's name escaped as a key's name is), each delivering one message at a time, acknowledged once its catalog
revision and `take-in.taken` are written. For each repository the event moves it reads the trees at the named commit
through `git.trees` (one of its kept calls, A-190), computes the build-input fingerprints of the modules registered for
that repository (a registration's git origin), and takes as movers those whose fingerprint differs from their entry
in the branch's newest catalog revision, and every module, in any repository, that imports a package of one of them
(A-476a; ADR 0573 §3). It writes one catalog revision of that branch, keyed `<branch>.<source>` by compare-and-set,
pins it for the pipeline of §746 in the same write (A-449a, key `pipeline.<branch>.<source>`) when it has a mover,
and says `take-in.taken` once, with `branch`, `source`, `revision`, `call` and `entries`, one `{module, commit,
fingerprint, repository}` per mover (§763 corrects G §476 and §648: no `pull`). It validates no manifest and holds
nothing for the indexer a revision names; it takes in with the build it runs (A-476 (a), (b)). A `source-moved` with
no mover, a branch's first revision (`start` true, §768) included, says `take-in.taken` with no entries and pins
nothing (H-A-19); its entries are the base the branch's later take-ins are held against. Built in H1: the indexer's
take-in part written again (today `Arrive`, `evaluate`, `takeInWhole`, `takenWhat`, `nextSource` and
`ApplyBranchSource`, fed by the bridge's messages); its storage, entries, pins' state, versions and verbs
(`KeepView`, `keepBranchEntries`) stay. How it is checked: whole (indexer): a `source-moved` naming a mesh-sdk commit
names mesh-sdk and its importers; one changing only a document names no mover and pins nothing; a restart between
the revision's write and the acknowledgement writes no second revision. Replaces G §16, §19, §509, §712 to §715 and
the main file's *Taken in by the indexer it names* (§764).
- §744 **A take-in that cannot be written is held, said and skippable.** (ADR 0682 §4; H-A-12; principles 1, 3 and 4) A take-in that
cannot be written (the catalog store `full`, G §181) is held: its `source-moved` stays unacknowledged and is not
redelivered on the acknowledgement timer (the consumer's wait for an acknowledgement is longer than any hold, and it
is redelivered with no delay only when the indexer asks); the indexer takes it again when the bus module's `room` for
that store wakes it, with no timer. A `git.trees` call unanswered is its kept call's own (A-190) and holds the same
way. While held the indexer says `take-in.pending` once per source, with `branch`, `source` and `why`, and the
watchman raises `take-in-waits` when no `take-in.taken` for that source follows within `take-in-bound` (a setting,
default 60 minutes; G §497). Take-ins of one branch go in source order; another branch's go on. `indexer_pending`
reads each branch's held `source-moved` and why; `indexer_take_in_pending` takes it again now; `indexer_take_in_skip`,
with `branch`, `source` and `why`, acknowledges it unwritten, says `take-in.taken` with no entries and `skipped`, and
writes the hand-act log. The state `pending`, its keys `m.<sequence>` and `g.<group>`, and `indexer_drop_pending` go
(§764). Built in H1 (indexer). How it is checked: whole (indexer): with the store full a `source-moved` is held, said
once, taken in on `room`; a second branch's take-in goes on meanwhile; a skip ends it in the hand-act log (acceptance
test 116). ADR 0682 §4 keeps `take-in.pending`, `take-in-waits`, `indexer_pending` and `indexer_take_in_pending` and
ends `indexer_drop_pending`.
- §745 **Manifest validation is the pipeline step `validate`.** (ADR 0682 §5; H-A-6; A-476 (a); ADR 0666 §17) The step
`validate` is served by the assigner as an accepted kind `step` of the seat `mesh-wants`, the seat that serves the
composition step (G §22), asked by the runner for each mover as the step before its composition step: the assigner
runs the manifest validator built from the take-in's own heads, in the sandbox, as it runs the composer (G §718), over
the revision's manifests, refusing only a problem in a manifest the revision alters or one between manifests an
altered one takes part in (ADR 0578 §2). It ends `failed` with the reason `manifest-refused` naming each problem, or
`cycle` naming the cycle the revision makes (the cycle refusal A-326 and G §569 made at the take-in), and `passed`
otherwise. Its status is `required` in every policy by default: the seed (A-441) writes it, and §22's list of steps
gains it. The indexer's events `manifest-refused` and `manifest-accepted` (G §500, §668) go; their readers read the
step's end (§764). Built in H1 (assigner: the step; runner: the order; indexer: the seed's default). How it is
checked: whole (assigner, runner): a take-in whose manifest names an unknown field fails `validate` naming it and asks
no composition; a take-in making a cycle fails it naming the cycle (acceptance test 117).
### One take-in, one pipeline, none in parallel
- §746 **One take-in is one pipeline.** (ADR 0682 §6; the operator, 2026-11-20 01:10 CET; ADR 0666 §1, §7, §9 read per
source revision) On a `take-in.taken` with entries the runner starts exactly one pipeline, keyed `<branch>.<source>`
by compare-and-set on its absence, whose movers are the event's entries. It asks each mover's pipeline steps in the
catalog graph's build order (a mover's build waits for the `package-published` of each mover it has a "published
before it builds" edge to, toolchain edges first), and computes the movers' delivery order in ADR 0666 §14's order as
tiers (acting parts first, providers before their consumers; movers no edge orders share a tier). The pipeline passes
when every mover's `required` steps passed, a `required` step ending `nothing to run` counting as passed (ADR 0682 §8; H-A-3), and
fails when one did not. Its events (G §73) move to `mesh.seat.mesh-pipeline-runner.event.<event>.<branch>.<source>`
for `pipeline-started` and `pipeline-ended`: `pipeline-started` names `pipeline`, `branch`, `source`, `call`,
`revision` and the movers; `pipeline-ended` is said once with `pipeline`, `branch`, `source`, `call`, `revision`, the
outcome and, per mover, `module`, `fingerprint`, the released `version` (absent when none was released), its step
results and its `tier`. The step events keep their subject per module (`...<event>.<module>.<branch>.<commit>`) and
gain `pipeline`. Who hears them now: the assigner (wants, §750, §751), the gate (the record, §738), the indexer (pins,
§749), the status view, and a session, which waits on `record-opened.<call>` (§737) and then on
`pipeline-ended.<branch>.<source>`, never per module (G §74's wait per module and G-4's subject row are corrected so,
§763). The runner says no `group-ended`. Built in H1 in the runner (the key `<branch>.<source>.<module>` becomes
`<branch>.<source>`, its step keys gain the mover) and in the seat's definition. How it is checked: whole (runner):
a take-in naming mesh-sdk and one importer is one pipeline whose importer builds after mesh-sdk's
`package-published`, and one `pipeline-ended` naming both with their tiers (acceptance test 112).
- §770 **`pipeline-failed` is one finding per failing mover.** (H-A-19; ADR 0666 §24; ADR 0505 §5, §6) The runner's
finding `pipeline-failed` (G §33, §113) has id `<module>.<branch>`, one per mover whose `required` step failed or
could not run, naming the pipeline and the step, the commit's author's for two hours then the operator's; it is
cleared by that module's next passed pipeline on that branch. Built in H1 (runner). How it is checked: whole (runner):
a pipeline with two failing movers raises two findings; a later pipeline passing one clears that one.
- §747 **A start names an event, never a sweep or a comparison.** (ADR 0682 §6; H-A-19; G §12, §112; principle 1) The
runner starts a pipeline on `take-in.taken` with entries, and on `package-published`, whose body gains `branch` and
`source`, the source revision whose pipeline published: when that pipeline runs and moves the dependent, the
dependent waits in it (§746) and nothing starts; otherwise (a publish again of an older build, G §588, or a kind-1
dependent that pipeline does not move) one pipeline keyed `<branch>.<source>.published.<package>.<version>` starts,
its movers the package's kind-1 dependents it does not move. That pipeline pins the publishing source's catalog
revision and belongs to that source's delivery record (§738). A hand `start`, `run`, the runner's other verbs (ADR
0666 §10, §18, §25) and its taking up of pipelines in flight after a restart (§19) stay, as do the set-aside keys of
A-446. The sweep (`StartUnstarted`, `Unstarted`; `StartHeld` of the held pull request mesh-delivery #262), the
start-time `outdated` of G §375 and `runner_catch_up` go: the runner's durable consumer of `package-published`
delivers a publish heard after a stop. Built in H1 (runner: `Starter`, `run`'s `resume`). How it is checked: whole
(runner): a restart with modules whose fingerprint has no pipeline starts none; a publish heard after a stop starts
its dependents once.
- §748 **At most `pipelines-at-once` pipelines run; the others wait in order.** (ADR 0682 §7; H-A-14; the operator,
2026-11-20 01:10 CET; principles 3 and 11) The runner's setting `pipelines-at-once` (default 1, from 1 to 8) bounds
the pipelines whose steps are asked at once. A pipeline written over the bound gets `seq`, the next value of the
runner's own counter in its state (one key `queue.next`, written by compare-and-set), and state `waiting`, with no
step asked; it says `wait-held` with `step` `pipeline` and `on` `pipelines-at-once`. The runner places the lowest `seq`
waiting at its start, from its own state, and on every end of a running one: `pipeline-ended`, a cancel and a
set-aside alike, with no timer, and says `wait-ended`. A pipeline waiting on the operator's `approval` step does not
count against the bound, and when its approval is answered it is placed first, ahead of every `seq`. When pipelines
wait and none runs after a placing, the runner raises its finding `pipelines-stalled`, naming the first waiting, and
clears it when one runs. At most `pipelines-waiting-cap` pipelines wait (a setting, default 64): a start over it
leaves its `take-in.taken` unacknowledged, raises `queue-full`, and is taken on the next end. Superseding (ADR 0682 §7): a waiting
pipeline's mover that a newer waiting pipeline of the same branch also moves leaves the older one, which ends
`superseded` when no mover is left; a running pipeline runs to its end (G §31's supersession of running steps no
longer applies across pipelines). A waiting pipeline is progress not owed (§741). The runner's tool `runner_waiting`
lists the waiting pipelines in their order; `runner_place`, with a pipeline's key and `why`, places one ahead of its
turn, written in the hand-act log. Built in H1 (runner). How it is checked: whole (runner): two take-ins in a row run
one after the other, the second said waiting; a cancel places the next; an approval answered places first; a newer
take-in moving the same module supersedes it in the waiting one (acceptance test 112).
- §749 **A pipeline's pins are the indexer's, freed on its end.** (ADR 0666 §9; A-449a; principle 12) The indexer pins
a take-in's revision in the take-in's write (§743), and a pipeline §747 starts on its `pipeline-started`, which
carries `revision`; it frees the pin on that pipeline's `pipeline-ended`, both heard on durable consumers of its own.
At its start it judges each kept revision against the pins its own state holds, never against a count of pins
stated. The runner says no `pinned` or `freed`; §153's rule that only the gate's account says them stands. Built in
H1 (indexer: `JudgePins`; runner). How it is checked: whole (indexer): a restart while a pipeline runs on revision
510 keeps 510. Corrects A-449 (§763).
@@ -0,0 +1,197 @@
---
layer: to-be
status: in-progress
design: 47
milestone: H
updated: 2026-11-20
decisions:
- "ADR 0682"
- "ADR 0666"
- "ADR 0656"
- "ADR 0596"
- "ADR 0595"
- "ADR 0573"
- "ADR 0548"
- "ADR 0525"
---
# 47, milestone H: the want rule, the builder's lines, the reads of other parts and the missing read tools
One file of milestone H's part of [design 47](../47-delivery-from-commit-to-delivered.md); the part opens at
[H-deliver-the-delivery-system](H-deliver-the-delivery-system.md), and its lines are numbered after
[H-1](H-1-the-merge-call-and-the-take-in.md)'s and cited as "design 47 H §n". These are items 8 and 11 to 13 of the
plan's section 4, and A-434 as a line. The wake-up rule (item 9) is one rule for every delivery part and is in the
main file, *Every delivery part wakes on its inputs alone*; §762 here says which part moves onto it where.
### The want rule
- §750 **A want names the newest released version whose pipeline passed.** (ADR 0682 §8, §9; A-478; the operator,
2026-11-17 about 21:45 CET; principles 2 and 7) For every machine whose assignment places module M on watched branch
B (the placement's branch, the registration's first by default), the assigner writes `want.<node>.<M>` naming
`version`, the highest version of M released on B (`version-released`) whose pipeline's `pipeline-ended` passed, with
`pipeline` (the pipeline's key) and `rank` (the mover's tier in it, §746); it writes it as §751 orders, by
compare-and-set on a newer version, so an older end heard later writes nothing (G §352's state `written` keeps, per
module and branch, the newest passed version). A want names a version, never a part's digest (A-430): A-436's
stand-in ends, and G §379's digests go. The one exception is a return to an earlier build: the placement's value
`build` (G §361), set through the seat `mesh-module-assigner`'s `propose` like any placement change and read through that seat's verb
`placement` (A-62), whose answer shows the `build` value beside the branch, is read before
the rule and names the want's version (H-A-5; ADR 0682 §8). The want is held for nothing else. These holds end, with
their states, findings and tools: the canary and the "at most `step` ahead" rule, every provider healthy, the grant
exists (ADR 0429 §6), the served verb, the bus state (rule 8), the acting part's new version first (ADR 0656 §4; G
§266, §268, §269, §274, §276: `acting-part-late`, `assigner_waits`'s acting rows, `assigner_release`), the
dependency's `version-delivered` (G §376, A-189, A-246, A-250: `delivery-waits`, `assigner_delivered_again`, the event
`version-delivered`), and the parts check (G §524, the `want-parts` capability). The backup before the apply and the
operator's yes on a failed backup and on a held revert are the node-engine's and the asker's (A-478 (c); G §296,
§710); a failing new version holds on its own machine while the old runs, by its node-engine's apply (ADR 0682 §9).
Built in H1: the assigner's want rule (`wants.go` and the want half of its pass) written again; the old holds
deleted. How it is checked: whole (assigner, in mesh-delivery): a module released with a passed pipeline is wanted on
every placing machine, whatever build an acting part runs; a version whose pipeline failed is not wanted; an older
end heard after a newer one writes nothing; a placement `build` names the want's version (acceptance test 113).
Replaces G §29's per-revision want, §291 and §292 in their want half, and design 47's six rules in *Where it goes*.
- §751 **One pipeline's wants are written tier by tier.** (ADR 0682 §8, §9; H-A-10; ADR 0666 §14; principles 3 and 7) On a
passed `pipeline-ended` the assigner writes the wants of its tier-1 movers on every placing machine; once each of them
reports `running` at that version on every present placing machine (the node-engine seat's `running`), it writes the
next tier's, and so on. Each tier has `tier-bound` (a setting of the assigner, default 30 minutes): a tier whose
movers do not all run within it ends the delivery loudly: the assigner writes no later tier, raises its finding
`tier-late`, scope `delivery`, id the pipeline, naming the movers and machines not running, and says `wanted` with
`outcome` `tier-late`, on which the gate ends the record `failed` (§738). A machine away is not waited for; it
applies its wants when back. The validate and composition steps composed the candidate against the whole graph
before any want (ADR 0682 §9). The wants of one tier are written at once. Built in H1 (assigner). How it is checked:
whole (assigner, in mesh-delivery): a pipeline of a provider and its consumer writes the consumer's want only after the
provider's `running`; a provider that never runs raises `tier-late` within the bound and writes no consumer want.
- §771 **The order on one machine is the node-engine's apply order, by rank.** (H-A-9; ADR 0682 §9; A-478b; principles
9 and 12) The node-engine, which applies its wants one module at a time (R6), takes its changed wants in this order:
the holder of the seat `node-backup` and the bus module first; then each pipeline's wants by `rank`; then the rest by
name; its own want last; at a revert, the reverted module first. It reads the order from the wants alone and consumes
no event of the runner. The assigner's order waits (`order.<node>.<module>.<part>` in its state `waits`, step
`placing`, G §298, A-245, A-258 to A-260) go. The node-engine's tool `engine_apply_order` answers the order it will
apply its changed wants in, with why; the assigner's `assigner_delivery_order` of the first version of this part is
not built. Built in H1 (node-engine apply pass; assigner). How it is checked: whole (node-engine, in `mesh-host`):
wants of the bus module, the backup holder and two movers of ranks 2 and 1 changed at once apply in that order.
- §772 **A machine away at the clean state is placed by hand when it is back.** (H-A-15; principle 2) A machine away at
milestone H1's clean state runs a node-engine that cannot read a want without digests, so the assigner keeps its wants
held, said as `held` with why `awaits its hand placement`, until its node-engine says, through the seat
`node-engine`'s capability event `manifest-reads-changed` (G §590), that it reads a want by version; when the machine
is back, its hand placement (the plan's section 5, step 5) runs before anything else on it. Built in H1 (assigner).
How it is checked: whole (assigner): a node whose node-engine lacks the capability gets no want of the new form and is
said held; live, plan step 5 reads it.
### Accounts
- §752 **Every account may call every verb; a module publishes only what it declares.** (ADR 0682 §10; A-434, H-A-11;
ADR 0666 §21; principle 6) The grants part grants every module's account every verb and every subscription; it grants
a module's account publication only of the events, step results and state its manifest declares (ADR 0666 §21 stays),
and an enrolling node's account keeps its own grant. No seat definition names callers, and the `callers` lists of
every seat contract go (G §610's class defaults are not built); approval marks stay. Acceptance tests 40, 51 (its
callers row) and 85 are not checked. Built in H1 (seat contracts in mesh-contracts and mesh-delivery-contracts; the
grants part). How it is checked: unit (seat definition reader: a definition naming callers is refused); whole (grants
part, in mesh-delivery): a module calls a verb its manifest does not invoke and is answered; it publishes an event
its manifest does not declare and is refused; an enrolling node's account is held to its grant (acceptance test 118).
### Builds and the builder
- §753 **A build module takes deliveries first.** (ADR 0525 §4; G §86; H-A-8) Each build module pulls its delivery queue
first, then its check queue, then its recheck queue, taking from a later queue only while the earlier is empty, and
waits on the bus's pull with no idle re-pull: the five-second idle pull of the shared step runner in
mesh-delivery-sdk goes. The turns the code takes today cite an operator's yes of 2026-11-15 that no record holds; if
H2's prepare finds it, ADR 0525 gets an insight. Built in H2 (mesh-delivery-sdk's step runner, `buildrun`). How it is
checked: whole (Go build module, in mesh-catalog): with three queues full the delivery asks run first (acceptance
test 119).
- §754 **A done build with no reservation is reserved by the builder itself.** (G §369, A-455; principle 3) When the
runner's build step asks a fingerprint whose build is done and carries no reserved version, the builder reserves the
next version for it at that ask, says it in its journal and in the step's end, and answers the step carried. Nothing
else reserves an old build. Built in H2 (builder: its reserver, and the look that writes "done build no version
reserved", which goes). How it is checked: whole (builder, in mesh-delivery): a build step for a done, unreserved
fingerprint ends carried with a version (acceptance test 119).
- §755 **A failed fingerprint built again keeps its reservation.** (G §369, A-490b; principle 7) When a key is built
again (a step ask, an event or `builder_build_again`), the reserver takes that key's own standing reservation as its
number; a key with none takes one above the higher of the highest released and the highest any other key reserved.
No number is skipped. Built in H2 (builder's reserver). How it is checked: unit (reserver): a failed 0.1.19 built
again is 0.1.19; acceptance test 119.
- §756 **A build is entered only by the runner's build step.** (G §473, A-97) The builder builds only what a pipeline's
build step asks: it builds no catalog entry it hears and follows no gate `source`; the pre-G numbering beside the
reservations goes, with its floor wait. `build.ended` carries the reserved `version` (A-430b), for every watched
branch. Built in H2 (builder). How it is checked: whole (builder): a catalog entry heard with no step ask builds
nothing; `build.ended` names its version (acceptance test 119).
- §757 **The steps a mover can run are asked at every pipeline.** (ADR 0666 §7; G §8, §22; H-A-3) At each pipeline's
start the runner asks each build seat's `steps` for each mover's artifacts, never only at the seed; a step the answer
says has nothing to do ends `nothing to run`, which counts as passed for the pipeline (§746; ADR 0682 §8). Built in H2 (runner).
How it is checked: whole (runner): a library module whose unit tests vanish ends that step `nothing to run` and its
pipeline passes (acceptance test 119).
- §758 **A registration whose pipeline would have no step is refused.** (ADR 0666 §2; acceptance test 67) `register`
and `policy` refuse a registration whose policy, read against the build seats' `steps` for its artifacts, leaves no
step that runs, naming each artifact and step. A package module with no registration has no pipeline and is never
built; the seed (A-441) registers every package module the graph holds, `mesh-delivery-sdk` among them. Built in H2
(indexer: `register`, `policy`, the seed). How it is checked: whole (indexer): a registration with every step
`bypass` is refused naming them (acceptance test 119).
### Reads of another part's data become its events and verbs
- §759 **The status view and the watchman read no other part's bus state.** (ADR 0596 §2, §4; principles 1 and 9)
Each read of another module's bucket becomes the owner's event, and the first view the owner's seat verb, in H4:
the node-engine's `have` becomes the seat `node-engine`'s event `running` and, for the first view, the verb `haves`
the seat gains (what the node runs per module, from its own state); the builder's `builds` becomes the seat
`mesh-builds`' `build.ended` and the verb `builds` with `since`; the indexer's `taken` becomes `take-in.taken` and the
seat `mesh-catalog`'s verb `revision`; the watchman's `presence` and `stuck` become the seat `mesh-watchdogs`' verbs
`presence` and `stuck` and the findings it says (`*.finding-written`, `*.finding-cleared`, G §386's form); the
indexer's `candidates` and the teller's `answers` go with their uses (G §512). The watchman's read of the
provisioner's `provisions` becomes §773's event. Every other bucket watch of the view the H4 prepare lists against an
owner's event; one with none stops that subtask for the lead. How it is checked: unit (each manifest declares no
`reads` of another module); whole (view, watchman, in mesh-delivery) against the owners' events.
- §773 **Four owners say their changes as events, for H4b.** (H-A-4; ADR 0596; principle 1) Each is a seat event kept
on its seat's stream for 7 days (a setting of its owner), with a first-view verb; until H4b builds them, the reads
they replace stand and are named leftovers in the gap table:
- the seat `mesh-provisioner`: `provision-changed` on `mesh.seat.mesh-provisioner.event.provision-changed.<provision>`,
body `{provision, reference, state, at}`; first view its verb `provisions`;
- the seat `mesh-vault`: `account-changed` on `mesh.seat.mesh-vault.event.account-changed.<user>`, body `{user, hash,
at}`, and `node-login-changed` on `...node-login-changed.<node>`, body `{node, hash, at}`; first view its verbs
`accounts` and `node-logins`;
- the seat `mesh-grants`: `user-list-changed` on `mesh.seat.mesh-grants.event.user-list-changed`, body `{revision,
users, at}`, consumed by the bus node's agent in place of its watch of the key `user-list`; first view its verb
`user-list`;
- the enrolment module: `enrolment-changed`, body `{node, state, at}`; its seat and its first-view verb are read
from the module's manifest at H4b's prepare, which writes them into this line.
The grants part then consumes them, and calls the bus module only through the seat `mesh-broker`'s verbs (`list`,
`show`, and `users` and `streams`, which the seat gains in place of the module tools `nats_users` and `nats_streams`
the grants part calls today). Built in H4b. How it is checked: unit (the owners' seat definitions); whole (grants
part): a changed account reaches the user list with no read of the vault's bucket.
- §760 **Rates on the bus are read on the seat `mesh-broker`.** (ADR 0595 §5; principle 4) The seat `mesh-broker`
gains the read verb `rates`, with `window` (seconds, default 10, from 1 to 60): it reads the bus server's counters of
each connection and each stream twice, `window` apart, within the call, and answers per connection (its account's
module and node) and per stream the messages and bytes in and out per second, and the slow consumers counted. Its call
is bounded at `window` plus 5 seconds, at most 65 seconds. A verb's call timeout is a bound per verb in its seat
definition, `timeout`, defaulting to the tool runner's own (principle 11); `rates` names 70 seconds, and the tool
runner waits a verb's `timeout` when its definition names one. The field, its schema, the tool runner's wait and
`merge-pull`'s 120 seconds are built in H1 (H-A-43: a merge call of several members is cut at a fixed 30 seconds
otherwise); the `rates` verb itself in H4 (the bus module). How it is checked: whole (bus module): a fixture publishing 100 messages a second reads between 90 and 110.
### The read tools found missing
- §761 **One revision, narrowed, and its differences, in the indexer.** (ADR 0666 §25; principle 4) The indexer gains
the tools `indexer_revision` (`branch`, `revision`, optional `module` and `kind`: one catalog revision of any watched
branch, narrowed to a module's entries and edges or to one edge kind), `indexer_difference` (`branch`, `from`, `to`:
the modules and edges that differ), `indexer_build_order` (`branch`, `modules`: the order the build edges give, its
cycles named) and `indexer_built_from` (`fingerprint`: the source revision and commit it was taken in at);
`indexer_edges` (G §380) gains `kind`. The node-engine's `engine_build` answers its own fingerprint, version and
commit, and `engine_apply_order` is §771's. An answer over the bus's per-message limit is an object's name (ADR 0666
§16). The callers of tools no part serves go: the checker's of `indexer_read_candidates` (it goes, §764), and the
health checks' and the first views' of `indexer_read_modules`. Built in H4. How it is checked: unit (each tool's
answer); the principle-4 table of each module's pull request.
- §762 **Every delivery part moves onto the wake-up shape.** (ADR 0682 §11; main file, *Every delivery part wakes on
its inputs alone*) The parts and what goes, from the gap table of 2026-11-20 (part F, question 2): mesh-config's
key-by-key pass, five-minute sweep and minute's read (`sources`, the held pull request mesh-catalog #708, H1); the
grants part's whole recompose (`Compose`, H4b); the indexer's re-read of every module, graph key and manifest after
each write (`KeepView`, `viewEntries`, `heldModules`, H4); the builder's start sweep and `builder_stuck`'s read of
every key (`Sweep`; it answers from the builder's own index of what builds, H4); the assigner's input sweep and its
ten-second poll of `backed-up` (`inputKeeper.sweep`, `snapshotTheBus`, H4); the watchman's five-minute recount of
dead letters (H4); the build modules' stale round and idle pull (`EndStaleRound`, §753, H2); mesh-sdk's first view
asked again with a backoff (`views` reader, H4); the view's five-second retries (§759, H4); the forge module's
one-minute poll and ten-minute timer (H3). The decided pulses (G §496's minute of `step-progress`, G §606) are real
deadlines and stay. Its cost: the shape lives in mesh-sdk's state package, so its release rebuilds nearly every
module (every module that imports that package, about 120 by the graph of 2026-11-20); it is carried by H4 alone, one
release, delivered by the pipelines of the movers it names; H1's mesh-sdk change is only the seat definitions' copies
and the `timeout` field (H-A-42, H-A-43), published first. How it is checked: as the main file's rule says.
### Questions this file leaves open
None: the enrolment event's seat and first-view verb (§773) are owed to H4b's prepare.
@@ -0,0 +1,125 @@
---
layer: to-be
status: in-progress
design: 47
milestone: H
updated: 2026-11-20
decisions:
- "ADR 0682"
- "ADR 0681"
- "ADR 0667"
- "ADR 0666"
- "ADR 0633"
- "ADR 0595"
---
# 47, milestone H: the corrections made in place, the stale state, and what retires
One file of milestone H's part of [design 47](../47-delivery-from-commit-to-delivered.md); the part opens at
[H-deliver-the-delivery-system](H-deliver-the-delivery-system.md), and its lines follow
[H-2](H-2-wants-builds-reads-and-tools.md)'s, cited as "design 47 H §n".
- §763 **The small contradictions are corrected in place.** (plan item 10; each line names its source) These lines of
milestone G and the main file now say what holds, each marked "corrected in place by design 47 H §763":
- G §5: `register` and `watch` refuse no unprotected branch (ADR 0682 §13, which ends ADR 0666 §3; the operator,
2026-11-14, G-23 *Met during G's start*). G §17, G §38's and G §41's protection judgement of `merge-pull`: no part
reads a branch's protection (ADR 0682 §13, which ends ADR 0666 §11's protection judgement and §6's condition).
- G §654: mesh-config consumes no pull request event and calls no `git.pull` (ADR 0681 §1).
- A-449 and A-449a: the runner says no `pinned` or `freed`; the indexer pins and frees a pipeline's revision (§749).
- Acceptance test 53: its `forge-made` and checker rows go (G §151; §740; the checker goes).
- G §476 and §648: `take-in.taken` carries one entry per mover, `{module, commit, fingerprint, repository}`, no
`pull` (§743).
- G §40: the operator channel keeps no count of open asks per asker and offers every ask (ADR 0682 §14, which ends
ADR 0666 §12's three; A-424, the operator, 2026-11-15).
- G §386: the grants part makes one durable consumer per process that hears a `*` event, named as A-397 names it.
- G §73's subject table and G-4's row of the runner's events, and G §74's wait per module: `pipeline-started` and
`pipeline-ended` are said per source revision and waited for so (§746).
How it is checked: hq's review of this amendment reads each corrected line against its source.
- §764 **What retires at milestone H's delivery.** (ADR 0682 §1 to §4, §6, §8, §9, §13, §14; ADR 0681; ADR 0667 §12) Each is
removed in the release named, with its code, its state (dropped by §765), its tools, its events in the seat
definitions, its readers and its tests; nothing of it is kept unread.
- **The forge module (H1, H3):** the single-pull form of `merge-pull` (§734); the merge-check machinery: the
`mesh/merge-gate` pending statuses, the weighing of trunk and sibling moves (`announceHead`, `passPulls`,
`trunksMoved`, `siblingMoves`, `GateWriter`) and its files on disk, its one-minute poll and ten-minute timer (H3).
The forge's branch protections drop the required statuses `mesh/merge-gate` and `mesh/repo-check` in H1, through
the forge module's protection setting, so no merge waits on a status nothing writes (H-A-15).
- **The branch protection (H1; ADR 0682 §13):** `merge-pull`'s judgement of a branch's protection, the bridge's
read of it on every move and its finding `branch-unprotected` (G §17), and the protection rows of G §41's test.
- **The operator channel's limit (ADR 0682 §14):** any count of open asks per asker, and ADR 0666 §25's tool for
asks held back, which reads empty and goes.
- **The forge bridge (H1):** the messages `merged` and `pushed` and their `base` (§739), with their readers, the
indexer's take-in and the gate's `ForgeMerged`; its state `inputs` and its read of trees; its call of `git.pull`
for a merged commit; the group it sets on a member, `forge.group.merged`, `bridge_groups` and
`bridge_group_end_again` (G §43, §45, §157, §715); its hearing of the checker's `owed` and the merge act of a
switched key (`OwedActMerge`, its `invokes` of `seat:git.merge-pull`, G §510).
- **The gate (H1, H3):** the source per merged message (A-464 to A-464e: `ForgeMerged`, `moveHead`,
`gate_source_follow`, the start's comparison with the forge's heads) and its consumption of the bridge's `merged`;
the source moved on `take-in.taken` (A-34; §737); the groups from a branch name (G §42, §716, A-279): its states
`deliveries` and `groups`, its consumption of `pull-updated`, `pull-merged` and `pull-closed`, its start
reconciliation through `open-pulls` and `pull`, and the verb `groups` (the verbs `deliveries` and `show` answer
from the records, A-24); the switch and the carried approval (G §156, §169, §542, A-281: `approval-carried`,
`approval-carried-open`); the states `merges`, `owed` (the old bucket), `asks`, `pulls` and `turns`, and the seat
definition's entries for them.
- **The indexer (H1):** the take-in fed by the bridge's messages; the state `pending` with its keys `m.<sequence>`
and `g.<group>`; the hold for the indexer a revision names (ADR 0633 §1, the main file's *Taken in by the indexer it
names*) and `indexer_drop_pending`; the validation at the take-in (G §16, §19, §237, §712, §713; A-475, A-475a),
which becomes §745's step; its events `manifest-refused` and `manifest-accepted`, with their reader, the watchman's
findings `manifest-refused` and `manifest-outgrown` raised from them; the catch-up by comparison (A-463, A-463a,
A-490k, G §177's first-start half, G §362: `kickBehind`, `TakeInBehind`, `sayUntaken`, `takeInHeads`,
`kickCatchUp`, `CatchUp`) and its condition `merge-untaken`; the source numbering in memory (A-431,
`nextSource`); the 30-minute group wait of `take-in-waits`; the field `pull`; the config candidates (ADR 0681 §2).
- **The runner (H1):** the sweep (`StartUnstarted`, `Unstarted`; `StartHeld` of mesh-delivery #262, closed
unmerged); the start-time `outdated` (G §375) and `runner_catch_up`, never built; `group-ended` (G §332) and its
stream `SEAT_mesh-pipeline-runner_group-ended`; the pipeline key per module; its `pinned` and `freed`.
- **The assigner (H1):** the holds §750 names, with `acting-part-late`, `delivery-waits`, `version-delivered`,
`assigner_delivered_again`, the acting rows of `assigner_waits` and `assigner_release`; the order waits
(`order.<node>.<module>.<part>` in its state `waits`, §771); the parts check and the `want-parts` capability; A-436's
digests in a want.
- **mesh-config (H1):** its pull request path (ADR 0681 §1, §7) and A-490l's pull sweep.
- **The checker (H1):** `mesh-delivery-checker` whole, with its verdict verb, its calls of
`indexer_read_candidates`, `mesh/check` and its try; it is not placed at the clean state.
- **Every seat contract (H1):** the `callers` lists (§752).
- **Acceptance tests:** 38, 49, 81, 82, 105 and 106 as written; tests 110 to 119 hold what remains of them, and test
67 is read with a merge call naming its three repositories.
How it is checked: after each release, a search of the trunks of mesh-catalog, mesh-delivery, mesh-sdk,
mesh-contracts, `mesh-host` and mesh-tools for each name above finds it in no file outside tests that assert its
absence, and `mesh-broker.list` shows no bucket or stream of a state above.
- §765 **The stale state goes once, after the new programs run.** (ADR 0682 §12; H-A-15, H-A-19; the operator,
2026-11-20 about 01:10 CET: "I don't care about old data at this moment, I want a working mesh again") At milestone
H1's clean state, after the new programs run on every present machine (the plan's section 5, after step 5.6) and
before the runner starts, every state §764 retires goes, and with it: the gate's 833 open records; the indexer's pins
that name no running pipeline; the pipelines of the earlier model; the stopped runner's queue of `take-in.taken` of
the old form, with the consumer that holds it; every held or failed want. Each goes as ADR 0595 §4 removes kept data,
on that word, through the tool named for it: a record through `gate_record_end`, a pin through the indexer's
`indexer_pin_drop` (with `key` and `why`, which H1 adds), a pipeline through the runner's `cancel`, a held or failed
want through the assigner's `assigner_want_drop` (with `node`, `module` and `why`, which H1 adds), and every retired
state, once no running module declares it (an orphan `orphans` lists), through the seat `mesh-broker`'s `drop`. The
bus module's stream snapshot taken before the first drop is the rollback. The pack says what it deleted, counted.
Every durable consumer the new programs make starts at the newest message of its stream and refuses an event of the
old form (one with no field its new form requires) with a journal line naming it. The delivery records of the one
model start from the clean state. How it is checked: the lead reads `mesh-broker.list` and each module's read tools
before and after, and quotes the counts; the snapshot's restore is tested before the first drop (ADR 0595 §4).
- §774 **Tests per module, against the budget, and where H1's run before its merge.** (H-A-16, H-A-17; the test
budget, 50 integration and 500 unit tests per module) Each acceptance test of H is one whole test per module, in that
module's repository (tests 110 to 119 name which); the chain of test 110 is the live proof. The counts, read from the
gap tables of 2026-11-20 (unit, then integration), and what H adds:
| module | today | H adds (whole tests) | H retires |
|---|---|---|---|
| forge module (seat `git`) | 177 Go and 196 TypeScript, 1 | 1 TypeScript whole test of the seat `git` (§734 to §736, §766, §767; tests 114, 115) | the merge-check machinery's tests |
| forge bridge | 45, 8 | none: its whole test is rewritten for `head-moved` | the `merged`, `pushed`, group and owed-merge tests |
| gate | 38, 2 | 1 (the merge call, the record, the watchdog; tests 110, 111) | the source-follow, group and switch tests |
| indexer | 194, 13 | none: its whole test is rewritten for the take-in (tests 112, 116) | the pending, catch-up, validation and group tests |
| runner | 78, 3 | 1 (the queue; test 112) | the sweep's tests and `group-ended`'s |
| assigner | 126, 13 | none: its whole test is rewritten for §750 and §751 (tests 113, 117) | `TestEachRuleHoldsTheWantBack` and the hold tests |
| node-engine | 1,719, 10 | 1 (the apply order; test 113) | the want-parts tests; over the unit budget, every H change to it deletes at least as many unit tests as it adds |
| grants part | 78, 1 | none: its whole test gains §752's cases (test 118) | — |
| builder | 154, 18 | none: its whole test gains §754 to §756 (test 119) | the pre-G numbering's tests |
| tool runner | 152, 7 | 1 that starts it whole (H-A-16's gap) | — |
Every module stays within 50 integration tests; acceptance tests 38, 49, 81, 82, 105 and 106 retire (§764). Before
H1's merge its integration tests are run once locally against an embedded bus, as the mesh's instruction
*how-work-is-done* allows, and each pull request says so; the chain's live proof (the plan's section 5, step 7) is
H1's check, and from H1 on the pipeline makes the check (H-A-16, a rule weighed and said). The counts above are per
module; the count per release and the number of tests each retired test frees are each milestone's prepare's, written
in its milestone file. How it is checked: each pull request of H states its module's counts against the
budget; the review refuses a count over it.
@@ -0,0 +1,479 @@
---
layer: to-be
status: in-progress
design: 47
milestone: H
updated: 2026-11-20
decisions:
- "ADR 0682"
- "ADR 0681"
- "ADR 0667"
- "ADR 0666"
- "ADR 0656"
- "ADR 0633"
- "ADR 0596"
- "ADR 0595"
- "ADR 0573"
- "ADR 0548"
- "ADR 0525"
---
# 47, milestone H — the plan to deliver the build-and-delivery system properly
The operator, 2026-11-19 about 23:30 CET: "This mess has been going on for too long and I want to fix it properly.
Write out a plan to deliver our mesh build-and-delivery system. Thoroughly. The research and design 47 should have
all the details of what it should look like and what it should do. Unfortunately, the code is very very poor. So we
need to fix it. What is our best approach? I'm willing to do some hand-deploys to get us into a clean state." And:
"Nobody should 'take in a merge', only commits to the configured branch of registered modules."
This file is that plan. It was written by the lead (lead-47) on 2026-11-20 between 00:00 and 01:00 CET from seven
readings by agents: one specification per part of the design (what design 47, its milestone G, research 086 and their
records say the part does; no code read) and one gap table per part (main's code of 2026-11-19 held line by line
against that specification, with what runs live on each machine). The readings are in the working folder
`.work/plan-47/` of the lead's hq working copy (`spec/`, `gaps/`, `live-state.md`); their findings are summarised
here, and every number below comes from them unless it says otherwise. The plan is the lead's; the operator reads it
before any code is written.
**Milestone H's part.** This file opens the part; its design lines (section 4's items as lines, numbered after G's
§733 and cited as "design 47 H §n") are in three files, each read whole:
- [H-1-the-merge-call-and-the-take-in](H-1-the-merge-call-and-the-take-in.md): the merge call, the gate's source
revision and record, the forge bridge's one message and the watchdog, the take-in, the step `validate`, one pipeline
per take-in and none in parallel, a pipeline's pins, the clean state's seed (§734 to §749, §766 to §770).
- [H-2-wants-builds-reads-and-tools](H-2-wants-builds-reads-and-tools.md): the want rule, the order on one machine,
every account may call every verb, the builder's lines, the reads of other parts made events, rates on the bus, the
missing read tools, which part moves onto the wake-up shape in which release, the four owners' events and a machine
away at the clean state (§750 to §762, §771 to §773).
- [H-3-corrections-and-what-retires](H-3-corrections-and-what-retires.md): the contradictions corrected in place, what
retires, the stale state dropped once, and the tests per module against the budget (§763 to §765, §774).
The wake-up rule itself is in the main file, *Every delivery part wakes on its inputs alone*. Where an H line and a G
line or lead's answer differ, the H line holds, and the G line it corrects in place says so.
## 1. Where the system stands (read 2026-11-20 00:05 to 00:30 CET)
**What works.** One pipeline delivered one module end to end with no hand act on 2026-11-19 at 22:48 (the image
registry, main.1039). The shape of the design is in the code: of 619 table rows of design lines held against main, 413 are built by their contract: 78
per cent of the 527 rows judged (the rest belong to another part's code) (per part: runner 87 of 100 judged, gate and bridge 71 of 93, indexer 70 of
89, assigner and node-engine 56 of 76). No part needs to be written again whole.
**What does not work.** The system cannot deliver itself today:
| symptom (live) | count | cause (from the gap tables) |
|---|---|---|
| wants held or failed on the four machines | 205 held, 66 failed | the catalog half holds no entry for their fingerprints; the node-engines compose nothing new |
| compositions fail on shanks | entries `asked` since 2026-11-18 03:09, unready answers seen from 2026-11-19 11:56 | an entry ask that failed in transport is never asked again; the node-engine's shadow waits on it without bound (the surveyor's reading); other compositions failed on the removed revision trunk.510 and on a call-token refusal |
| delivery records open, none ended | 833 open, 0 ended since 2026-11-14 | the gate hears only `take-in.taken`; nothing writes a record's trace, so its difference is never known |
| pipelines about to start at the next take-in | about 60 (reasoned from the sweep's rule; no tool reads the runner's waiting starts) | the runner's sweep starts a pipeline for every module whose fingerprint has none (b9c50ef0, 2026-11-16); no design line allows it |
| source revision moved twice | both paths on main | A-464 (per merged message) wins live over A-34 (on the take-in): a group's members become separate revisions; a pushed commit reaches the source only through the take-in or the gate's comparison at its next start |
| catalog revision removed while pipelines name it | trunk.510 at 22:51 | pins are judged at restart against a key written once; only the take-in pins pipelines |
| stale bus state | 753 merges, 6,951 owed, 898 asks, 1,256 pulls keys; 5 pipelines running since midday; 31 stale pin records | buckets of the pre-G model never dropped; no tool reads their contents |
| bus load | 98 slow consumers, 117 consumers behind; the watchman 39,037 events behind | mesh-config passes every 5 minutes and polls every minute on main (fixed on the held pull requests); the forge module polls every minute |
| hand-placed programs | every delivery module but the indexer, the image registry and the tool runner (on ace and g14) | G's live start of 2026-11-16 and after; 16 drop-in files under `/run/systemd/system` on the four machines (4 of them `.pre-g` copies), lost at a reboot |
**Every node, every module** (the assigner's assignments at revision 339, unchanged since 2026-11-18 02:08 UTC; the
node-engines' haves; read 00:47 to 00:54 CET):
| machine | placed | running | held | failed |
|---|---|---|---|---|
| the control node | 90 | 10 | 53 (entry missing) | 24 |
| the build node | 80 | 12 | 59 (entry asked) | 7 |
| the desktop | 83 | 24 | 33 (entry missing) | 26 |
| the laptop | 76 | 6 | 60 (45 entry asked, 15 entry missing) | 9 |
65 of the 66 failed wants fail for one reason: the want names no digest for the module's files, code or tools (the
A-430 stand-in, a want naming its parts, left half built). The one other is the desktop's node-engine ("operation not
permitted" unpacking a version). No service is failed on any machine; the 25 open conditions are warnings. Every
delivery module's want is held by the assigner's rule 2 (no passed pipeline at the graph's fingerprint); before it,
rule 8 held the 23 bus-state modules because the bus module's have is held although the bus runs. The haves are wrong
about what runs for three modules (the tool runner on two machines, the indexer): `node-engine.process-builds` is the
tool that says it, found only in this second reading.
**The module graph** (the indexer, trunk revision 512, source 1038): 193 modules; 269 package edges (published before
it builds), 178 toolchain edges, 208 provider and requirement edges, 212 served-verb edges (19 verbs served by no
module), 30 await edges; the acting-part and contribution edges are not in the graph's state. The other watched branches
stand at revision 1 since 2026-11-14. No tool answers a build order or a deploy order for a set of modules, the
difference between two revisions, a revision narrowed by module or edge kind, or which source commit a fingerprint was
built from; the whole revision is 1.3 MB, over the bus's message size. Those reads are tools milestone H4 builds in the
indexer and the assigner (one-question-one-tool; section 4, item 12).
**No pull request is checked today.** The checker's verdict verb asks the indexer `indexer_read_candidates`, which no
running indexer answers (503 no responders), so no `mesh/check` status is ever set; the forge calls every pull request
mergeable because main requires no status. Nine open conditions ask `indexer_read_modules` the same way. Tools were
removed or renamed while their callers still name them; by design 47 G the pipeline makes the check and the checker
goes (G-2), which milestone H1 builds for hq too (hq is a registered module whose one step is its Go checks).
**The root of it, in one sentence.** The design settled on one model (a change is a commit on a watched branch of a
registered module; the forge bridge says it, the indexer takes it in per branch, the runner starts exactly the movers,
the gate follows), but the code carries that model and the earlier per-pull-request model side by side, and the live
patches of 2026-11-16 and 2026-11-17 (the sweep, the source per merged message, the catch-up by comparison) were
written on the earlier one. Every list above is a consequence.
## 2. The approach, decided
1. **One model, no leftovers.** A change is a commit on a watched branch of a registered module (design 47 G §11,
ADR 0666 §6). The forge bridge says every move of a watched branch's head, and at its start reads every watched
branch's head and says it (G §11, A-236). The indexer takes a move in per watched branch (A-476: the commit, the
fingerprints of that repository's modules, a catalog revision; `take-in.taken` names the movers by A-476a). The
runner starts one pipeline per mover and nothing else (G §12, §112: never by a sweep). The gate moves the
source revision on `take-in.taken` alone (A-34, G §592). Everything of the earlier model goes: the gate's and the
bridge's pull-request paths, the source per merged message (A-464 series), the catch-up by comparison (A-463,
A-490k), the runner's sweep, the start-time `outdated` by digest comparison (G §375; a kind-1 dependent is started
by `package-published`, which the runner's durable consumer delivers also after a stop), the forge module's
merge-check machinery and its polls. Principles 1 and 7 settle this; design 10 names a change found by comparison
as a failure mode.
2. **Fix in place, every module; write one part again.** Each gap table judged its modules by the share of lines
built and by whether their events, verbs and kept data are the design's. All say fix in place, mostly by deletion,
but two parts: the assigner's want rule (about 2,600 lines of `wants.go` and the want half of its pass, written again
to A-478) and the indexer's take-in part (about 3,400 of its 34,000 lines: pending, source numbering, behind, the
branch take-in, the pull-request candidates and previews) is written again as one take-in per watched branch fed
by the forge stream alone; its storage, entries, pins, versions and verbs stay.
3. **The design first, then code.** Where the operator's words and the lead's live answers departed from the
written lines and records, the lines and records are brought in line before any code (implement-the-design,
design-it-as-it-should-be): section 4 lists them. No lead answer patches a design question this time.
4. **One clean state by hand, once.** The system cannot deliver the fixes to itself, so milestone H1's programs are
built from main after its merges and placed by hand on the four machines as one mutually consistent set, the
stale bus state is dropped, and from that moment nothing is placed by hand again: every later milestone is
delivered by a pipeline, and that delivery is its proof (principle 5). Section 5 is the step list.
5. **Milestones in dependency order, one release each**, each proven by one pipeline delivering it, with the
guards of section 7 so it does not rot again.
## 3. The six parts: findings and the decision per module
Each part's specification and gap table hold the line-by-line detail; this is the summary a reader needs to judge
the plan.
### 3.1 The forge bridge and the gate (mesh-delivery: the bridge and both seat contracts; mesh-catalog: the gate and the forge module)
Built 71, built otherwise 5, contradicts 10, missing 7, not judged 15 (of 108). No gap is large alone.
- **No delivery record ever ends.** The gate consumes only `take-in.taken`. It must consume the runner's
`pipeline-ended`, the node-engines' `running`, the `wait-held` and `wait-ended` events and the assigner's families
(G §5, §146) to write a record's trace and end it (G-3 "The delivery record").
- **Two paths move the source revision**; the per-merge one (A-464, about 330 lines) acts live and the design's
(A-34, about 75 lines) is left without effect. A-464 goes.
- **The per-pull-request model** is about 1,300 of the gate's 5,475 lines and about 1,050 of the bridge's 4,281.
It goes, apart from what the operator's answer on groups keeps (section 4, question 1).
- **The forge module** holds no failed pipeline step (`failed-steps` has 0 entries, so `merge-pull` refuses
nothing), and still runs the old merge-check machinery: `mesh/merge-gate` statuses, weighing kept in files on disk,
a one-minute poll and a ten-minute timer. All of it goes.
- **Decision:** fix in place, mostly deletion. Gate: consume the two events, end records, one source path, drop the
pull-request model. Bridge: drop the pull-message path, the checker's owed acts and the merge act. Forge module:
drop the merge-check machinery and the polls; find why `seen` is empty (the surveyor's guess: the hand-placed
runner says `step-ended` without commit and fingerprint). Seat contracts: drop `asks` and `turns` and the callers
lists (small).
### 3.2 The indexer and mesh-config (mesh-delivery: the indexer; mesh-catalog: mesh-config)
Built 70, built otherwise 8, contradicts 7, missing 4, not judged 18 (of 107). Seven medium gaps in the indexer are
one problem: the take-in's shape.
- **The take-in is the trunk's alone, fed by two paths**: a `merged` or `pushed` message of any branch lands in the
trunk's one view (a `pushed` message has no base), other watched branches get revisions only by comparing against
the gate's `source-moved`, one `pending` holds every branch, and the catch-up takes heads in outside `pending`.
- **Revisions are made at every keep, not only at a take-in**: "27 changed" counts entries whose stored form differs
from the newest kept revision; two revisions were kept at the 22:51 restart with no `take-in.taken`.
- **Pins** are judged at restart against a `made` key written once in the state's life and against the server's
pending count, which is 0 at start, so the last restatement counts at once and trunk.510 was removed while
pipelines named it. Source numbers live in memory and restart from the gate's number while later pipelines exist.
- **mesh-config** on main passes every 5 minutes, polls every minute and reads its states key by key (issue 1982);
the held pull requests mesh-catalog #708, mesh-contracts #16, mesh-sdk #200 and mesh-delivery #261 make it wake
on events and one deadline timer, reading from watched copies, and remove its pull-request path (ADR 0681).
- **Decision:** write the take-in part again (one take-in per watched branch, from the forge stream alone, movers by
A-476a, source numbers on the bus, pins restated by every asker and judged against what is pinned now, never
against a count); keep storage, entries, pins' state, versions and verbs. mesh-config: merge the four held pull
requests; its reads of the indexer and the assigner through module tools (principle 9) and its full pass per
event are a later milestone's.
### 3.3 The pipeline runner (mesh-delivery: mesh-delivery-pipeline)
Built 87, built otherwise 5, contradicts 2, missing 6, not judged 19 (of 119). No large gap; 4 medium, 9 small.
- **The sweep** (`StartUnstarted`) runs after every take-in and at every start; removing it is small. The held pull
request mesh-delivery #262 adds a third start path (`StartHeld`); it is closed, not merged.
- **The refused step ends naming an empty pipeline** are the assigner's composition runs on the Go build seat: the
runner reads the pipeline field before the asker field. Skip an end whose asker is not the runner (small).
- **Missing, medium each:** `toolchain-agreed` starts no kind-3 rebuild (G §376); nothing consumes `approval-carried`
(G §542); each repository's own checks are not a step (G-2); the start-time `outdated` and `runner_catch_up` (G §375)
do not exist, and by decision 1 they are not built: a kind-1 dependent is started by `package-published`, which the
durable consumer delivers also after a stop (section 4, item 7 says it as a line).
- **A bound on pipelines at once** has no design line; principle 11 makes it a setting (`pipelines-at-once`). A start
over the bound is written with no steps and placed when the runner's own `pipeline-ended` comes, no timer
(medium). What a waiting pipeline reports against G §12's `pipeline-stands` is a design line (section 4).
- **Timers found:** mesh-sdk's view catch-up asks again on a timer (against A-490k's "no timer"); a one-minute
ticker while the step log is full. Both go in the class fix of section 3.6.
- **Decision:** fix in place. The self-hold A-446 removed is already gone from main (commit 070a5dd3).
### 3.4 Builds, the builder and the registries (mesh-delivery: the builder, the artifact holder; mesh-catalog: the build modules, the forge's package registry, the image registry)
Built 73, built otherwise 3, contradicts 8, missing 2, not judged 24 (of 110). 85 per cent of the judged rows
built; no large gap, five medium.
- **A failed key taken again skips a version** (G §369): the reserver treats a key's own standing reservation as its
floor unless the source says rebuild, and only the hand tool `builder_build_again` says so; a failed key taken again
by a look, an event or the start moves its number on by one. This is the 0.1.19 skip. Small.
- **The pre-G numbering still runs beside the reservations** (design 51's versions, replaced): it numbers every trunk
build from its own state and floor, can hold a build waiting for that floor, and its late look writes the live "done
build no version reserved" line. Medium; it goes.
- **The builder is not entered only through the runner's `build` step** (G §473, A-97): it builds every catalog entry
it hears and still follows the gate's `source`. Medium; it goes.
- **`build.ended` carries no version** (A-430b), and builds of other branches reserve nothing.
- **The build modules** pull their three queues in turns, not deliveries first (ADR 0525 §4, G §86; a code comment
cites an operator's yes of 2026-11-15; the surveyor did not search the knowledge base for it, so H2's prepare does: if
the yes was recorded, the record and the lines are out of date, not the code), and when idle re-pull every 5 seconds, a poll. The artifact
store is read by raw HTTP in group runs; the registries are reached through bus verbs. Two copies of the client that
asks `reserved` differ (the forge's and the image registry's).
- **The publish again** runs in one shape, A-457's retried items with A-473's ask at build end as their first try;
§588's event path is gone and its queue remains as plumbing. Live, all 42 publish retries ended after 8 tries,
unreleased, and 37 reservations built since 2026-11-18 were never published (cause a guess, in the table).
- **From the specification:** a module with no registration has no pipeline and is never built, so
mesh-delivery-sdk's missing registration blocks mesh-delivery-contracts by design; a `required` step ending
`nothing to run` holds the delivery and no line works it out again later; no line has the builder reserve an old done
build by itself. Each is a design line to write (section 4, item 11).
- **Decision:** fix in place, all three. Builder: remove the numbering, the gate-source follow, the builds without a
step ask and the other pre-G mechanisms the table lists; fix the reserver's floor and the version on `build.ended`.
Build modules: deliveries first and no idle poll in the shared step runner; the store read over the bus. Registries:
one `reserved` client, the forge's form moved into `mesh-delivery-contracts` as the surveyor recommends. The delivery
budget (G-2) is checked by nothing; medium, H2.
### 3.5 The assigner, the node-engine and the tool runner (mesh-delivery: the assigner; `mesh-host`: the node-engine; `mesh-tools`: the tool runner)
Built 56, built otherwise 8, contradicts 9, missing 3, not judged 7 (of 83). Assigner 31 of 44 built, node-engine
17 of 27, tool runner 8 of 12.
- **The want rule A-478 is not built** (the newest released version whose pipeline passed, on every machine that
places the module; the operator, 2026-11-17). Every old rule still holds wants: canary, grant, provisions, served
verbs, bus state, the acting-part waits (about 450 lines), the `version-delivered` holds (about 500), the parts
check (about 250), the rest about 300. All of it goes; the backup, the yes on a failed backup or a held revert, and
A-478b's order waits on one machine stay (the order waits reshaped to order one machine's apply, not hold wants).
- **The assigner reads every have** whole and with history (§520 says it goes); wants are not written from the kept assignment graph revision. Medium each, H1. (The node-engine's read of
every node's record stays: it feeds the composer; H-A-60 corrects the earlier reading of §459.)
- **The shanks shadow fault**: the entries client leaves an entry `asked` after a transport failure and never asks
again; about 110 entries on shanks have stood `asked` since 2026-11-18 03:09. Ask again on failure, with a bound
(small). The surveyor's reading, to be confirmed by `entry-again` on one entry during the clean state.
- **The tool runner:** A-490m is ready on mesh-tools #77 (merge); it lacks a test that starts it whole.
- **Tests:** the node-engine has 1,719 unit tests against a budget of 500; every change to it shrinks that.
- **Decision:** fix in place, all three. The want rule is rewritten to A-478 in `wants.go` and the want half of the
pass; the old holds are deleted.
### 3.6 The status view, the watchman, health, the tools and the grants part (mesh-delivery: the view, the watchman, the grants part; mesh-catalog: the bus's module; mesh-tools; mesh-sdk)
Built 56, built otherwise 11, contradicts 5, missing 11, not judged 9 (of 92).
- **The status view, the watchman and the grants part read other modules' buckets directly**, against ADR 0596: the
view watches twelve such buckets, retries each every 5 seconds, and still follows the teller's answers and the gate's
source, which §512 retires. Large, with the want rule, the take-in and the SDK's generic state tools (the fill against
a cap in every module's tools, the grants part's `filling` and `room`) the large gaps of the survey.
- **The kind of fault behind issue 1982 is in more modules than mesh-config.** The grants part recomposes everything
and reads its own states key by key on every change; the indexer re-reads every module, every graph key and every
manifest after each verdict, import or source move (its modules state stands at revision 1,789,962); the view
retries twelve watches every 5 seconds; the forge module polls every minute; the build modules re-pull every 5
seconds when idle; mesh-sdk's view catch-up asks again on a timer. The gap table's section 5 has the table of
modules, functions and intervals. The fix belongs in mesh-sdk, for every module at once (fix-a-bug-met-twice).
- **`builder_stuck` times out** for a fixed cost, measured in the tool runner's journal on shanks: 1,010 values took
52.5 seconds and 53 values took 44.5 seconds. Each call starts a tool process that lists every key of `builds` and of
every build seat's `steps`, then reads one value per request. The builder keeps its own index of what is building
and answers from it. Medium.
- **The runner's live seat serves five read verbs and no act**; main has them all (commit 8fa1b8ae). The surveyors'
guess is a hand-placed build from before it, or an older tools bundle; the clean state ends either.
- **Who may call what** stands in two forms: §610's class defaults and A-434's stand-in that every account may do
everything. **The lead's answer: principle 6 settles it** (inside the mesh every module is trusted, who may do what
is not a question): A-434 stands, the callers lists go from every seat contract, and tests 40, 51 and 85 are not
checked while it holds.
- **No line names rates on the bus**; the bus's module gives connections and consumers, not rates. A tool to build in
the bus's module (principle 4), named in section 4 so the design says it.
- **Decision:** fix in place, every module. View: replace the bucket watches with events (the large gap, H4).
Watchman (16 of 23 built) and condition keeper (14 of 18): fix in place; their live builds predate milestone G. Grants
part: the re-read through the SDK shape in H4, its move from reads to events a release of its own (H4b). Bus module and tool runner: the
rate read and the SDK's generic state tools.
- **The wake-up rule is a class fix:** one shared shape in mesh-sdk (watched copies of a module's own states, a pass
only on a changed input, one deadline timer for its real deadlines, no poll), every delivery module moved onto it in
H4, and the review refuses a timer, a poll or a full re-read in any delivery module from H1 on.
## 4. What the design must say before code (milestone H0)
The design's lines and the records carry the operator's later words only in the lead's answers, and in places not
at all. Before any code, `hq-design-writer` writes these as design lines in this milestone's part and
`hq-record-writer` proposes the records, in one batch read by the four reviewers by architectural domain (playbook
02). The lead decides none of them.
**Questions for the operator** (each changes how the operator works or removes data; the lead's recommendation first):
1. **How a change across repositories lands together.** The design forms a group from pull requests' head branch
names (G §42, §716); the operator's rule is that the mesh knows only commits on watched branches. Recommendation:
a group is a trailer in each member commit's message (`Mesh-Group: <name>`, the name chosen by the author); the
forge bridge reads it from the commit and the gate forms the group from it; nothing is read from a pull request.
The alternative is no groups at all, with the order between modules left to the artifact dependency edges, which
does not cover a seat contract change that several modules must follow at once.
2. **Data the clean state removes:** the gate's `merges`, `owed`, `asks` and `pulls` buckets, its 833 open records,
the indexer's stale pins and the 5 pipelines running since 2026-11-19 midday, every held or failed want, and the
19 drop-in files. Recommendation: remove all of it; none is a record anyone reads (no tool reads what the buckets
hold), and the delivery records of the one model start from the clean state.
3. **Pipelines one at a time** until the operator raises it: the setting `pipelines-at-once` with default 1.
Recommendation: yes for the clean state and the first milestones; the operator raises it when the chain has proven
itself on about ten deliveries.
4. **The one hand placement** of milestone H1 at the operator's terminal, one command at a time as before, after
which no pack is run again except the ones this plan names. Recommendation: yes; it is what the operator offered.
**The operator's answers (2026-11-20 about 01:10 CET).** Question 2: "I don't care about old data at this moment, I want
a working mesh again": the stale state goes. Question 3: "for my part, pipelines don't exist, or a group of commits
form one big pipeline, but parallel pipelines sounds like a very bad idea": no two pipelines run at once, and one
take-in is one pipeline (design 47 G §44 already says it for a group: one source revision, one pipeline per watched
branch; the single commit is the case of one member); the setting of item 7 keeps the bound at one by its default and
the design says serial. Question 4: "I don't mind doing hand placements": yes. Question 1: the operator's first idea was a
build plan and a deploy plan per change (ADR 0238 §2, "one commit, one change plan", the glossary's *delivery plan*,
deleted at release 6 of ADR 0429 and replaced by ADR 0666's pipelines per module ordered by edges, from research 086);
what remained open was how the mesh learns which commits belong together, since ADR 0666 §13 reads a shared branch
name through pull requests. The operator, 01:15 CET: since every version is deployed by a merge into main, an action,
they have no objection to kicking the build-and-delivery process off by hand instead of listening to the forge's
hooks; the lead named principle 3 (a start a person forgets is a silent non-delivery) and proposed the merge through
the mesh as the start with the forge's event kept as the watchdog; the operator chose it (01:20 CET, "A, go
ahead"): one call of the seat `git`'s verb `merge-pull` naming one or more pull requests is the delivery's start and
the group; nothing reads a pull request after it; a head moved outside the mesh raises a condition and delivers
nothing. Item 1 below is answered so; item 6 is written from it. The operator also said they would not mind a start from scratch; the lead's answer is in
section 2 (fix in place, two parts written again) with the reason: 78 per cent of the lines are built by their contract,
and the parts that are the earlier model are named and bounded.
**Design lines and records to write** (no operator decision; `hq-design-writer` and `hq-record-writer`):
5. The take-in by A-476, as a record: it changes ADR 0666 §6 and ends ADR 0633 §1's hold for manifest validation
and the "pending revision"; the manifest validation becomes a pipeline step with a name, a server and a policy
default (A-476 names none), and `manifest-refused` and the cycle refusal move with it.
6. The forge bridge's one message for a watched branch's head moving, replacing the `merged`/`pushed` pair where
they differ (a `pushed` message carries no base today), and the end of A-464 to A-464e, A-463 and A-490k.
7. One take-in, one pipeline, and no two at once: the pipeline is per source revision of a watched branch, its steps per
mover in the graph's build order, delivered in the deploy order (the operator, 2026-11-20); the setting
`pipelines-at-once` with default 1 (principle 11), the order waiting take-ins are placed in, and that a waiting
pipeline is not owed progress under G §12. A record, since ADR 0666 §1 and §7 speak per module.
8. The want rule A-478 and A-478b as design lines, with the list of holds they end (§266, §268, §269, §376, A-189,
§524) and the reshaped order wait on one machine.
9. The wake-up rule for every delivery part (section 3.6) as a design line in design 47's main file, with how it is
checked (the review and the integration test that starts the module whole).
10. The small contradictions the specifications list, corrected in place: G §5's unprotected-branch refusal (removed
by the operator, still in ADR 0666 §3), G §654 (replaced by ADR 0681 §1), A-449's `pinned`/`freed` publisher,
test 53's `forge-made`, G §476's "one entry per module of that revision", ADR 0666 §12's three open asks (A-424),
§386's per-module durable consumer (A-397).
11. The builder's lines the specification found missing: a done build with no reservation is reserved by the builder
itself (not only by a hand tool); a failed fingerprint built again keeps its reservation; a `required` step with
nothing to run is worked out again at every pipeline, not only at the seed; a registered module whose pipeline has
no steps is refused at registration, naming the step (test 67); and the record of the build modules' queue order
(deliveries first, ADR 0525, or the turns the code takes on an operator's yes no record holds).
12. The reads the status view, the watchman and the grants part make of other parts' buckets, each turned into the
event or verb of the part that owns the data (ADR 0596), and the rate read on the seat `mesh-broker`.
13. The read tools the second live reading found missing (one-question-one-tool): in the indexer, one revision
narrowed by module or edge kind, a non-trunk branch's revision, the edges by kind, the difference between two
revisions, a build order for a set of modules, the source commit a fingerprint was built from; in the assigner, a
deploy order per machine; in the node-engine, its own build. Also the checker's and the conditions' calls of
indexer tools no indexer serves (`indexer_read_candidates`, `indexer_read_modules`), removed with their callers.
## 5. The clean state, by hand once (milestone H1's delivery)
Milestone H1 is "the one chain": the smallest set of fixes that makes a commit on a watched branch reach a running
module through the design's path with no hand act, in every module of the chain. It is built, reviewed and merged
like any milestone, but its delivery is by hand, because the running system cannot carry it. The steps, in order;
the operator runs each pack one command at a time and says "done"; the lead checks each with the tools before the
next:
1. **Freeze.** No merge to any delivery repository but H1's own. The runner on novox is stopped through the service
seat so no take-in starts a pipeline during the freeze; the assigner stays, so nothing moves.
2. **Merge H1's set to main**, in this order, each reviewed: mesh-contracts #16 and mesh-sdk #200 (held), then
mesh-catalog #708 and mesh-delivery #261 (held, mesh-config), mesh-tools #77 (held), and H1's own pull requests
(indexer take-in, runner, gate, bridge, assigner, node-engine). mesh-delivery #262 is closed as superseded. The
stopped runner starts nothing; the indexer takes the commits in and the gate follows, as the old programs still
running during the freeze do. The new programs take every head in once, at the seed (step 5.9).
3. **Build H1's programs from main** with the packs (the g-start phases, one per program), on shanks and novox as
before, from main's head after step 2 so that main is what runs.
4. **Take the stream snapshot that is the drop's rollback** (design 47 H §765) and test its restore; nothing is
dropped yet. The drop-ins are removed and the units the node-engine writes take their place as the wants compose.
5. **Place the programs** in the order the graph gives (derived by the reader of 00:54 CET from the package,
toolchain and provision edges; no tool answers it yet), each step read with the tools before the next:
1. keep the bus, the forge and the image registry running; make the bus module's have `running` (else the
assigner's rule 8 holds the 23 bus-state modules);
2. the packages first: `mesh-sdk`, `mesh-contracts`, `mesh-delivery-sdk` published (every Go module needs them;
`mesh-delivery-sdk` needs its registration first, section 3.4);
3. build with the running Go build module (it is what breaks the cycle "every Go module, the build module itself
included, is built by a running build module"), the builder first, then the indexer, the forge bridge, the
runner, the assigner; then the three build modules (they need the builder's package), the gate (needs the
indexer's and the bridge's packages) and the grants part; then the node-engine;
4. place on the control node: the indexer, the forge bridge, the gate, the forge module, mesh-config; on the build
node the builder; on the desktop the assigner;
5. place the node-engines: the control node, then the build node, the desktop, the laptop;
6. place the watchman, the status view, the teller, the grants part and the hand-act log on the control node;
the checker is not placed (it goes, section 1);
7. after each machine, read its health, its first views, its `process-builds` against its haves, and the bus's
consumers;
8. **drop the stale state** (question 2's yes), now that the new programs run: every state design 47 H §764
retires, the records, the pins, the pipelines of the earlier model, the stopped runner's queue of old
`take-in.taken`, the held and failed wants, each through the tool H §765 names, the pack saying what it
deleted, counted;
9. **seed every watched branch's first source revision** with `gate_source_seed` (H §768): each branch's current
heads, taken in with no mover, so the newest catalog revision names them; read and counted, and a bridge
restart then raises no `head-untaken`;
10. a machine away at this moment gets its hand placement when it is back, before anything else (H §772).
6. **Start the runner last**, with `pipelines-at-once` at 1.
7. **Prove it.** One single-module change (a comment in a module with lint, unit and composition steps only), merged
to main through a merge call (`merge-pull`, design 47 H §734), followed from `take-in.taken` through `pipeline-ended` to the node-engine's `running` and the gate's
ended record, with no hand act; then one two-module change along a dependency edge (mesh-sdk and a module that
requires it published), merged in one merge call, the same way. Each proof read with the tools and quoted with its times.
8. **Lift the freeze.** From here every merge is delivered by its pipeline; a hand pack is a breach the lead names.
## 6. The milestones after the clean state
Each is one release of playbook 07 (prepare by `mesh-preparer`, develop, one review, one check, delivered by its
own pipeline), in this order, the next starting when the previous one's delivery is read on the machines:
| # | outcome | repositories | proof |
|---|---|---|---|
| H0 | the design says the one model whole (section 4) | hq, the knowledge base | the four reviewers by architectural domain approve the batch; the operator's four answers are in |
| H1 | the one chain works with no hand act (section 5) | `mesh-delivery`, `mesh-catalog`, `mesh-host`, `mesh-tools`, `mesh-sdk`, `mesh-contracts` | the two proofs of step 7 |
| H2 | the builder, the build modules and the registries follow the latest lines (section 3.4; versions, publish again, the floor, a step with nothing to run, a module with no registration) | mesh-delivery, mesh-catalog | a version is skipped by nothing; every registered module has a pipeline with steps |
| H3 | the gate and the forge module carry only the one model; groups by the operator's answer; the forge module without its merge-check machinery | mesh-catalog, mesh-delivery | a group of two repositories delivers as one; every delivery record ends |
| H4 | every delivery part wakes on its inputs alone and is steered by its tools (sections 3.6, 7): the shared wake-up shape (mesh-config's full pass per event and its reads through module tools, the indexer's re-read after every verdict, the node-engine's retry clocks, the build modules' idle pull), the view and the watchman without reads of another part's state, the SDK's generic state tools with the fill against a cap, the missing tools per module, rates on the bus | `mesh-sdk`, `mesh-delivery`, `mesh-catalog`, `mesh-host` | the bus's slow consumers and behind counts at 0 for an hour; every module's tools pass the principle-4 table |
| H4b | the grants part says what it holds as events and reads nothing of another part's bus state (ADR 0596) | `mesh-delivery` | the grants part's own `filling` and `room` and no bucket watch in its code |
| H5 | tests to budget and one integration test per module that starts it whole (ADR 0548); the catalog check that refuses a module whose state has no read tool | every delivery repository, mesh-catalog's check | the check reports each module's counts within budget |
Milestone H1 is large for one release (six repositories); it is not cut (how-work-is-done), its subtasks are split
at the prepare, one developer per repository on the larger model for the indexer and the gate (their parts meet at
`take-in.taken` and the source revision), the smaller model for the rest.
## 7. The guards, so it does not rot again
- **An integration test per module that starts it whole** against a real bus and store (ADR 0548), run in the one
check per release, and the change to how a module runs carries it in the same commit.
- **The review reads against the one model:** a sweep, a poll, a timer for another part, a comparison in place of an
event, a read of another part's state or a per-pull-request path is a finding that fails the review.
- **The tools table per module** (principle 4) in every pull request of a delivery module, and the catalog check of
H5 that refuses a module without its read tool.
- **No hand pack after the clean state.** A delivery that fails is fixed in its repository and delivered by its
pipeline; a pack is a breach the lead names, with the one exception the operator grants in words.
- **The test budget** per module (50 integration, 500 unit) reported by the check; a module over it shrinks with
every change.
- **Live answers become design lines the same day** (implement-the-design): a lead answer that patches a model
question is a breach the reviewer reports.
## 8. What the plan does not do
- It does not merge the six held pull requests now; they merge in H1's step 2, and one (#262) is closed.
- It does not touch the controller, the vault or the recovery module beyond what the chain needs; the recovery
module is not placed and is a later milestone's.
- It does not change the catalog's module set or the build nodes (novox's build agent stays paused).
- It leaves what the surveyors could not read as named guesses, each with how to check it, in the gap tables'
last sections; the prepare of each milestone checks them.
## 9. H1's develop: who builds what, and why (the lead, 2026-11-20 02:27 CET)
Six `mesh-developer` agents, one branch each, from the six build maps of the prepare (the lead's working folder,
`prepare/H1-*.md`) and the lead's answers H-A-21 to H-A-62; each commits locally and cites its lines; none pushes before
the one review (`mesh-code-reviewer`, at most three rounds).
| repository | branch | scope | model and reason |
|---|---|---|---|
| `mesh-delivery` | feat/47-H1-contracts-indexer-bridge | the seat contracts (C0 to C3, C-R, C-W), the indexer's take-in written again, the forge bridge | larger: its parts meet at `source-moved`, `take-in.taken` and the contracts every other branch starts from |
| `mesh-delivery` | feat/47-H1-runner-assigner | the runner's queue and events, the assigner's want rule and tiers | larger: it meets the indexer at the pins and `take-in.taken`, and the node-engine at the want's form |
| `mesh-catalog` | feat/47-H1-gate-forge | the gate's merge-call path and record, the forge module's merge call and `merge-called`, #708 onto main | larger: the gate and the forge module meet at `merge-called` and the bridge at `head-moved` |
| `mesh-host` | feat/47-H1-node-engine | the entries asked again, the apply by rank and `apply-order`, the want by version | smaller: a leaf, its contract given |
| `mesh-tools` | feat/47-H1-tool-runner (from #77) | the per-verb `timeout`, no `admitted` hook, the whole test | smaller: a leaf |
| `mesh-sdk` and `mesh-contracts` | feat/47-H1-seats (from #200 and #16) | the copies, `timeout` in `seatdef`, no `Callers`, `placement`'s `build`, `apply-order`, `held-back` gone | smaller: contracts and copies, published first |
The order between branches is the prepare's dependency table: contracts and copies first (published as the next
mesh-sdk and mesh-contracts versions at the clean state's step 5.2), then the modules; the runner-and-assigner branch
rebases onto the contracts commits before the review.
@@ -0,0 +1,570 @@
---
layer: to-be
status: in-progress
design: 47
milestone: H
updated: 2026-11-20
decisions:
- "ADR 0682"
---
# 47, milestone H — the lead's answers
The lead's one-line answers to the questions the design writer and the builders could not answer from the records,
each tested against the principles (`00-META/principles.md`) and naming the one that settles it (implement-the-design).
A question no principle settles goes to the operator or to a record; it is marked so here.
## From the design writer (2026-11-20 01:36 CET)
- **H-A-1 A member refused after every member was judged** (a forge failure between the judgement and the last merge):
the forge module says the members that landed, marks the call `refused` and raises the finding `merge-partial`;
the landed members are taken in as one source revision, so what landed is delivered and what did not is loud.
Principle 3.
- **H-A-2 The delivery order once the want holds for nothing**: the runner's `pipeline-ended` carries its movers in
ADR 0666 §14's delivery order, and the node-engine applies one pipeline's wants on its machine in that order
(A-478b's order on one machine). The order travels in the event; no part reads another part's state for it.
Principles 1 and 9.
- **H-A-3 A `required` step whose run ends "nothing to run"** counts as passed for the want, as `bypass` does: a hold
with nothing behind it protects nothing. Principle 7. A clause of ADR 0682 says it (asked of the record writer, 01:38).
- **H-A-4 Reads whose owner says no event today** (the provisioner's provisions, the vault's bus accounts and node
logins, the enrolment module's enrolments, the grants part's user list): each owner says one change event and the
readers consume it, in milestone H4b; until then the read stands and is named a leftover in the gap table.
Principle 1 (ADR 0596).
- **H-A-5 A revert against the whole want rule**: a return to an earlier build stays one value in the placement, read
by the assigner before the rule, the one exception; ADR 0682's review raised it and the record says it. Principle 4.
- **H-A-6 The seat that serves `validate`**: the seat that runs the composition step, which G §22 names `mesh-wants`;
ADR 0682 §4's "mesh-module-assigner" is corrected to that seat (asked of the record writer, 01:38). Principle 9.
- **H-A-7 The main file's size** (258 KB against the 80 KB bound of playbook 02): the reshaping of design 47's main
file into its milestones' parts is a change of its own after H0, not folded into this one (rules-are-weighed: the
bound protects a reader, and the plan's reader has the three H part files; a 180 KB move in the same pull request
would hide the lines under review). Said to the operator at 01:38.
- **H-A-8 The build modules' queue order**: deliveries first (ADR 0525 §4), as §753 writes; the operator's yes to the
turns of 2026-11-15 is in no record, so the record holds. If H2's prepare finds it, the record gets an insight.
## From the review by architectural domain (2026-11-20 01:40 to 01:44 CET; four reviewers, one round)
- **H-A-9 How the apply order reaches the node-engine**: the want carries its `pipeline` and its `rank` in that
pipeline's delivery order, written by the assigner from `pipeline-ended`; the node-engine applies a pipeline's wants
on its machine by rank and consumes no event of the runner. The apply-order read tool is the node-engine's.
Principles 9 and 12.
- **H-A-10 Order across machines** (ADR 0666 §14's "delivered before" edge between machines): the order lives inside the
one pipeline, never as a hold of a want on another delivery (A-478 b: "the order between modules moves into the
pipeline"): the assigner writes one pipeline's wants tier by tier in §14's order, a later tier once the earlier
movers report `running` on their machines, with a bound per tier (a setting) whose end ends the delivery record `failed` with the finding `tier-late`,
the pipeline's own outcome staying what its steps said; the
composition step composed the whole graph before any want was written. ADR 0682 says it (the record writer, 01:50).
Principles 3 and 7.
- **H-A-11 Publishing an event family stays the producer's**: every account may call every verb (ADR 0682), but a
module publishes only the events its manifest declares, as ADR 0666 §21 has it for a step's result; the mechanism
stays whole, who is on which side is not asked (who-may-is-not-the-operators-question). An enrolling node's account
keeps its own grant. Principle 6.
- **H-A-12 The indexer's consumer and a blocked `source-moved`**: one durable consumer per watched branch; a
`source-moved` the indexer cannot take in (the store full) is held, unacknowledged, until the bus module's `room`
wakes it, never on the ack timer; a hand tool skips it with `why`, written in the hand-act log. `take-in.pending`,
`take-in-waits` and `indexer_pending` stay (ADR 0633 §5). Principles 1, 3 and 4.
- **H-A-13 A source revision written once per call**: the gate keys its write by the call id with compare-and-set, so a
redelivered `merge-called` numbers nothing twice. Principle 3.
- **H-A-14 The waiting queue of the runner**: ordered by the runner's own counter at the moment it writes the pipeline;
placed at the runner's start from its own state and on every end, a cancel and a set-aside included; a finding when
pipelines wait while none runs; a pipeline waiting on the operator's `approval` step does not count against
`pipelines-at-once` and re-enters the queue first when approved; the queue's cap is a setting. Principles 3 and 11.
- **H-A-15 The clean state's seams** (plan section 5): before the runner starts, every watched branch's current head is
its first source revision with no movers (§745) and the newest catalog revision names it, read and counted; the state
of the earlier model is dropped after the new programs run (after step 5.6), each item through the tool named for it,
the backup snapshot being the rollback; every new durable consumer starts at the newest message and refuses an event
of the old form with a journal line; the stopped runner's queue of old `take-in.taken` goes with the state; a machine
away at the clean state gets its hand placement when it is back, before anything else; the forge's branch protections
drop the required statuses `mesh/merge-gate` and `mesh/repo-check` in H1. Principles 2 and 3.
- **H-A-16 Where H1's integration tests run before its merge**: once, locally against an embedded bus, said in each
pull request (how-work-is-done allows it); the chain's live proof (plan step 7) is H1's check; from H1 on the
pipeline makes the check. Rules-are-weighed, said.
- **H-A-17 The whole tests**: one per module, in that module's repository; tests 110, 112 and 113 are split per module
and the chain stays the live proof; the design gives the count per module per release and names the retired tests
that free room (the budget is per module, 50 and 500).
- **H-A-18 The forge module's call**: a call key written before the first merge and said again at start; `refused` in
the event's body; `merge-partial` keyed by the call, cleared when a later call lands the refused members or by a
hand clear; tools `forge_calls` and a say-again; `by` recorded beside the caller's bus account; `failed-steps` is
written from the runner's `step-ended` (mover, commit, fingerprint). Principles 3 and 4.
- **H-A-19 Small forms**: no pin for a take-in with no mover; a call with two members on one branch names the last
member's commit in call order; `package-published` carries the source revision that published and its pipeline pins
it and belongs to that merge call's delivery record; `pipeline-ended` carries per mover the released version and
fingerprint and the delivery order, on the subject `...pipeline-ended.<branch>.<source revision>`; `pipeline-failed`
is one finding per failing mover, cleared by that module's later passed pipeline; every fixed value (30 days, 64 MiB,
60 minutes, 10 minutes) is a setting with that default; §765 drops every state §764 retires. Principles 7 and 11.
- **H-A-20 The main file's size**, a must for three reviewers: the lead's bending of the bound (H-A-7) is said to the
operator at 01:48 with the question whether the reshape follows H0 as its own change; until the answer the bound
stands bent, said.
## From H1's prepare (2026-11-20 01:55 to 02:20 CET; six build maps, 92 questions merged in the lead's working folder)
Each answer names the item of the merged list (Q1 to Q92), then the principle. "Both" means the preparers agreed.
Where two preparers disagreed, the chosen one is named.
- **H-A-21 (Q1)** `merge-pull` answers once, at the call's end, naming `call` and each member's outcome. P1.
- **H-A-22 (Q2)** Its caller `id` goes (the call names it); an already-merged member is refused `already-merged` with
its commit; the answer is `{repository, number, state, commit|why}` per member; `protected` goes; `calls` is declared
in the seat definition. P3, P7.
- **H-A-23 (Q3)** A kept stream's age and cap are its definition's `keep` (30 days, 64 MiB as defaults there); the
word "setting" is dropped from those lines. H-A-19 is read so: a value stated once outside code. P11.
- **H-A-24 (Q4)** A call is 16 lowercase hex characters, one subject token; `hand`, `seed`, `by` and the time live in the
record's trace (sdk). P8.
- **H-A-25 (Q5)** `record-opened` and `record-ended` carry `branch`, `source`, `call`; `milestone` stays as
`<branch>.<source>` for the teller and the watchman (cat); the verb `source` answers `call`. P7.
- **H-A-26 (Q6)** `source-moved`'s wildcard has two tokens; its body as §737 with `v`; every decoding reader is named
with its change in its repository's map. P7.
- **H-A-27 (Q7)** `head-moved` on `forge.<owner>.<name>.<escaped branch>.head-moved` under the module stream prefix; its
form in `forgemsg`, series 0.1. P8.
- **H-A-28 (Q8)** Everything of the earlier model leaves in H1: the gate's states and `approval-carried`, `groups` and the
group model, the forge module's merge-check machinery, the knowledge base's `groups` invoke (cat); `heads` is declared
in the seat definition. One model, no second one left running. P7; ADR 0682 §2.
- **H-A-29 (Q9)** §738's delivery record is built in H1: the gate consumes the new `take-in.taken` for entries and
outcome and reads a record by its key `<branch>.<source>`, never every record. P3.
- **H-A-30 (Q10)** The grants part makes one durable consumer per registered watched branch from the registrations
(`MaxAckPending` 1), as it makes every consumer today; mesh-sdk binds a consumer by the name its caller gives;
`source-moved` gets a kept stream (sdk, with ib's stream). P7 over P12 here: the maker of consumers exists, and a
second maker would be a second mechanism.
- **H-A-31 (Q11)** `take-in.taken` and `take-in.pending` carry only the lines' fields, plus `by` and `why` for a hand act;
readers pair by `<branch>.<source>`. P7.
- **H-A-32 (Q12)** The catalog revision's number is the source's; its name `<escaped branch>.<source>` for every branch,
the trunk included; written once by read-before-write under the branch's one-at-a-time consumer. P3.
- **H-A-33 (Q13)** The source value carries `source` and `call`; the next number is the branch's newest plus one; a
`merge-called` whose call the newest already carries writes nothing. P3 (with H-A-13).
- **H-A-34 (Q14)** The assigner builds the validator from the change's heads on `node-build-go` as it builds the
composer, its input on `mesh-wants`' `input`; `RefusedCycle` moves to `mesh-delivery-sdk/graph` first; the program's
source stays where it is until a later cleanup (ra). P9; ADR 0682 §5.
- **H-A-35 (Q15)** A pipeline's kept value holds `movers` as `{module, commit, fingerprint, repository, tier}`; each step
result names its module; `list` filters by a mover; the runner's contract lands first. P7, P10.
- **H-A-36 (Q16)** A runner state `queue` holds `queue.next` and one key per waiting pipeline; the pipeline's outcome
value is `waiting`; an over-cap start is acknowledged and kept there, never left unacknowledged. P1, P3.
- **H-A-37 (Q17)** A want keeps `fp`, `hold` and `words`; `parts`, `rule`, `newest`, `kept-on` and `composed` go; `rank`
is an integer from 1, absent for a placement's `build` and for the node's own want (host's set, ra's drops). P7;
ADR 0682 §8 ("never digests" reads `fp` as the build's name).
- **H-A-38 (Q18)** A candidate's `added` names a version alone; the shadow resolves it as it resolved parts; `parts`
leaves the assigner seat's contract. P7.
- **H-A-39 (Q19)** `ModulePlacement` gains `build` (one row per branch and build), a fingerprint; the assigner wants the
version released at it. P7 (with H-A-5).
- **H-A-40 (Q20)** The next tier is released on `running` matched by the mover's fingerprint from `pipeline-ended`;
`running` gains no version in H1 (ra). P7.
- **H-A-41 (Q21)** The callers type and functions, every holder's list, the step runner's asker judgement and the tool
runner's `admitted` hook go in H1, in one group; the vault's enrolment caller stays; a step run carries `asker`. P6;
ADR 0682 §10.
- **H-A-42 (Q22)** H1 carries a mesh-sdk change beside #200: the new byte-equal copies of the seat definitions and the
checker's copy dropped, published first as the next version; §762's "mesh-sdk in H4" is read as the wake shape alone
and corrected in place in this branch. P5.
- **H-A-43 (Q23, Q24)** The per-verb `timeout` is built in H1 (a merge call of several members is cut at a fixed 30 s
otherwise): the field in the seat definition and its schema, whole seconds 1 to 600, others refused; `merge-pull`
names 120; the tool runner's own default is the setting `call-timeout`, 30 s; every caller and the serving runner wait
it; module processes through mesh-sdk. §760's "H4" is corrected in place. P3, P11.
- **H-A-44 (Q25)** Writers the maps did not name: the watchman's and the view's take-in readers, `sourcebranches` and
`meshcatalogclient` are the indexer-and-bridge developer's; the teller, the knowledge base's and the issue tracker's
readers are the gate-and-forge developer's; mesh-sdk's copies are the sdk developer's; the knowledge base reads
`pipeline-ended` in place of `group-ended`. The lead's order of work.
- **H-A-45 (Q26)** The word is `want:version`, a constant of the node-engine seat contract, stated in `manifest-reads`
and read from `manifest-reads-changed` alone; the assigner holds any node not stating it (host). P8, P1.
- **H-A-46 (Q27)** The apply order is a read verb `apply-order` on the seat `node-engine` (contract and sdk copy first),
one row per unapplied changed want: `module, pipeline, rank, position, why, waits`; H-A-9's "read tool" is read so.
P4, P9.
- **H-A-47 (Q28 to Q36)** A member whose base is no watched branch is refused at judging, naming the base (P3). The
call's value gains `said` (P3). A call past retention is deleted on each new call's write, oldest first (P1). A merge
call with no caller account the tool runner proved is refused, saying why (P3). FORGE-1 builds `step-unread`; the
clean state reads `failed-steps` after the first failed step; a one-event read tool is H4's (P3, P4). A session reads
its record by `call` through the gate's verbs after the answer, then waits on `pipeline-ended`, re-reading (P1).
"The tool runner's `merge-pull` tool" reads as the seat `git`'s definition and its sdk copy (P8). The seed opens no
record (P7). `gate_source_start` and `gate_source_again` stay, the first writing through the seed's path (P4).
- **H-A-48 (Q37 to Q43)** Every indexer and bridge part no line names (`Src` stamps, the builder-log index,
`TakeInMerge`, the own-build hold, previews, `taken`, the `pull` and `open` messages, the bridge's finding writer) goes
in H1 and is added to §764's lists (P7). `pins-stated`, `heads-replaced`, `pinned`, `freed` and the gate's pin
operations go; the indexer pins in its own state (P7). The gate's stalled conditions go, the watchman's `record-open`
being the one signal; `untaken-bound` runs on the gate's one timer; the start loops are waits for the bus, said (P7;
ADR 0682 §11). `transition` leaves `emits` (P7). A record's member is `<repository>@<number>` for a call and
`<repository>@<commit 12>` for a hand take (P8). The bridge's `heads` gets a read tool, no say-again (P4).
`git.branch` becomes a kept call that waits on the holder's return, no redelivery timer (P1).
- **H-A-49 (Q44, Q45)** Every reader of `manifest-accepted` and `manifest-refused` (the builder's floor, the watchman,
mesh-config on #708, the tool runner's ledgers, membership) reads the mover's entry on `take-in.taken`; #708 moves to
`catalog-revision` before it merges. Only a take-in keeps a catalog revision; a toolchain change reaches pipelines
through `toolchain-agreed`. P1, P9; ADR 0682 §4.
- **H-A-50 (Q46)** The event that wakes a held take-in is the indexer's own `room` on the seat `mesh-catalog` (the store
is the indexer's); H-A-12 is read so. P12.
- **H-A-51 (Q47 to Q51)** One `validate` run per take-in revision, every mover's step reading it; the runner's and
`mesh-wants`' contracts add `validate` (P7). Every step of a mover waits for its dependency's publish (P3; ADR 0682
§6). Tiers come from edge kinds 2, 4, 5 and 6; a cycle among movers is one tier, said in the journal (P3; ADR 0666
§14). `widens` is per mover in its entry (P10). Lines already built (`branch` and `source` on `package-published`,
`pipeline` on step events) are read as built (P7).
- **H-A-52 (Q52 to Q59)** The runner consumes `version-released` in H1 and writes the version into the mover's entry
(P1). A run on request stays per member, in the same queue (ADR 0682 §7). A hand `start` keys the newest revision's
`<branch>.<source>` with that module as its one mover, refused when the key exists (P4). The runner asks `outdated`
no more; kind-1 dependents come from the edges it holds (P1; ADR 0682 §6). A waiting pipeline says
`pipeline-started` at its write, so its revision is pinned while it waits (P3). Its wait is said on
`…wait-held.<branch>.<source>`; the enums gain `pipeline` and `pipelines-at-once` (P8). `pipelines-stalled` and
`queue-full` have scope `delivery`, one each for the mesh; `queue-full` clears when the over-cap start is taken (P3).
- **H-A-53 (Q61 to Q64)** One `approval` step per mover whose policy names it (P10). `runner_place` is backed by a verb
`place` (`pipeline`, `why`) on the runner's seat (P4). A mover left in an older pipeline whose dependency was
superseded out builds against the newest published package, said in its step's why (P3; ADR 0682 §7). "A set-aside"
reads as `pipeline-ended` and `cancel` alone (P7).
- **H-A-54 (Q65)** The indexer's §749 pins land with the runner's R3 and are reviewed together. The lead's order.
- **H-A-55 (Q66, Q67)** Rule 2, rule 7, the network hold and A-428's exceptions end; the record-policy yes and the
outside wait stay as the asker's. `assigner_release` stays for puts and outside waits, losing its acting, order and
delivery steps. P7, P4; ADR 0682 §9.
- **H-A-56 (Q69, Q70)** Tier progress is the assigner's state `written`, key `pipeline.<branch>.<source>` with
`{tier, since, movers}`, one deadline timer; a tier counts placed movers only; `wanted` gains `pipeline`; `tier-late`
clears when the late movers run or a later pipeline passes; the end is said on `…wanted.<branch>.<source>` with
`outcome`. P4, P7, P1, P3.
- **H-A-57 (Q71 to Q74)** `acts` stays for the builder's build seats; `bound`, `DefaultBound`, `Wait` and the other
acting entries go (P7). The seat `mesh-delivery-checker` goes with every copy and its `check` and `recheck` kinds
(P7). The operator channel's limit, `held-back` and `ask-again` go in H1 with their state, verb, sdk reads and the
tool runner's row (P7; ADR 0682 §14). `indexer_unpin` is renamed to take the pin's key; one tool (P7).
- **H-A-58 (Q75 to Q78)** "Every durable starts at the newest and refuses the old form" holds for consumers of event
streams, not for watches of last-per-subject state; the refusal is in the shared follower; old means a
`take-in.taken` without `call` or a pipeline value without `movers`; the gate's and the forge module's durables are
dropped at the clean state's drop step, named there (P2, P3). The new `cancel` reads no old pipelines: they are
dropped (ADR 0682 §12). `assigner_want_drop` deletes the key, written in the hand-act log; the next pass writes the
new form (P4). A run's candidate graph is dropped on the `pipeline-ended` of its last member (P1).
- **H-A-59 (Q79 to Q84)** `pipeline-failed` has scope `delivery`, its owner as today (P8). A higher rank waits for a
lower rank of its pipeline on this machine alone, said as `wait-held` step `placing` (P3; ADR 0682 §9). When two
pipelines' wants changed at once, the older source first (ADR 0682 §7). "Its own want last" is the node-engine's
module; the node's composition stays first (P8). The bus module and the backup holder are found by claim
(`mesh-broker`, `node-backup`) from the entries the composer holds (P9). A revert is a want naming a fingerprint this
node ran before, read from the applied record (P1, with H-A-5).
- **H-A-60 (Q85 to Q88)** The entry asked again: on `catalog-revision` and on the link's reopening, at most `entry-asks`
times (default 5), then `unanswered` and the finding `entry-unanswered`; built in mesh-delivery's `meshcatalogclient`
(P3, P11). The node-engine's read of every node's record stays: it feeds the composer; the plan's section 3.5 and gap
table E row 67 are corrected (P9). Its wait on `restic.backed-up` by name is out of H1, a leftover in the gap table
(P9). Existing timers, polls and whole reads in files H1 touches stay for H4 (§762); new runner code reads only its
`queue` keys (ADR 0682 §11, P1).
- **H-A-61 (Q89 to Q92)** mesh-contracts' five modules are registered at the clean state's step 5.2 (P5). The whole
tests: the forge module's is a Go integration test running its bundle; the tool runner's `main` gains an entry; the
request-count assertion is H4's (with H-A-17). #708's description drops the words "config candidates" before it
merges (P8). Tests that use the checker's state as a fixture rename it to a state that stays (P7).
- **H-A-62 (Q68, for the operator)** Until H2, a version is released only for a registration's first watched branch, so
a placement on another branch gets no want and is said. Recommendation: accept for H1; H2 releases per branch. Asked
of the operator 2026-11-20 02:25 CET.
## From H1's develop (2026-11-20 from 02:27 CET)
- **H-A-63 (seats developer, 02:31)** The `apply-order` row's `waits` is a list of module names the want waits on
(strings), empty when it waits for nothing; `why` stays the one sentence. P8 (one form, named by the rows' own
modules).
- **H-A-64 (seats developer, 02:33)** The word `want:version` is a generated constant of the node-engine seat contract:
the seat generator's identifier function splits on `:` as it splits on its other separators (one more rune, no
existing output changes; a subtask of the contracts-indexer-bridge developer in `mesh-delivery-sdk/seatgen`), and the
node-engine definition gains the enum `ManifestReadWord` with that value. P9 (every reader takes it from the generated
package), P7.
- **H-A-65 (tool runner developer, 02:33)** How the setting `call-timeout` reaches the tool runner: not in H1. The
node-engine composes the tool runner's unit as a special case (two environment variables, no config file), so the
setting's path is the generic manifest `config` with its file resource, built in H4 when that special case goes
(principle 7, no module special-cased in shared code). In H1 the runner's own bound is a constant of 30 seconds and the
per-verb `timeout` carries every real need (`merge-pull` 120); the manifest declares no setting it does not read
(principle 3). Rules-are-weighed: principle 11 bent for one release, said to the operator.
- **H-A-66 (seats developer, 02:36)** H-A-41's set is the whole of it: mesh-sdk's `go/callers.go` (`stdio.Callers`,
`VerbCallers`, `ParseVerbCallers`, `ConfiguredVerbCallers`, `HoldVerbCallers`), the `callers` parameter of
`seatlist.NewPart`, `Tools` and `Refusal`, of `views.FirstViewTool`, `ParseCallers`, `Refusal` and `CallersEnv`, and
`approval.ViewCallersPrefix` go in H1, as the sdk developer's subtask `sdk-holder-callers-gone`; every gated first-view
tool and seat-list tool is open to every caller. The holders' call sites in mesh-delivery and mesh-catalog are their
developers' subtasks, from the signatures the sdk developer reports (the lead relays them). P6, P7 (code that gates
callers protects nothing once every account may call every verb); ADR 0682 §10.
- **H-A-67 (node-engine developer, 02:37)** The node-engine seat's `Wait.step` enum gains `placing` (the seats developer
in mesh-contracts, with the sdk copy); the hold of a higher rank behind a lower one is said with it. P8.
- **H-A-68 (node-engine developer, 02:37)** A revert is a want with no `pipeline` and no `rank` (a placement's `build`,
H-A-37) naming a fingerprint other than the one running; the applied record keeps no history for it. "Read from the
applied record" in H-A-59 reads so: the running fingerprint is what it compares with. P7.
- **H-A-69 (seats developer, 02:38)** How the approval gate knows a module's own first-view tool once the callers env
marker goes: by the tool's name, the forms `<module>_first_views` and `<module>_first_view_again` that mesh-sdk's views
already serve; no environment marker and no tool runner change. P7 (what exists over a new side channel).
- **H-A-70 (contracts-indexer-bridge developer, 02:40)** `acts` stays everywhere as a declaration: it is what the
catalog graph's kind-5 edge ("B has acted for A") is built from, and H-A-51 orders the pipeline's tiers by that edge.
So the `acts` entries of the seats mesh-catalog (take-in, validate), mesh-wants (place) and the build seats stay, as do
`ActingRule` and `CatalogEntry.acts`; what goes is the hold and its wait: `bound` on the build seats' `acts` and every
reader that waits or holds on an acting part (the builder's bound reader and its re-asks, the assigner's acting holds,
the runner's acting waits), each in its own developer's subtask citing this answer (builder and assigner readers:
the runner-and-assigner developer; the pipeline's: the same; the indexer's edge builders stay unchanged). P7 (the
declaration serves the order; only the hold had no need left); ADR 0682 §9.
- **H-A-71 (contracts-indexer-bridge developer, 02:49)** (1) The seat mesh-delivery's state `owed` stays in H1: the
bridge's owed path and the client's check read it (P7 removes only what nothing reads); its stale keys of the earlier
model go at the clean state's drop step (ADR 0682 §12), and §764's "owed (the old bucket)" reads so. (2) The call's
kept member gains `branch`, so the forge module can say `merge-called` again after a stop (P3). (3) The step runner's
asker judgement in `mesh-delivery-sdk/steprun` (H-A-41) is the contracts-indexer-bridge developer's subtask, since
that package is mesh-delivery's shared code and it already writes there; the three build modules' wiring stays
mesh-catalog's BUILD-1. The lead's order of work.
- **H-A-72 (contracts-indexer-bridge developer, 02:53; the runner's and mesh-wants' seats)** (1) The pipeline value:
`group` goes; `module`, `commit`, `fingerprint`, `repository`, `version`, `widens`, `tests` and `over-budget` move into
each mover; `run`, `published`, `unlogged` and `end-count` stay on the pipeline (P7). (2) The runner's state `queue`:
`queue.next` and one key `waiting.<sequence, 20 digits>` with value `{pipeline}`, so the lowest sequence is read by key
order (P1). (3) `wait-held` and `wait-ended` move to the one subject form `<branch>.<source>`, the body naming the
module for a step's wait (P8). (4) The step events and `run-started` keep `<module>.<branch>.<commit>`, the
commit-or-group token becoming `<commit>` (P8). (5) `pipeline-ended` carries per mover its step results as a list of
step results, each naming its module (H-A-35, P10). (6) The tier end is a second event of the seat mesh-wants,
`pipeline-wanted.<branch>.<source>` with `outcome`; `wanted` stays per node and module (P8); H-A-56 reads so.
- **H-A-73 (runner-and-assigner developer, 02:55; the map's Q-32)** Where §772's "held, awaits its hand placement" is
said: in the want's own `hold` field (H-A-37 keeps `hold` and `words`), with the word `awaits-hand-placement`, written
by the assigner for a node whose node-engine does not state `want:version`; no second event. P7.
- **A note on placement (02:55):** a developer's `go.work` per module pointing at the H1 worktrees is the build map's
placement, whatever small script wrote it; it is not a placement of the developer's own. The lead's reading.
- **H-A-74 (runner-and-assigner developer, 02:59)** H-A-70's builder part is only a reader of the build seats' `bound`
field, now absent from the contracts; the builder has none, so nothing of the builder changes in H1. G §672's
toolchain wait (the state `waits`, its events, tools and setting) is a build's wait for its toolchain, not a want's
hold on another delivery, and stays: §764 names no builder item, and the builder's lines are H2's. P7; ADR 0682 §9.
- **H-A-75 (contracts-indexer-bridge developer, 03:02; the seat mesh-wants)** (1) `pipeline-wanted.<branch>.<source>` is
said once per pipeline, at its end: `outcome` is `running` when every tier's movers run (the node-engine seat's word,
P8) and `tier-late` when a tier passed its bound, the body `{v, pipeline, branch, source, outcome}` and, for
`tier-late`, the movers and machines not running. (2) H-A-37's drop of `composed` is the module want's alone; the
node's own want `want.<node>.node` `{composed, words, self}` is unchanged, and `rank` is absent from it. P7.
- **H-A-76 (gate-and-forge developer, 03:04)** (1) `gate_source_again` says `source-moved` again with the `heads` the
source value keeps (every watched repository's head at that source) and `start` false; the indexer finds the movers
from the fingerprints, not from `heads` (A-476a). P7. (2) The clean state drops the source bucket (plan step 5.8)
before the seed (5.9); the seed numbers from 1 on an empty bucket and refuses to run while a key of the earlier form
exists, naming it, so a missed drop is loud and nothing is overwritten. P2, P3; ADR 0682 §12.
- **H-A-77 (node-engine developer, 03:08)** With `kept-on` gone from the want (H-A-37), nothing names the build node that
keeps a kept build, so the node-engine fetches every artifact from the artifact store; G §526's fetch through a build
node's `kept-artifact` has no input left and retires in H1 (added to what retires). P7 (a want names a version; the
store answers for it).
- **H-A-78 (contracts-indexer-bridge developer, 03:10)** (1) A want counts as applied when the have's fingerprint is
the want's (ADR 0682 §8 reads `fp` as the build's name); `motion.Want.Composed` goes from mesh-sdk (the seats
developer's subtask), the view passes the fingerprint. P7. (2) The view's source readers, the watchman's source-moved,
manifest-accepted and manifest-refused readers, and the grants part's readers of the removed caller invokers in
mesh-sdk are the contracts-indexer-bridge developer's subtasks after B4, with the watchman's acting-part-late (H-A-44
gives them the view's and watchman's readers; the grants part is mesh-delivery's shared code). The lead's order of work.
- **H-A-79 (indexer developer, 03:13)** With the validation gone from the take-in (§745, ADR 0682 §4): the state
`verdicts` leaves the seat mesh-catalog's contract (a small commit C1b by the contracts-indexer-bridge developer, the
seat's one writer) and the indexer with its tools `verdicts` and `indexer_verdicts`; the grants part's invoke of
`verdicts` and the watchman's invoke of `indexer_verdicts` go from their manifests (the contracts-indexer-bridge
developer's readers subtasks, with H-A-49: what they need they read from the mover's entry on `take-in.taken`). The
`oldsections` count and `indexer_old_sections` stay as they are: the validator still reads the foundation list there
(ADR 0619 §3), no line moves it in H1. P7; ADR 0682 §4.
- **H-A-80 (indexer developer, 03:14)** The indexer's tool `indexer_consumer` goes with the state `consumers` no ledger
writes any more (P7). A run's ended modules are read from `movers` (C-R), as the developer did. A developer who stops
with work in the tree commits it as "in progress: <subtask>" so the next developer starts from a commit, never from a
dirty worktree (the lead's order of work).
- **H-A-81 (gate-and-forge developer, 03:15)** `head-untaken` (scope delivery, id `<repository>.<branch>`): severity
warning, resolver operator (only `gate_head_take` delivers it, §742), no owner: nothing runs wrong, something is not
delivered. `merge-partial` (§767, keyed by the call): severity urgent, resolver operator: the repositories stand
half-merged until the refused members land or the call is cleared. P3.
- **H-A-82 (indexer developer, 03:16)** With `version-delivered` retired (§764, C-W), the indexer reads nothing in its
place: the fill of delivered versions into the graph's edges (versions_edges, versions_fill_edges) served the
assigner's holds, which are gone; the kind-2 edge stays as a declaration (H-A-51's tiers), its "delivered" fact is
nobody's to keep. The indexer's consume of `mesh-wants.version-delivered` goes from its manifest. P7; ADR 0682 §9.
- **H-A-83 (runner developer, 03:19)** (1) A run on request is one pipeline per run with the named commits' modules as
its movers (one take-in, one pipeline, ADR 0682 §6; the run delivers nothing), keyed `<branch>.run.<run id>` in the
runner's state `pipelines`: the seat gains that key form (a small commit C-R2 by the contracts-indexer-bridge
developer); H-A-52's "per member" is corrected to this. P8. (2) Inside a pipeline, a mover's build waits on a kind-1
edge for the From mover's `package-published`, and on a kind-3 edge (a build module, an image base, which publishes no
package) for the From mover's build step ending passed, which the runner hears as its own step end; toolchain edges
first. P3, P7.
- **H-A-84 (runner developer, 03:20)** A mover's heard dependencies are kept on the mover: a list `published` of the
artifact references whose `package-published` (and, for a kind-3 edge, whose build step's passed end) was heard at
that pipeline's branch and source, written with the pipeline by compare-and-set, so the wait survives a restart (a
C-R2 field, the contracts developer's). P3, P1.
- **H-A-85 (contracts-indexer-bridge developer, 03:21)** (1) The bridge's read tool of its state `heads` is
`bridge_heads`, in the bridge's tools, answering the state's keys with each head, read-only (H-A-48). P4. (2) A run's
pipeline puts in `source` the newest source revision of its branch at the run's start (what the run is checked
against), its key being `<branch>.run.<run>`. P7.
- **H-A-86 (assigner developer, 03:22)** The assigner's pins of assignment graph revisions go whole (the state `pins`,
the tools `assigner_pins` and `assigner_unpin`, the hearer of the gate's pin events): nothing feeds them once the gate's
pin operations are gone (H-A-28, H-A-48). An assignment graph revision that is not current and that no want or
running build names is removed at the same judging moments as today, pins aside. P7; §764.
- **H-A-87 (gate-and-forge developer, 03:24)** (1) The forge module's whole test is a Go integration test that runs
its bundle against an embedded bus (H-A-61 stands over §774's "TypeScript"; §774 is read so). (2) The gate's source
key is read with the same escaping it is written with (the fault in `merges.go`, where `Set` escapes the branch and
`Read` does not): fixed in GATE-9 with a test for a branch holding `/`. P3.
- **H-A-88 (assigner developer, 03:27)** The seat mesh-wants' `StepRun` gains `asker`, the seat name of the part asking
the step (the runner writes `mesh-pipeline-runner`), a small commit C-W2 by the contracts-indexer-bridge developer;
the assigner names a step's progress with that asker (A11, H-A-41). P8.
- **H-A-89 (gate-and-forge developer, 03:29; the record's pipeline families, §738)** All five as recommended: (1) each
per-mover step result becomes one record outcome, its step the result's step, its outcome the result's own word
(passed, failed, nothing to run, bypass), its why the result's, the entry (module, fingerprint, no node). (2)
`wait-ended` writes over `wait-held`'s `held` under the same outcome key, the outcome the wait's `how` (answered,
ran, superseded). (3) A `pipeline-ended` not passed (failed, refused, cancelled, superseded) ends the record `failed`
naming the pipeline's why. (4) `pipeline-wanted` with outcome `running` ends the record (every tier's movers run on
every present placing machine); `running` events are still written as running-healthy outcomes; `tier-late` ends it
`failed` naming the movers and machines not running. (5) The trace's `from` is the lowest and `to` the highest
assignment graph revision of this pipeline's `wanted` events, each want an entry (module, fingerprint, node) with
outcome `wanted`; a removed want writes nothing. P3, P7.
- **H-A-90 (indexer developer, 03:31)** (1) mesh-sdk gains `process.Account.NamedEventConsumer(ctx, js, stream, suffix)`,
binding the durable `<node>_<module>_<suffix>` (the seats developer's subtask); the indexer binds one per watched
branch with the escaped branch as the suffix (§743). (2) The indexer binds a newly watched branch's consumer on its
own `registration-changed`, not only at its start. P1, P8.
- **H-A-91 (bridge developer, 03:34)** (1) A newly registered branch adopted while the bridge runs says `head-moved` with
`start` true, like the start pass: `start` means "a read of the head, not a move", so the indexer takes it in as the
branch's first revision with no movers (§745). P8. (2) The watchman's `delivery-waits` finding goes with
`acting-part-late` (§764), in the same pass. (3) The seat mesh-delivery's contract drops the owed act `merge`
(OwedActMerge), a small commit C3b by the contracts developer; the bridge's refusal of it as undecodable then has
nothing to refuse. P7.
- **H-A-92 (indexer developer, 03:37)** A catalog revision carries its movers (option c): a redelivered `source-moved`
whose revision is already kept says `take-in.taken` again from the kept revision's movers, recomputing nothing, so the
pin judge may remove the revision before as today and a restart between the write and the acknowledgement loses
nothing. P3 (an idempotent re-say), P7 (no extra kept revision, no extra record).
- **H-A-93 (gate-and-forge developer, 03:41)** The forge module's finding `step-unread` (G §585) is one per unread
step: scope `module`, its subject the step's module, its key `<run>.<module>.<step>`, severity warning, resolver
operator, no owner, cleared by that step's later `step-ended` carrying its fingerprint; the body's `fingerprint`
alone is read. P3, P8.
- **H-A-94 (readers developer, 03:42; the watchman)** (1) The watchman learns the bus-state entries a mover's manifest
asks for from the mover's catalog entry, read by fingerprint through the seat mesh-catalog's verb, never from a
retired event (P1, P9); `bus-state-overdue` stays, timed from `take-in.taken`. (2) Its take-in reader is rewritten now to
the seat's `TakeInTaken` and `TakeInPending`, seeded from `indexer_pending`'s answer in the contract's form
`{branch, source, why}` (build against the contract, not an indexer commit); the wait's key is `<branch>.<source>` and its
`since` the message's own time; the bound is the watchman's setting `take-in-bound`, default 60 minutes; the
30-minute group bound goes (§764). (3) §741: the watchman learns no head; the head timing of pipelines.go goes with
its source-moved reader's change. P3, P11.
- **H-A-95 (readers developer, 03:44)** (1) A want with `hold` (awaiting its hand placement, §772) is not timed by the
watchman: a hold is not progress owed (as H-A-14 reads a waiting pipeline). (2) The Owed definition's `ask` and
`group` fields, kept "for a merge", go with the act (C3c, the contracts' writer); nothing reads them. P7.
- **H-A-96 (runner developer, 03:48)** The runner's state `queue` gains a third key form `running.<pipeline key>` with
value `{pipeline}`, written by compare-and-set when a pipeline is placed and deleted at its end, its cancel or its
approval wait; `pipelines-at-once` is judged against the `running.*` keys alone (a C-R3 field, the contracts' writer).
P1, P3; H-A-60 holds (the runner reads only its queue keys).
- **H-A-97 (indexer developer, 03:49)** A held take-in is kept in a small indexer state `held`, one key per branch
`{branch, source, why, since}`, written when a `source-moved` is held and deleted at its take-in or skip (option c):
`indexer_pending` reads it, and the watchman seeds from it (H-A-94). P4 (what the module holds is read, not derived
from a scan of events), P3 (`since` is kept for the bound). The retired state `pending` was the pending revision of
the manifest validation (ADR 0633 §1); this is not it. `indexer_take_in_pending` and `indexer_take_in_skip` are asks in
the state `takeins`, the existing hand-act path. P7.
- **H-A-98 (runner developer, 03:50)** No shared follower exists, so in H1 each consumer's own handler refuses an
old-form event: the contract's strict decode (a required field such as `call` missing) is the refusal, the handler
writes one journal line naming the subject and ends the message. H-A-58's "shared follower" is read so for H1; a
shared shape, if any, is H4's with the wake-up rule. P3, P7.
- **H-A-99 (assigner developer, 03:51)** The placement's `build` is read as the want's target before the rule (option
b): the want's `fp` is that build and its `version` the one released at it, with no pipeline, no rank and no put steps;
`placementPuts` goes. "Read before the rule" literally, with no hold. P7.
- **H-A-100 (gate-and-forge developer, 03:51; the forge module's merge-check machinery)** As recommended: the poll that
goes is the pass's pull-passing half with its 60-second retry, the timer the 10-minute keep-main-rules step (it keeps
a protection ADR 0682 §13 no longer judges); the repository poll (`repo.created`) and the start's whole read stay.
`announceHead` goes whole with its pending status and sibling weighing; `pull.updated` is said by the webhook path
alone; the poll of merged and closed pulls stays. The seat verb `status` writes a head's status through the client
directly; `GateWriter` and its chain go. P7; §764.
- **H-A-101 (runner developer, 03:52)** (1) The runner's verbs `rerun`, `accept` and `ask-again` name the mover's
`module` beside the pipeline and step, so a pipeline of several movers is never ambiguous (a C-R4 change of the
runner's seat by the contracts' writer); a call naming no module on a pipeline of one mover is taken as that mover.
P8, P3. (2) A step of a mover whose dependency did not build ends with its own `step-ended`, outcome `not asked`,
naming the dependency, so nothing ends silently. P3.
- **H-A-102 (readers developer, 03:53)** `indexer_pending` answers `{"held": [{branch, source, why, since}]}`, `since`
in RFC 3339, one form with the indexer's state `held` (H-A-97); the watchman reads it with a local type over the
seat's `TakeInPending` plus `since`. P7, P8.
- **H-A-103 (indexer developer, 03:55; two things held in memory)** (1) Between take-ins the indexer's view of a module
change (a build's imports, a toolchain change, the reconcile) is derived state: the kept revision is the truth and a
restart rebuilds the view from it and the events after it, so nothing kept lives only in memory. (2) A held
message's acknowledgement handle is the bus consumer's; after a restart the bus redelivers the message and the state
`held` says why it was held. Neither is a kept truth in memory; the review reads them so. P2, P3.
- **H-A-104 (assigner developer, 03:57)** (1) The assigner keeps each node's words in its own state `engine-reads`
(key the node, value `{words, at}`), written when it hears `manifest-reads-changed`, so a restart loses no node's
words and no verb is asked; the shared client stays as it is, used for the event alone. P1, P2. (2) `RefusedCycle`
moves into `mesh-delivery-sdk/graph` on the contracts branch (a small commit by the readers developer, the branch's
writer of shared code), so the assigner's `validate` step and the indexer both read it from there (H-A-34). P9.
- **H-A-105 (indexer developer, 03:59)** (a) `indexer_take_in` goes whole with the builder-log index (H-A-48, §742:
the gate's `gate_head_take` is the one hand take of a head and nothing else delivers an untaken head; acceptance
test 36 and G-5's restore step are corrected to name `gate_head_take` when the design is next amended). P7. (b) The
run candidate's check of a module named twice with different manifests stays: it guards a run on request, which
stays (H-A-58); only the pull-request candidates go. P7.
- **H-A-106 (runner developer, 04:06; the queue)** (a) A start over the cap is written as an ordinary waiting pipeline
(its value `waiting`, its `waiting.<seq>` key) beyond the cap; `queue-full` is raised naming it and cleared when an end
brings the waiting count back to the cap or below; nothing is left unacknowledged (H-A-36 over §748's wording). P1,
P3. (b) A pipeline whose approval is answered and still runs has its `running.<pipeline>` key written at once, over the
bound, since it asks nothing new but its end; "waiting on its approval step" means every step not ended is an approval
asked of the operator. P3. (c) `pipelines-stalled` and `queue-full` have the kind as their id, severity warning, no
owner (the operator's), like `pipeline-failed`'s writer. P8.
- **H-A-107 (runner developer, 04:09)** The pipeline value keeps `superseded`, the modules a newer queued take-in took out
of it (written with the movers by compare-and-set; a C-R5 field of the seat, the contracts' writer); at the pipeline's
placing the runner writes a mover's build-step why from it (H-A-53). P1, P4.
- **H-A-108 (assigner developer, 04:10; the validate step)** All four as recommended: (1) the revision's validate run is
a key `<pipeline>.validate` of the state `composition-runs` in the CompositionRun form with no machine, taken up at
start like the composer's runs (P1, P7); (2) the validator's request carries no predefined key list and no known
package names until a design line names the verb that gives them: the validator takes every fallback as declared and
judges no package name (P7); (3) "a manifest the revision alters" is judged against the branch's previous revision,
read through the seat mesh-catalog's verb `revision` (P1); (4) one run, one verdict per revision: every mover's
validate step ends the same, failed with `manifest-refused` naming every problem of the revision (P3).
- **H-A-109 (readers developer, 04:12; the view's held take-ins)** A held take-in names no module in the view: its
movers are not known until it is taken in, so the view says the branch's held take-in in the tiers' pending words
(branch, source, why, since) and holds no module for it (option b). P3 (no module named on a guess).
- **H-A-110 (node-engine developer, 04:13)** Within one link, the node-engine asks its unanswered entries again when it
hears the seat mesh-catalog's `catalog-revision` (one per take-in): the composer consumes that event for this alone and
calls the entries client's `AskAgain`; and once more at each link's start. No timer. P1.
- **H-A-111 (node-engine developer, 04:15)** With no verdict state (H-A-79, ADR 0682 §4), the node-engine's reading of
a taken module (frombus `taken`) drops the verdict branch and decodes the manifest alone: a module in a catalog
revision is accepted by being there; the validate step judged it before any want. P7.
- **H-A-112 (node-engine developer, 04:18)** The composer (mesh-compose) likewise drops the verdict branch of its parse
and its case verdict, and the request field `Case.Verdicts` goes: its one writer was the checker, which goes in H1
(H-A-57); the assigner's composition step sets no verdicts (H-A-108: the validate step judges before the composition
step). P7.
- **H-A-113 (readers developer, 04:20; the grants part)** (1) `grants_callers` goes whole with the callers lists;
`grants_why_refused` stays, answering the bus's own verdict for the account and the verb's call subject (a refusal
is still readable, P4). (2) §752 and ADR 0682 §10 mean every verb of every seat: the grants part grants each module's
account every seat's verbs, not only the seats its manifest invokes (acceptance test 118's case); that change is this
developer's, in the same pass. P6.
- **H-A-114 (indexer developer, 04:40; a held take-in across a restart)** A held take-in is acknowledged on the bus and
kept only in the indexer's state `held` (H-A-97); the indexer takes the held ones in again from that state on its own
`room` and at its start. No message stays unacknowledged, so the consumer's acknowledgement wait bounds nothing and a
restart loses nothing; H-A-12's and H-A-50's "held, unacknowledged" are read so. P3 (a stuck message for the length of
the acknowledgement wait would be a silent failure), P2.
- **H-A-115 (indexer developer, 04:40; the movers' code dependents)** In H1 a code dependent of a mover is a module
whose Go module requires the mover's (the go.sum reading built), which moves a module that requires the package
without importing a changed package; the finer reading by imported package is H2's refinement of the same line
(§743, A-476a). P7 for H1, said.
- **H-A-116 (readers developer, 04:43; the builder's floor)** In H1 the builder's version floor wakes on every
`take-in.taken` (it consumes the indexer's `take-in.taken` in place of the retired `manifest-accepted`, H-A-49) and
looks at every module again as it does today; the finer wake on a mover's new entry is H2's with the builder's own
lines. The mechanical changes SDK-C forces on the builder's seat-list calls are in scope. P1, P7.
- **H-A-117 (build-modules developer, 04:57)** The build seats' step kinds `check` and `recheck` stay in H1: they carry
the composer's and the validator's runs (a compose or run ask is kind `check` of the Go seat), and §753's queues are
H2's to rename with the builder's lines; H-A-57 is narrowed to the checker's seat, its copies and its own verdict
verb. The runs' input moves from the checker's `input` to the seat mesh-wants' `input`, which the assigner serves
(H-A-34), not to mesh-catalog. P7; rules-are-weighed, said.
- **H-A-118 (join developer, 05:03; the held take-in's form)** (1) The state `held`'s value carries the held
`source-moved` whole beside H-A-97's four fields: `call`, `heads` and `start`, so a take-in after a restart reads
nothing from any verb (P2). (2) A later `source-moved` of a held branch is held behind the earlier one: the branch's
key holds its held events in order, each acknowledged once written there, and the indexer takes them in in that order
on `room` and at start; no source revision ends silently (P3).
- **H-A-119 (join developer, 05:09; the teller)** `pipeline-ended`'s per-mover entry gains `commit` and `widens`
(a C-R6 change of the runner's seat, the runner writing them from its pipeline value), so the teller says one widening
per mover from the event alone and reads no other module's state (P10, P1). The teller's hearing of `record-opened`
and its consume go: the record's H1 form carries no forge-made, and §763 retires acceptance test 53's rows (P7).
- **H-A-120 (fixer, 05:29; the build seats' queue order)** The build seats serve their kinds in turn, as lead answer
A-440 (the operator's yes of 2026-11-15) decided and the code since 2026-11-15 does; the red test still asserting ADR
0525 §4's deliveries first is corrected to the turns (delivery, check, delivery), citing A-440. H-A-8 and H-A-51's
"deliveries first" were wrong: a line existed. ADR 0525 §4 gets a progressive insight in H2 with the builder's lines.
P7 (build what the last decided line says).
- **H-A-121 (the clean state's drop tools, 05:46; corrected 05:52)** The plan's step 5.8 needs tools no module serves
(the reader of 05:43): (1) the bus module gains two tools beside the seat mesh-broker's verbs: `nats_drop_bucket`
(bucket, why: drops one named bucket after a snapshot) and `nats_consumer_delete` (stream, durable, why: deletes one
durable consumer); both written in the hand-act log. They are the operator's hand tools for one module's own kept
data, not a need another module fills: a module tool, not a seat verb (P9; the seat's definition stays as reviewed).
(2) `gate_record_end`, `assigner_want_drop` and `indexer_pin_drop` take an `all` form (every open record, every held
or failed want, every stale pin) beside their one-item form, each saying what it dropped, counted. (3) The source
bucket's old keys go by `nats_drop_bucket` of that bucket before `gate_source_seed`. Built in H1 on the H1 branches as
the clean state's own tools. P4 (everything that goes wrong can be recovered through tools), P3.
- **H-A-122 (the mail reader's callers list, 05:52)** Goes, like every other holder's (H-A-41, H-A-66; ADR 0682: every
account may call every verb). The mesh has one user and is assumed not compromised (the operator, 2026-11-11); who
inside the mesh may read mail through a verb is who-may, a design detail, not public exposure: nothing here puts mail
outside the mesh. ADR 0459 §2's protection of mail data stays in what the verbs answer, not in who may call them. P1.
- **H-A-123 (what `assigner_want_drop all` drops, 05:53)** There is no failed want: a want has `hold` and nothing
else that marks a fate (the node-engine's report of a failed have is the node-engine's state, not the want's). The
`all` form drops every want key in the wants bucket, held or not, and every value of an earlier form the decoder
refuses; it answers the count per kind (held, current, undecodable). The clean state drops every want before the
seed (plan step 5.8; the operator, 2026-11-20: old data may go). H-A-121 (2) is read that way. P4.
- **H-A-124 (what `indexer_pin_drop all` drops, 05:56)** Every pin in the indexer's pins state, as H-A-123 drops every
want: the indexer knows a running pipeline only through its own pins, so no other reading of "stale" exists without
a read of another module's state (P9 forbids it). The `all` form is the clean state's tool, run while no pipeline
runs (plan step 5.8, after the runner is stopped); it answers the count dropped and the pipelines they named. P4, P9.
- **H-A-125 (the knowledge base's delivery block, 05:58)** Goes whole: the overview's delivery source, its
`delivery-groups` block, the progress verb's groups lines, and with them the knowledge base's consumes of
`pull-merged`, `transition` and `group-ended`; `pipeline-ended` is then not consumed by the knowledge base at all,
which corrects H-A-44's reading. The block showed another module's state (the deliveries), which only the status view
owns (P9: a module's state is read through its owner's view, never copied); a pipelines block belongs in the status
view and is parked as an idea, not built in H1. Told to the operator as a change of what they see: the knowledge
base's overview page no longer shows deliveries. P9, P5.
- **H-A-126 (the bus module's drop tools exist; H-A-121 (1) withdrawn, 06:12)** No new tool: the bus module already
serves them (the developer's reading of 06:11). A retired state bucket (the gate's `merges`, a seat's `asks` and
`turns`, mesh-config's `pulls`, the indexer's `pending`) is kept data no accepted manifest asks for once H1's
manifests are taken in, so the keeper's `nats_drop` drops it, by warrant, after its snapshot, as today. The source
bucket's old keys go by `nats_bucket_remove` with the prefix, before `gate_source_seed`. A durable consumer goes by
`nats_consumer_remove`; it refuses only a consumer an accepted manifest asks for, and no longer every consumer of a
kept-data stream: one no manifest names is an orphan on that stream, removed by warrant (`why` required) and
written in the hand-act log as the keeper's requests are. The clean state's step 5.8 names these three. P4, P7 (no
second tool for one act).
- **H-A-127 (`nats_consumer_remove`: what it refuses and what warrants it, 06:16)** (1) No manifest field names a
durable; the mesh names a module's durable `<node>_<module>`. The tool refuses a durable whose name is that form for
a module of an accepted manifest on that node (the keeper's accepted set), on any stream; every other durable is an
orphan and goes. The judgement is by name, not by stream (H-A-126). (2) `why` is the warrant: the tool acts at once
and writes the hand act through mesh-hand-acts as `nats_bucket_expiry` does. The clean state runs at the operator's
terminal, so an ask of the operator would answer the operator's own call (P7). P4, P7.
- **H-A-128 (`nats_consumer_remove`: the name rule is a prefix, 06:24)** H-A-127's rule is read as a prefix: a durable
is a module's when its name is `<node>_<module>` or begins with `<node>_<module>_` for a module of an accepted
manifest on any node (the keeper's accepted set holds modules, not nodes; the developer's reading of 06:23), since a
module may name more than one durable with a suffix. The tool refuses those and removes every other. P3 (a wrong
removal of a live consumer is a silent failure), P4.
+4 -1
View File
@@ -22,7 +22,10 @@ try in process. Four more carry playbook 07: `mesh-preparer` (the prepare step),
`mesh-code-reviewer` (the review) and `mesh-watcher` (every wait). A tenth, `hq-design-writer`, writes one set of design amendments
(playbook 02, *Amending an existing design*) from accepted decisions and a milestone's findings, in a worktree of its
own, ready for the review by architectural domain. An eleventh, `hq-record-writer`, proposes one decision record or
research effort to the knowledge base from a draft (playbooks 01 and 02) and follows its review to approve. An agent that researches, decides, designs or reviews a design starts
research effort to the knowledge base from a draft (playbooks 01 and 02) and follows its review to approve. Two more serve a design
delivered again part by part (design 47, milestone H): `spec-reader` writes one part's specification from the design and its
records (no code read), and `gap-surveyor` holds main's code against that specification and writes the part's gap table
(read-only on the mesh and every working copy). An agent that researches, decides, designs or reviews a design starts
from the knowledge base and maps each interaction onto a mesh path. An agent that prepares, builds or reviews code
starts from the milestone's part of its design alone, builds what its design lines say, and decides nothing: what
the design does not answer is a design question, answered in an hq commit before any code (playbook 07). The prepare