Adds design 57: the work of ADR 0348 (config is a domain of its own) and ADR 0349 (provisioning) as 37 work packages, each with its repository, size, what changes and its proof, in dependency order. The shape is design 38's.
Each package is an issue: 650-685, plus 647 (a node's secrets come from the vault), each linked with depends-on to the packages it waits for and to ADR 0348 or 0349.
At most five are worked at once; the first wave is T1 (654), F1 (650), F3 (652), P1 (667) and F4 (653).
Counts come from the code at mesh-controller 1601cb1, mesh-host 1a26778, mesh-sdk a2d2af0, mesh-catalog ebf7f37.
Adds design 57: the work of ADR 0348 (config is a domain of its own) and ADR 0349 (provisioning) as 37 work packages, each with its repository, size, what changes and its proof, in dependency order. The shape is design 38's.
- Each package is an issue: 650-685, plus 647 (a node's secrets come from the vault), each linked with `depends-on` to the packages it waits for and to ADR 0348 or 0349.
- At most five are worked at once; the first wave is T1 (654), F1 (650), F3 (652), P1 (667) and F4 (653).
- Counts come from the code at mesh-controller 1601cb1, mesh-host 1a26778, mesh-sdk a2d2af0, mesh-catalog ebf7f37.
- The README's to-be table gains its row.
hq's merge-check.sh passes locally (records, index, cycle, words).
So config and provisioning are built in small, provable steps with at most five worked at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Deliverynovox/hq@e5b98e3b0a42 — delivered since 2026-10-13T23:57:27Z
Delivery plan — builds nothing: the change touches no module of the mesh's graph
Transitions
2026-10-13 23:54 (new) → proposed (announced): novox/hq#350's head announced
2026-10-13 23:55 proposed → checked (checked): a verdict names this commit and the trunk's head as its trunk commit
2026-10-13 23:55 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
2026-10-13 23:57 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move
2026-10-13 23:57 published → delivered (done): nothing for a walk to move
The commit's note under refs/notes/mesh-plan keeps every transition: git log --notes=mesh-plan.
<!-- mesh-delivery:view -->
**Delivery** `novox/hq@e5b98e3b0a42` — **delivered** since 2026-10-13T23:57:27Z
**Delivery plan** — builds nothing: the change touches no module of the mesh's graph
**Transitions**
- 2026-10-13 23:54 (new) → proposed (announced): novox/hq#350's head announced
- 2026-10-13 23:55 proposed → checked (checked): a verdict names this commit and the trunk's head as its trunk commit
- 2026-10-13 23:55 checked → ready (accepted): the gate passed or warned, and the repository's own check did not fail
- 2026-10-13 23:57 ready → published (merged): on the trunk its modules follow: merged there, and its walk opened — or nothing for a walk to move
- 2026-10-13 23:57 published → delivered (done): nothing for a walk to move
The commit's note under `refs/notes/mesh-plan` keeps every transition: `git log --notes=mesh-plan`.
The operator chose one delivery over one per package; the packages meet on one branch per repository and reach main as one delivery group.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Verdict: REQUEST_CHANGES. I posted this as a comment because the forge refuses a tool's REQUEST_CHANGES (ADR 0266).
Independent review at 1405dd39. hq's merge check passes (records, index, cycle, words). The frontmatter and the README row are right. All 37 packages are present. Every dependency in the table matches the issue tracker (details in 5). Must change:
Node names in a public repository. README.md:68 and AGENTS.md:85-89 forbid node names and ask for role names. Design 57 names one at lines 46, 202-203 (K7), 239 (P6) and 243 (P7). Use a workstation or the operator's workstation instead.
The failure rule misreads the ADRs (lines 48-49). Under ADR 0276 §3, a failed walk fails all its own commits. ADR 0299 §1 walks the others again only after a stop, and says "nothing failed here". ADR 0299 §2 also leaves out every merge that shares modules with the stopped one, and in this release that is almost every member. Say that a failed release walk fails the whole release and that the fix is a new merge.
The issues and the design disagree on when a package starts. Lines 27 and 105-106 say a package starts once its dependencies are merged into the release branches. Most issue reports (655-685) say "Waits for: X delivered", and nothing is delivered before the release. Make one of them match the other.
B2's proof contradicts V5 (lines 267-268, 280). Under ADR 0349 §8, the vault remakes the pair credential when it sees a consumer's need recorded. Moving PostgreSQL records each need for the first time, so the credential would be remade. Yet B2 promises "no rotation". Say how the move adopts the existing credential without that counting as a bind.
Dependency links (lines 105-106). The text says the tracker "holds the same links". Thirteen issues also link implied, transitive dependencies: 656, 657, 658, 671, 672, 673, 675, 676, 679, 680, 682, 684 and 685. None is missing or contradictory. Say "the same or implied".
Should change:
K5 (lines 194-195). The controller reading mesh-config_resolved conflicts with ADR 0348 §9 unless it is ADR 0349 clause 17's transitional exception. Say that it is, and have R1 remove that read.
K1 (lines 166-169). The refusal list leaves out five checks from ADR 0348's "How it is checked": no default and not required, an unknown or mistyped predefined key, a value used nowhere, a reload-on or restart-on naming an undeclared file, and an unnamed takes-config.
The counts at line 67 and line 69.${bound:} occurs 117 times at ebf7f37. ADR 0349 says eleven modules keep a second copy of their text, and line 69 says 12. The other counts I spot-checked match: 225/88/193 KB, 975, 1125, 15/9/4, 4/3/1, and the ten modules without a build.
Line 258. For "operator key's export", write "its public key". operator_key holds only the public half, and private keys never reach the vault (ADR 0349 §7).
Lines 35-37. One walk needs all five merged within one merge window (ADR 0276 §1). Say that, and say the order: node-engine, then controller, then catalogue (ADR 0239 §12).
**Verdict: REQUEST_CHANGES.** I posted this as a comment because the forge refuses a tool's REQUEST_CHANGES (ADR 0266).
Independent review at 1405dd39. hq's merge check passes (records, index, cycle, words). The frontmatter and the README row are right. All 37 packages are present. Every dependency in the table matches the issue tracker (details in 5). Must change:
1. **Node names in a public repository.** README.md:68 and AGENTS.md:85-89 forbid node names and ask for role names. Design 57 names one at lines 46, 202-203 (K7), 239 (P6) and 243 (P7). Use `a workstation` or `the operator's workstation` instead.
2. **The failure rule misreads the ADRs (lines 48-49).** Under ADR 0276 §3, a failed walk fails all its own commits. ADR 0299 §1 walks the others again only after a *stop*, and says "nothing failed here". ADR 0299 §2 also leaves out every merge that shares modules with the stopped one, and in this release that is almost every member. Say that a failed release walk fails the whole release and that the fix is a new merge.
3. **The issues and the design disagree on when a package starts.** Lines 27 and 105-106 say a package starts once its dependencies are merged into the release branches. Most issue reports (655-685) say "Waits for: X delivered", and nothing is delivered before the release. Make one of them match the other.
4. **B2's proof contradicts V5 (lines 267-268, 280).** Under ADR 0349 §8, the vault remakes the pair credential when it sees a consumer's need recorded. Moving PostgreSQL records each need for the first time, so the credential would be remade. Yet B2 promises "no rotation". Say how the move adopts the existing credential without that counting as a bind.
5. **Dependency links (lines 105-106).** The text says the tracker "holds the same links". Thirteen issues also link implied, transitive dependencies: 656, 657, 658, 671, 672, 673, 675, 676, 679, 680, 682, 684 and 685. None is missing or contradictory. Say "the same or implied".
Should change:
6. **K5 (lines 194-195).** The controller reading `mesh-config_resolved` conflicts with ADR 0348 §9 unless it is ADR 0349 clause 17's transitional exception. Say that it is, and have R1 remove that read.
7. **K1 (lines 166-169).** The refusal list leaves out five checks from ADR 0348's "How it is checked": no default and not `required`, an unknown or mistyped predefined key, a value used nowhere, a reload-on or restart-on naming an undeclared file, and an unnamed takes-config.
8. **The counts at line 67 and line 69.** `${bound:}` occurs 117 times at ebf7f37. ADR 0349 says eleven modules keep a second copy of their text, and line 69 says 12. The other counts I spot-checked match: 225/88/193 KB, 975, 1125, 15/9/4, 4/3/1, and the ten modules without a build.
9. **Line 258.** For "operator key's export", write "its public key". `operator_key` holds only the public half, and private keys never reach the vault (ADR 0349 §7).
10. **Lines 35-37.** One walk needs all five merged within one merge window (ADR 0276 §1). Say that, and say the order: node-engine, then controller, then catalogue (ADR 0239 §12).
Role names instead of node names, the failure rule as ADR 0276 and 0299 have it, readiness and adoption of existing credentials said exactly, and the counts and refusals corrected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Independent re-review at 2f3e6e6641 (against review 236). MESH_CATALOG_DIR=<mesh-catalog at ebf7f37> sh merge-check.sh passes: records, index, cycle (437 documents) and words. No node name is left in the document.
Nine of the ten earlier changes are answered correctly. Role names replace the node names. The failure rule now follows ADR 0276 §3 and ADR 0299 §1-2 (two small points in 5 below). "Waits for … delivered" is now defined as merged into the release branches (lines 111-112). V5 says how the move adopts the existing pair credential, and B2 points to it. The tracker links now say "thirteen issues also link what they wait for through another package"; I counted the same thirteen. K5 is called ADR 0349 clause 17's exception to ADR 0348 §9, and R1 removes it. K1 now gives all eleven registration refusals from ADR 0348's How it is checked, word for word. The counts are right: I measured 117 ${bound:} in module.json at ebf7f37, and 12 modules with 34 copy files (lemurs has config/config.toml and a manifest_test). The public half of the operator key is named. One merge window is stated. The group order is only partly answered (finding 4).
The dependency table matches every depends-on link of issues 650-685 and 647 (I checked all 37 issues), except for the case in finding 3.
The 67 documents in R2 are correct. I added setting/settings to the Not: line of F4's glossary (hq#351 at 1f6d1b3), and words.py named 68 files: the glossary itself and the same 67 listed in the record.
Must change:
F4 contradicts R2 and F4's own pull request (lines 135-137 against line 328). F4 says "setting is named on config value's Not: line", and its proof is "fails on a planted settings". R2 says that F4 left it off, and hq#351 at 1f6d1b3 does leave it off ("The word is not yet on a Not: line … checked by review"). Reword F4, including its proof, to say what it actually does.
The table is broken at lines 72-73. A line break sits inside the "file text" row ("…context counted\neleven; the twelfth is lemurs)"). GFM ends a row at the end of the line, so the row's Becomes cell is empty and line 73 renders as a separate two-cell row. Put the row on one line.
P3's dependency on F3 is stated but not linked (lines 113-114 against line 93 and issue 669). The prose says "P3 waits for F3 because both change internal/broker", and P3 is left out of the first wave. Yet the table gives P3 as "—", and issue 669 has no depends-on link to 652. Because the tracker decides which packages are ready, P3 could start alongside F3. Add P3 F3 to the table and the link to issue 669, or drop the sentence.
Should change (these may follow):
The group order (line 37) is not what ADR 0239 §12 infers. §12 gives node-engine before controller, and controller before a member that changes manifests. Otherwise it orders by repository name, which puts hq first and mesh-sdk last, not mesh-sdk first and hq last. Either declare the order with after: lines in the pull request descriptions (ADR 0239 §11) and say so, or state only what §12 gives: mesh-host, then mesh-controller, then mesh-catalog.
Lines 50-52. "Only after a person's stop" is too narrow. Under ADR 0299 §1, mesh-delivery itself stops a group's later members after a failed member (ADR 0239 §17). The text also still does not say that the fix is a new merge, which a person then releases (ADR 0239 §17).
Issue 660's report is shorter than K1. Its Expected lists eight refusals. It leaves out a predefined key that is not on the list or has another type, a value neither in a template nor read at runtime, and a moved module's mergeable file not naming its keys. The agent that takes K1 works from the issue, so add a note there that points to the design's full list.
Line 269 (nit). "…and the operator key's public half are copied into the vault as it is": the subject is plural, so write "as they are".
Verdict: CHANGES at 2f3e6e6641d0
Independent re-review at 2f3e6e6641d0 (against review 236). `MESH_CATALOG_DIR=<mesh-catalog at ebf7f37> sh merge-check.sh` passes: records, index, cycle (437 documents) and words. No node name is left in the document.
Nine of the ten earlier changes are answered correctly. Role names replace the node names. The failure rule now follows ADR 0276 §3 and ADR 0299 §1-2 (two small points in 5 below). "Waits for … delivered" is now defined as merged into the release branches (lines 111-112). V5 says how the move adopts the existing pair credential, and B2 points to it. The tracker links now say "thirteen issues also link what they wait for through another package"; I counted the same thirteen. K5 is called ADR 0349 clause 17's exception to ADR 0348 §9, and R1 removes it. K1 now gives all eleven registration refusals from ADR 0348's How it is checked, word for word. The counts are right: I measured 117 `${bound:}` in module.json at ebf7f37, and 12 modules with 34 copy files (lemurs has config/config.toml and a manifest_test). The public half of the operator key is named. One merge window is stated. The group order is only partly answered (finding 4).
The dependency table matches every depends-on link of issues 650-685 and 647 (I checked all 37 issues), except for the case in finding 3.
The 67 documents in R2 are correct. I added *setting*/*settings* to the Not: line of F4's glossary (hq#351 at 1f6d1b3), and words.py named 68 files: the glossary itself and the same 67 listed in the record.
Must change:
1. **F4 contradicts R2 and F4's own pull request (lines 135-137 against line 328).** F4 says "*setting* is named on *config value*'s *Not:* line", and its proof is "fails on a planted *settings*". R2 says that F4 left it off, and hq#351 at 1f6d1b3 does leave it off ("The word is not yet on a *Not:* line … checked by review"). Reword F4, including its proof, to say what it actually does.
2. **The table is broken at lines 72-73.** A line break sits inside the "file text" row ("…context counted\neleven; the twelfth is lemurs)"). GFM ends a row at the end of the line, so the row's Becomes cell is empty and line 73 renders as a separate two-cell row. Put the row on one line.
3. **P3's dependency on F3 is stated but not linked (lines 113-114 against line 93 and issue 669).** The prose says "P3 waits for F3 because both change `internal/broker`", and P3 is left out of the first wave. Yet the table gives P3 as "—", and issue 669 has no depends-on link to 652. Because the tracker decides which packages are ready, P3 could start alongside F3. Add `P3 F3` to the table and the link to issue 669, or drop the sentence.
Should change (these may follow):
4. **The group order (line 37) is not what ADR 0239 §12 infers.** §12 gives node-engine before controller, and controller before a member that changes manifests. Otherwise it orders by repository name, which puts hq first and mesh-sdk last, not mesh-sdk first and hq last. Either declare the order with `after:` lines in the pull request descriptions (ADR 0239 §11) and say so, or state only what §12 gives: `mesh-host`, then mesh-controller, then mesh-catalog.
5. **Lines 50-52.** "Only after a person's stop" is too narrow. Under ADR 0299 §1, mesh-delivery itself stops a group's later members after a failed member (ADR 0239 §17). The text also still does not say that the fix is a new merge, which a person then releases (ADR 0239 §17).
6. **Issue 660's report is shorter than K1.** Its Expected lists eight refusals. It leaves out a predefined key that is not on the list or has another type, a value neither in a template nor read at runtime, and a moved module's mergeable file not naming its keys. The agent that takes K1 works from the issue, so add a note there that points to the design's full list.
7. **Line 269 (nit).** "…and the operator key's public half are copied into the vault as it is": the subject is plural, so write "as they are".
The word setting waits for R2, P3 waits for F3, the group's order is declared with after: lines, and a stop's fix is a new merge.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Independent third review at e5b98e3b0a, checked against review 239 (and review 236). MESH_CATALOG_DIR=<mesh-catalog at ebf7f37> sh merge-check.sh in a worktree of the head passes: records, index, cycle (437 documents) and words (28 retired words, 155 head words, each once). Every table in the document now has the same five-cell rows.
All seven findings of review 239 are answered:
F4 (lines 135-138) no longer says that "setting" is on the Not: line. It says that the word goes there only in R2, and that until then the glossary calls it the older word. That matches hq#351 at 21508a7, whose glossary entry says the word goes on the Not: line once issue 683 is done. The proof no longer claims a failure on a planted settings. R2 (line 330) agrees.
The "file text" row (line 73) is now one line.
The table gives P3 F3 (line 92), and issue 669 has a depends-on link to 652 (added 23:53:05).
Line 37 now declares the order with after: lines (ADR 0239 §11). Under ADR 0249 §2 a declared order outranks every inferred rule, so putting mesh-sdk first and hq last is allowed. The order also agrees with ADR 0239 §12's rules: node-engine before controller, and controller before catalogue.
Lines 50-53 now name mesh-delivery's stops as well as a person's, and say that the fix is a new merge into the release's branch.
Issue 660 has a note (23:53:05) pointing to design 57's K1 for the full list of eleven refusals.
Line 270 says "as they are".
Nothing from review 236 has regressed. No node name is in the document. The counts are the same (117 ${bound:}, 12 modules and 34 copy files, 225/88/193 KB). The failure rule still cites ADR 0276 §3 and ADR 0299 §1-2. "Waits for … delivered" is still defined as merged into the release's branches. V5 still adopts the existing pair credential, and B2 still points to it. K5 is still clause 17's exception, and R1 still removes it. The public half of the operator key is still named, and so is the one merge window.
I checked the dependency table against the depends-on links of all 37 issues (650-685 and 647). Every link matches the table, directly or through another package. The thirteen issues that also link what they wait for through another package are still the same thirteen: 656, 657, 658, 671, 672, 673, 675, 676, 679, 680, 682, 684 and 685. Issue 669's new link is a direct one, so the count of thirteen stays right.
Should change (these can follow; none blocks the merge):
Lines 50-53. A person's stop of a delivery on a shared walk asks first (ADR 0299 §5). Only with the way again are the others walked again; with walk, the whole walk stops. ADR 0239 §17 also says that a person releases what was stopped once the fix is merged. "The group is walked again" leaves out that step. Suggested wording: "…the fix is a new merge into the release's branch, and a person releases what was stopped (ADR 0239 §17)", with "(a person's stop, with the way again, ADR 0299 §5)".
Issue 669's report still ends "Waits for: nothing", while its link now says it waits for 652. The report cannot be edited, so add a note that P3 waits for F3 (both change internal/broker), so the agent that takes P3 does not read the report alone.
Verdict: APPROVE at e5b98e3b0a42
Independent third review at e5b98e3b0a42, checked against review 239 (and review 236). `MESH_CATALOG_DIR=<mesh-catalog at ebf7f37> sh merge-check.sh` in a worktree of the head passes: records, index, cycle (437 documents) and words (28 retired words, 155 head words, each once). Every table in the document now has the same five-cell rows.
All seven findings of review 239 are answered:
1. F4 (lines 135-138) no longer says that "setting" is on the Not: line. It says that the word goes there only in R2, and that until then the glossary calls it the older word. That matches hq#351 at 21508a7, whose glossary entry says the word goes on the Not: line once issue 683 is done. The proof no longer claims a failure on a planted *settings*. R2 (line 330) agrees.
2. The "file text" row (line 73) is now one line.
3. The table gives `P3 F3` (line 92), and issue 669 has a depends-on link to 652 (added 23:53:05).
4. Line 37 now declares the order with `after:` lines (ADR 0239 §11). Under ADR 0249 §2 a declared order outranks every inferred rule, so putting mesh-sdk first and hq last is allowed. The order also agrees with ADR 0239 §12's rules: node-engine before controller, and controller before catalogue.
5. Lines 50-53 now name mesh-delivery's stops as well as a person's, and say that the fix is a new merge into the release's branch.
6. Issue 660 has a note (23:53:05) pointing to design 57's K1 for the full list of eleven refusals.
7. Line 270 says "as they are".
Nothing from review 236 has regressed. No node name is in the document. The counts are the same (117 `${bound:}`, 12 modules and 34 copy files, 225/88/193 KB). The failure rule still cites ADR 0276 §3 and ADR 0299 §1-2. "Waits for … delivered" is still defined as merged into the release's branches. V5 still adopts the existing pair credential, and B2 still points to it. K5 is still clause 17's exception, and R1 still removes it. The public half of the operator key is still named, and so is the one merge window.
I checked the dependency table against the depends-on links of all 37 issues (650-685 and 647). Every link matches the table, directly or through another package. The thirteen issues that also link what they wait for through another package are still the same thirteen: 656, 657, 658, 671, 672, 673, 675, 676, 679, 680, 682, 684 and 685. Issue 669's new link is a direct one, so the count of thirteen stays right.
Should change (these can follow; none blocks the merge):
1. **Lines 50-53.** A person's stop of a delivery on a shared walk asks first (ADR 0299 §5). Only with the way `again` are the others walked again; with `walk`, the whole walk stops. ADR 0239 §17 also says that a person releases what was stopped once the fix is merged. "The group is walked again" leaves out that step. Suggested wording: "…the fix is a new merge into the release's branch, and a person releases what was stopped (ADR 0239 §17)", with "(a person's stop, with the way `again`, ADR 0299 §5)".
2. **Issue 669's report** still ends "Waits for: nothing", while its link now says it waits for 652. The report cannot be edited, so add a note that P3 waits for F3 (both change `internal/broker`), so the agent that takes P3 does not read the report alone.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Adds design 57: the work of ADR 0348 (config is a domain of its own) and ADR 0349 (provisioning) as 37 work packages, each with its repository, size, what changes and its proof, in dependency order. The shape is design 38's.
depends-onto the packages it waits for and to ADR 0348 or 0349.hq's merge-check.sh passes locally (records, index, cycle, words).
Delivery
novox/hq@e5b98e3b0a42— delivered since 2026-10-13T23:57:27ZDelivery plan — builds nothing: the change touches no module of the mesh's graph
Transitions
The commit's note under
refs/notes/mesh-plankeeps every transition:git log --notes=mesh-plan.Verdict: REQUEST_CHANGES. I posted this as a comment because the forge refuses a tool's REQUEST_CHANGES (ADR 0266).
Independent review at
1405dd39. hq's merge check passes (records, index, cycle, words). The frontmatter and the README row are right. All 37 packages are present. Every dependency in the table matches the issue tracker (details in 5). Must change:a workstationorthe operator's workstationinstead.Should change:
mesh-config_resolvedconflicts with ADR 0348 §9 unless it is ADR 0349 clause 17's transitional exception. Say that it is, and have R1 remove that read.required, an unknown or mistyped predefined key, a value used nowhere, a reload-on or restart-on naming an undeclared file, and an unnamed takes-config.${bound:}occurs 117 times at ebf7f37. ADR 0349 says eleven modules keep a second copy of their text, and line 69 says 12. The other counts I spot-checked match: 225/88/193 KB, 975, 1125, 15/9/4, 4/3/1, and the ten modules without a build.operator_keyholds only the public half, and private keys never reach the vault (ADR 0349 §7).Verdict: CHANGES at
2f3e6e6641Independent re-review at
2f3e6e6641(against review 236).MESH_CATALOG_DIR=<mesh-catalog at ebf7f37> sh merge-check.shpasses: records, index, cycle (437 documents) and words. No node name is left in the document.Nine of the ten earlier changes are answered correctly. Role names replace the node names. The failure rule now follows ADR 0276 §3 and ADR 0299 §1-2 (two small points in 5 below). "Waits for … delivered" is now defined as merged into the release branches (lines 111-112). V5 says how the move adopts the existing pair credential, and B2 points to it. The tracker links now say "thirteen issues also link what they wait for through another package"; I counted the same thirteen. K5 is called ADR 0349 clause 17's exception to ADR 0348 §9, and R1 removes it. K1 now gives all eleven registration refusals from ADR 0348's How it is checked, word for word. The counts are right: I measured 117
${bound:}in module.json at ebf7f37, and 12 modules with 34 copy files (lemurs has config/config.toml and a manifest_test). The public half of the operator key is named. One merge window is stated. The group order is only partly answered (finding 4).The dependency table matches every depends-on link of issues 650-685 and 647 (I checked all 37 issues), except for the case in finding 3.
The 67 documents in R2 are correct. I added setting/settings to the Not: line of F4's glossary (hq#351 at
1f6d1b3), and words.py named 68 files: the glossary itself and the same 67 listed in the record.Must change:
1f6d1b3does leave it off ("The word is not yet on a Not: line … checked by review"). Reword F4, including its proof, to say what it actually does.internal/broker", and P3 is left out of the first wave. Yet the table gives P3 as "—", and issue 669 has no depends-on link to 652. Because the tracker decides which packages are ready, P3 could start alongside F3. AddP3 F3to the table and the link to issue 669, or drop the sentence.Should change (these may follow):
after:lines in the pull request descriptions (ADR 0239 §11) and say so, or state only what §12 gives:mesh-host, then mesh-controller, then mesh-catalog.Verdict: APPROVE at
e5b98e3b0aIndependent third review at
e5b98e3b0a, checked against review 239 (and review 236).MESH_CATALOG_DIR=<mesh-catalog at ebf7f37> sh merge-check.shin a worktree of the head passes: records, index, cycle (437 documents) and words (28 retired words, 155 head words, each once). Every table in the document now has the same five-cell rows.All seven findings of review 239 are answered:
21508a7, whose glossary entry says the word goes on the Not: line once issue 683 is done. The proof no longer claims a failure on a planted settings. R2 (line 330) agrees.P3 F3(line 92), and issue 669 has a depends-on link to 652 (added 23:53:05).after:lines (ADR 0239 §11). Under ADR 0249 §2 a declared order outranks every inferred rule, so putting mesh-sdk first and hq last is allowed. The order also agrees with ADR 0239 §12's rules: node-engine before controller, and controller before catalogue.Nothing from review 236 has regressed. No node name is in the document. The counts are the same (117
${bound:}, 12 modules and 34 copy files, 225/88/193 KB). The failure rule still cites ADR 0276 §3 and ADR 0299 §1-2. "Waits for … delivered" is still defined as merged into the release's branches. V5 still adopts the existing pair credential, and B2 still points to it. K5 is still clause 17's exception, and R1 still removes it. The public half of the operator key is still named, and so is the one merge window.I checked the dependency table against the depends-on links of all 37 issues (650-685 and 647). Every link matches the table, directly or through another package. The thirteen issues that also link what they wait for through another package are still the same thirteen: 656, 657, 658, 671, 672, 673, 675, 676, 679, 680, 682, 684 and 685. Issue 669's new link is a direct one, so the count of thirteen stays right.
Should change (these can follow; none blocks the merge):
againare the others walked again; withwalk, the whole walk stops. ADR 0239 §17 also says that a person releases what was stopped once the fix is merged. "The group is walked again" leaves out that step. Suggested wording: "…the fix is a new merge into the release's branch, and a person releases what was stopped (ADR 0239 §17)", with "(a person's stop, with the wayagain, ADR 0299 §5)".internal/broker), so the agent that takes P3 does not read the report alone.