Triad migration evidence of 2026-10-08: run C of the adopter rehearsal at the post-landing main, and T006's root collisions and leg guidance patches - #192

…6-10-09

The 46 lines of gate-b-checklist.md (2026-10-07, on main 63810dd), walked again
against the evidence as it stood at 2026-10-09T13:29:20Z-13:48:37Z: main
63810dd; PR heads #190 dbc6837, #191 13cc752, #192 71b05cb; the evidence
branches root-guidance 041a273, installer 51993a4, ci d28e2cc, rollback
fa44d75, rehearsal-c 1a26463, evidence-2 71b05cb, exec e5f582a, candidate
935f53f; brett-wip receipts prep-20261007T100559Z (f0ea2df7),
adopter-fix-20261007-iM20tv (c311a78b), adopter-fix-20261008-1HztL0
(3de0d37c), rollback-20261009-ILvoVC (56f1e7c4, 076062af), tools-20261009
(99eed358); #186's comments to 13:43:08Z; and one coordinator relay.

Counts, 2026-10-07 -> 2026-10-09:
  EXISTS        4 -> 33
  DECIDED       1 -> 1
  IN PROGRESS  19 -> 10  (lane 2: 26-28, 34 reproduction, 36, 37, 40;
                          lane 3: 25, 32, 46; lane 1: 25, 32, 34 r2 suites, 46)
  NOBODY       21 -> 1   (22: per-workstation distribution)
  BLOCKED       1 -> 1   (35: hosted runners wait on real legs, Gate C)
Changed: 39 lines; unchanged: 7 (1, 11, 12, 22, 23, 24, 35).

Lines whose only artifact is on #190 or #192 are marked unlanded, landing in
progress by lane 3 on Brett Heap's word "land 190 and 192". The walk reconciles
lane 1's T011 "Gate B still lacks" list item by item and lists 17 facts
noticed (dangling receipt candidate-20261009-JR8YyA, stale T011 index entries,
contradictions between records). Facts only; it approves nothing and checks
no task.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…s, so it runs on today's single repository and in the adopted triad, and refuses composed acceptance without all of them (#190)

Tests only. The suite names four roots: code, assembly, spec, and the dependency nested in code. Standalone is the default, and a check whose root is absent skips by name. OPENREPOTOOLS_COMPOSED=1 refuses (exit 4) unless every root is present, the dependency is initialized and every leg sits clean at its gitlink. On today's single repository every existing test keeps its outcome, and none is removed or renamed. The lane suite's last two manual reads are guarded (b50f197), so a code leg with no spec root skips them by name instead of aborting under set -u.

Reviewed LAND by lane openRepoTools-3 (review 5464931249) and by lane 1's independent review (5464938427); the b50f197 delta reviewed LAND by lane 1 (5471080687). CI, tests-macos included, passed on b50f197. Follow-ups, including Copilot's pin-commit finding, are in #193. Landed on Brett Heap's word "land 190 and 192" (2026-10-09), after #192.

Part of #186

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ayload-source branch, so its proofs run against what CI merges.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…t note, narrow "only record", add the census row, cite 1abee1d

Part of #186. These answer the should-fix findings 1–3 and nit 4 of
lane 3's review of #193 items 8–13 (#186 comment 6092925484). Text only,
in the same three records. Every claim was checked again on 2026-10-10.

- root-guidance-2026-10-08.md, section 10 item 6: the 18 tests are now
  described as at 837928a. One dated bullet says what #190 (f87fd5a)
  changed. Through ROOTS.path_for, 17 of them read the assembly or spec
  root, and they skip in a standalone code leg. The line-ending test
  reads the code root's own .gitattributes. This was measured by direct
  calls on main's code paths.
- Runs A and B (both rehearsal records): "only record" becomes "their
  logs survive only in that receipt". Run C's legs keep 558a12b and
  48d98cd as their heads' parents. The splits, 6d52162 and the remotes
  are gone. Checked in --no-local clones of run C's legs.
- root-guidance-2026-10-08.md: section 6's 85-token census joins the
  list of rows that exist only in the document.
- adopter-rehearsal-2026-10-08.md: the plan was replaced on main in
  #192's squash 1abee1d. 1a26463 is kept and named as the branch merge.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…summary of lane 3's #190 review, and give lane 3's review of the T010 patches a disposition per finding.

Lane 3's review of #196 (5481348731) raised three points.
- Finding 4. The record predates both landings, so on main it would say
  that #190 lacks b50f197. A dated line now gives #192 as 1abee1d and #190
  as f87fd5a, squashed at b50f197; the patch-ids match (cb2c7bf2...).
- Finding 5. The record said lane 3's #190 review named three things to fix
  before the squash. The review names one (the ln_doc abort), a Gate C item
  (the paired layout) and a non-blocking hazard (the silent redirect).
- Finding 6. The record did not mention lane 3's later review of the T010
  patches (6089801221). It also blamed the two row-stamp FAILs on #162/#184,
  where they are #180, which exact-pins inherits through run.sh.

Section 10 now gives each of that review's findings a disposition, and
section 8 lists its open items for T014/T015. The two patches are
byte-identical. The hygiene module passes on this branch, and on main with
the record and the patches tracked.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

…ch on the tree the PR will be judged on

origin/main at f87fd5a carries #188's openRepoShape pin (7f84ca4), #189, #192
and #190's explicit test roots. A trial merge was clean; no file is changed
on both sides. The merged tree's counts are in the commits that follow.

Lane: openRepoTools-3
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

… CI patches, installer entry point, restore and rollback, #192's corrections, T018, Gate B status (#196)

* docs: preserve triad migration planning drafts

Preserve the existing OpenSpec and Speckit planning files at Brett Heap’s request during repository cleanup. No migration or adoption is executed.

* docs: refresh triad migration plan after worktree cleanup

* docs: refresh triad plan for October 4 feature updates

* docs: align triad migration with adopted amendment protocol

* docs: record triad preparation and adopter blocker

* docs: record successful triad adopter candidate rehearsal

* Record run C of the Gate B rehearsal at the post-landing main 837928a and regenerate the adoption plan from it.

Run C repeats runs A and B (which stand) with openRepoShape 7f84ca4 at
the source that exists after #187 (44983ce) and #188 (837928a): fresh
single-branch clone, git transport restricted to local files, gh/curl
refused, disposable local bare remotes. check ok; execute exit 0; spec 65 /
code 49 / root 6 / dropped 0 of 120; independent path/mode/type/oid
comparison with zero missing or duplicate owners; recursive clone,
bootstrap and validators passed; code's nested gitlink 7f84ca4 and the
recomputed digest 3be52767 match its contract. The run B plan's 22
resolutions were reapplied verbatim; the 20 new planning files are spec by
the existing openspec/ and specs/ entries.

adoption-plan.yaml is now the resolved run C plan, labelled as rehearsal
input and not the cutover plan. The arrangement is kept host-local as the
read-only input for T006-T010. Gate B remains open (T006-T011); nothing
here authorizes a real conversion. Private receipt adopter-fix-20261008-1HztL0. Part of #186.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* T006's five rehearsal patches are recorded: the assembly front door with two lockstep pin bumps, and the code and spec legs' own front doors, all made against run C's rehearsed split.

Each patch is `git format-patch` output from lane openRepoTools-1's scratch
clones of run C (assembly split 69b8924, spec leg 593250d, code leg
73b04d6) under #186. Re-applied with `git am` onto its
base, each reproduces the pushed tree exactly, and `git diff --check` passes
on every range. assembly-0002 and -0003 are bump-leg.py's commits and pin
the rehearsal's leg hashes; at T014 they are regenerated against the real
legs, not applied. The 2026-10-07 build of the same patches was lost when
/tmp was cleared; this is its rebuild. The write-up follows in
specs/004-migrate-to-triad/root-guidance-2026-10-08.md.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record T010's two workflow patches for the triad: the code leg's standalone tests workflow and the assembly's exact-pin integration job in its own file.

Checkpoint of the CI evidence for #186. The legs do not exist on GitHub yet
(Gate C), so these are patches against the rehearsed run C arrangement: the
code leg's tests.yml at 73b04d6 (its six jobs and trigger policy kept, the
standalone checkout context named) and a new .github/workflows/composed.yml
for the assembly at 69b8924, its own file because validate.yml is a pinned
shape copy. Both `git apply --check` cleanly against run C. The measurements
and local proofs follow in test-roots-and-ci-2026-10-08.md.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record what the triad split breaks in the test suite, the root rules and composed contract PR #190 answers it with, and the local proofs of both T010 workflow patches.

Measured on the rehearsed run C code leg 73b04d6 in three contexts (mounted
in the assembly, standalone, standalone without the dependency): the same 17
hygiene tests fail on files the split moves out of the code leg, and nothing
else in the measured selectors. PR #190 is red at origin/main in 18 tests and
188 passed with nothing skipped on today's layout; CI green per push. In a
composed candidate pinned with bump-leg.py, the validators pass, the composed
job refuses a missing spec leg (exit 1 at its context step, exit 4 from the
suite), and the composed and standalone leg runs fail only on the code leg's
missing .gitattributes (a T006 finding) and two lane-suite stamp assertions
the long run-root path pushes past the row cut. Hosted runs wait for real
legs (Gate C). Part of #186.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* T006's run C rebuild is written up: the collision inventory, 32 repaired links, the leg front doors, the 35-commit provenance correspondence and every bootstrap, validator, OpenSpec and Speckit result, with the assembly patches replaced by leaner ones that respect the root files' length caps.

specs/004-migrate-to-triad/root-guidance-2026-10-08.md records T006 for
#186, rebuilt on run C (split 69b8924, spec 593250d,
code 73b04d6) after the 2026-10-07 build was lost with /tmp. The assembly
patches are replaced: code/tests/test_repo_hygiene.py caps AGENTS.md and
README.md at today's 316 and 486 lines and admits lines only for a rule. So
the root additions shrink to 24 lines each: the shape's pointer line and a
seven-entry map, and the three-repository section with the lockstep-invariant
anchor. The leg patches are unchanged. All five patches re-apply to
identical trees, and `git diff --check` passes. Findings for lane 3, T007,
T009, T010, T014 and T017 are in its section 10. Nothing lands on main from
this task.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record T011's restore rehearsal of the source from the Gate A bundle: 90 checks pass, no captured ref moved except forward, and a dirty tree rebuilds only with the staged and unstaged diffs the October 7 capture does not take.

First commit of specs/004-migrate-to-triad/gate-ab-evidence-2026-10-09.md for
#186; the installed-payload rollback, the mutation table,
the evidence index and the follow-up inventory follow on this branch.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record T011's rollback of the installed payload, the mutation table, the Gate A/B evidence index and the follow-up patch inventory: both rollbacks return every placed file, mode and hook to main's install, the copy-back leaves the receipt stale, and nothing real was written.

Steps 2-4 of T011 in specs/004-migrate-to-triad/gate-ab-evidence-2026-10-09.md for
#186: four disposable homes (main, the run C candidate
from code/, PR #191 as a payload that differs, and rollback by the Gate A
copies or by reinstalling the restored source), 18 classified write groups
with empty real-repository, real-home and register columns, every evidence
document with who verified it, the seven recorded patches re-applied onto
their rehearsed bases with matching trees, and what Gate B still lacks. The
private receipt ID is rollback-20261009-ILvoVC; its commit follows.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Name the T011 private receipt's brett-wip commit, 56f1e7c4, and say why the hygiene selector did not run on this document while the two all-files hygiene checks find nothing in it.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record T007's assembly entry point as a patch and T007/T008's installer evidence for the triad migration.

The root openRepoTools of the assembly cannot land on main before the split,
because its path is the implementation's, so it is kept here as a
format-patch of the commit made in a clone of run C's rehearsed split; git am
on a fresh clone reproduces its tree. installer-2026-10-08.md records the
arrangement relied on, the pin grammar, the one-liner contract and how it
compares with T006's README sentence, the 15-row matrix with its tests and
red-at-base results, the runs, lane 2's installer rows, and what remains for
T009, T010, T014 and T015. Nothing here authorizes a real conversion.

Part of #186.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record the composed triad candidate that carries every Gate B follow-up: its inputs, apply order, commits and trees, its validators, and its install compared with today's main.

The suite runs on the candidate are queued behind the workstation lock and
are recorded in a later commit.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Bring the T009/T010 test-roots record up to PR #190's head dbc6837, with both reviews, the .gitattributes decision and the dated follow-ups.

The record stopped at c139e88. Since then e3a76b5 refuses a leg whose git
status cannot be read and keeps .gitattributes in the fixture legs (Copilot
round 1), and dbc6837 adds T006's assembly-root caps and fails a missing lane
manual by name. Readers of the migration plan need four things here that were
not in it: the red and green runs for those commits, the CI counts at the
head, the independent review's LAND, and Copilot round 2's pin-commit gap.
That gap is reproduced here, and validate-pins.py is shown to close it in the
assembly job.

The .gitattributes decision is now stated outright. It agrees with T006's leg
patch and disagrees with plan.md's placement table, which is a follow-up for
the plan's owner. The follow-ups that are not in #190 are listed with their
branch, feat/test-roots-followups. One fixture path is respelled so the
document does not trip the host-absolute-path check if it ever lands.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Correct the test-roots record: the standalone leg's lane suite aborted on an unguarded manual read, which this record had attributed to two stamp assertions alone.

P3's log shows `ln_doc: unbound variable` at line 6595. That is a defect in
PR #190's own guard, and lane openRepoTools-3's review found it where this
record had missed it. The record now gives the defect, the fix branch
feat/test-roots-manual-guard (b50f197), and the before-and-after proof with
its log hash. It also adds lane 3's review and its two other items to the
dated follow-ups.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record the candidate's state at the coordinator's checkpoint: no suite run has finished, the composed run was stopped at about 20% and must be rerun, and the CI numbers it will be compared with.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* T018 disposable half: feature selection, OpenSpec and park/resume/status in a triad fixture (#186)

Evidence of 2026-10-09 from the composed candidate (assembly 3bd95a7, spec
9e1d006, code f0af7d9, nested openRepoShape 7f84ca4), copied into fresh bare
remotes and driven from three temporary homes under env -i, a disposable
agent root and workspace record, file-only git and a private network
namespace. Selection resolves the same root and feature directory from the
root and both legs. OpenSpec answers only from the spec leg. park commits and
pushes one WIP per leg and leaves the root and its gitlinks alone. resume
refuses a root that the bootstrap itself dirtied, and silently drops a parked
nested gitlink move. Every refusal is relayed verbatim; nothing was forced.
No task is checked complete.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Dated corrections to #192's run C and T006 evidence: #193 items 8–13

Part of #186. Text only, in three Markdown records under
specs/004-migrate-to-triad/. No patch file, plan or YAML is touched.
Every figure was measured again on 2026-10-09.

- adopter-rehearsal-2026-10-08.md
  - Item 8 (review 5464893559 finding 5): one dated block says where the
    records are now, what run C superseded, and that run B's plan stays
    at 63810dd. It records the 2026-10-08 restart that removed runs A
    and B's repositories, and the openspec 1.2.0 observation. openspec
    1.13.1 now exits 0 on 837928a and on 71b05cb.
  - Item 10 (finding 17): `check` reads the clone's local main, not
    HEAD. This plan gives plan-stale, exit 1, against any newer main.
    Reproduced both ways.
  - Item 12 (finding 6): the "no ensurepip" cause is withdrawn. The
    refusal is real, and its cause is not established.
- root-guidance-2026-10-08.md
  - Item 9 (findings 8 and 9): 18 front-door test functions (17 + 1),
    not 17. The command test stays red after code-0001. The "23 in 9"
    total disagrees with its own per-file list (26 in 11), and which is
    right is not established. Also :6, and :195's consumer-reads is on
    main.
  - Item 11 (finding 16): no receipt is cited. The record now says what
    the patches reproduce from public inputs, and which rows exist only
    in the document.
- adopter-rehearsal-2026-10-07.md
  - Item 8: one dated line. The false "On this branch" sentence is in
    this file.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record v3 of T007's entry-point patch, which checks the gitlink before it runs any pinned implementation, and the network needs T014 should document.

Lane 3's review of the v2 patch (T3) required the remote path to confirm
the gitlink at the pin's submodule_path before it fetches or runs the
implementation the pin selects. v3 (6acc147 on 69b8924) does that, and git
am on a fresh run C clone reproduces its tree 87782af. The runs table rows
for the round-2 head and CI are still to be added.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Refresh the T011 record's index, inventory and Gate B list to origin at f87fd5a, and correct what lane 3's review and lane 2's receipt check found

Answers lane 3's review of fa44d75 (#186 comment 6082139424) and folds in lane
2's verification of receipt rollback-20261009-ILvoVC (#186 comment 6082370012).
Only specs/004-migrate-to-triad/gate-ab-evidence-2026-10-09.md changes. Every
refreshed passage says Refreshed or Corrected and keeps the 01:1xZ text it
qualifies.

- F11: no stage changed a hook, so the hook rollback is vacuous. The Gate A
  backup holds no settings.json, and RESTORE.md's copy-back restores only
  installed/. 849592b's subject overclaims hooks.
- F15 and F18: sections 4 to 7 re-read origin at 20:29Z-20:37Z, with main at
  f87fd5a after #192 and #190 landed. They now record T007's entry point on
  the installer branch at 51993a4, the candidate at 935f53f, every review
  since 01:13Z, and an inventory that maps each named follow-up to a patch,
  a PR, an owner or a deferral.
- F4-F9, F12-F14, F17, F19, F20, F22 and lane 2's corrections:
  - the restored bundle's identity, and the 90 checks by category against
    #189's 97;
  - the two near-tautological check families, and main's 70 files plus one
    symlink;
  - "live" meaning the shared checkout at 00:44:08Z;
  - 10 installs, not 12, and no placed file added or retired;
  - 076062af and the final snapshot;
  - T010 trees with no earlier record;
  - GIT_OPTIONAL_LOCKS, the copied capture block, and nested repositories
    rather than caches.

Part of #186.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Correct three rows of the refreshed T011 record from lane 3's review of c1283b5: the candidate branch's creation sha, the two #190 follow-up branches' creation and first move, and the owner of run C's ninth docs/ file

From lane 3's review (#186 comment 6089982753), N1 to N3, each checked at
its source:
- N1: 004-migrate-to-triad-candidate was created at b21c511 at 01:58:43Z,
  and b21c511 -> 935f53f was pushed at 02:43:24Z (GitHub activity API).
- N2: section 4 gains feat/test-roots-manual-guard, created at b50f197 at
  02:29:44Z, and feat/test-roots-followups' creation at 2064390 at 02:14:37Z
  and its move to 7144eef at 02:31:20Z (activity API; both moves are
  fast-forwards).
- N3: the plan's unnamed ninth docs/ file is tests/test_repo_hygiene.py.
  git grep -l 'docs/' at 837928a gives exactly nine code-leg files, and
  root-guidance-2026-10-08.md:191 routes that one to T009/T010.

Part of #186.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Answer lane 3's low findings N4 to N9 on the refreshed T011 record: collision rows, the dropped-check arithmetic, the copy-back's limits, deleted refs, the refresh's real time, and the verbatim run's public citation

From lane 3's review of c1283b5 (#186 comment 6089982753), each checked at
its source:
- N4: root-guidance-2026-10-08.md:69-76 holds the collision and shape/ rows
  (AGENTS.md is :69; CLAUDE.md, :70, is unchanged), so "resolved in the
  patch", not "merged".
- N5: October 7's verify.log:13 is one check of the 16 per-worktree HEAD
  refs, and main's status check moved to the singles (10-restore.log:160):
  12 - 2 + 1 + 11 = 22 and 85 - 17 = 68.
- N6: the copy-back route restores the 29 files and the receipt only, with
  no settings.json and no hook entry.
- N7: a captured ref moved forward or was deleted with its commit still on
  GitHub; re-checked against ls-remote at 22:05:16Z (15/9/9/0).
- N8: c1283b5 was committed at 20:38:20Z, so the hygiene line and the
  header's refresh window say 20:3xZ.
- N9: the public comment marks as inferred only its conclusion; the
  verbatim run's log is host-local and unpublished, so the record now says
  no public citation exists for the run itself.

Part of #186.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Gate B status walk of 2026-10-09: the 46 checklist lines (MET 30, PARTLY 15, OPEN 0, AFTER-GATE-C 1), T001-T019, the Gate C open questions with proposed wording, and a line-by-line reconciliation with lane 2's walk

specs/004-migrate-to-triad/gate-b-status-2026-10-09.md walks
gate-b-checklist.md against origin as fetched at 22:07:06Z and #186's
comments through 22:06:49Z. It approves nothing and changes no task
status.

- Section 1: each line's status, evidence, what remains and owner, and
  which lines moved since lane 1's unpublished first pass.
- Section 2: lane 2's walk, the same path on 004-migrate-to-triad-reads
  @ a7a8e00 (a cross-check that goes into no PR), line by line. It
  differs in substance on 22, 2, 12, 18, 19 and 21. It also answers
  lane 2's review of the first pass (6088829392) item by item.
- Section 3: T001-T019 against the evidence.
- Section 4: questions a-e for Brett, each with text to approve:
  - (a) openspec 1.2.0 fails migrate-to-triad and 1.6.0 passes;
  - (b) "account move";
  - (c) .gitattributes in plan.md against tests/conftest.py on main;
  - (d) hosted CI before the legs exist;
  - (e) e1-e10, including the six-commit openRepoShape drift and the
    post-cutover one-liner's API and jq dependency.
- Section 5: the blockers in dependency order.
- Section 6: the commands.

Private receipts are named by id only, and no host path appears.

Part of #186
Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record the runs of #191's review rounds, the strict composed result at 6568f08, and CI on each head, in the installer evidence.

The runs table had two rows still waiting at the first checkpoint. It now
records what each review round changed and every proof, with which ran under
the lock and which ran unlocked under the coordinator's rule for scoped
proofs. It also records the strict composed run of root v3 with the head's
whole code leg, where the only failure is the code leg's missing
.gitattributes (T006), and CI per job on 6568f08.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record the r2 candidate's three suite runs and classify every failure: two candidate defects with owners (the r14 KeyError fixed in #191, #193 item 14) and four environment classes, each shown the same on main or in isolation.

r2 is code 4dbcf6c, spec 9e1d006 and assembly 5d92962, built from #190 at
b50f197 and #191 at 13cc752. Composed: 1411 passed, 5 failed, 0 skipped.
Standalone: 1389 passed, 3 failed, 24 skipped. No submodule: 1168 passed,
6 failed, 242 skipped. The install from the assembly root differs from
main's only in the implementation it hands over to. The composed rerun at
#191's final head 6568f08 with T007's v3 entry point is still owed.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Lane 3's follow-ups on 096de93: name 837928a and #190 in the T006 test note, narrow "only record", add the census row, cite 1abee1d

Part of #186. These answer the should-fix findings 1–3 and nit 4 of
lane 3's review of #193 items 8–13 (#186 comment 6092925484). Text only,
in the same three records. Every claim was checked again on 2026-10-10.

- root-guidance-2026-10-08.md, section 10 item 6: the 18 tests are now
  described as at 837928a. One dated bullet says what #190 (f87fd5a)
  changed. Through ROOTS.path_for, 17 of them read the assembly or spec
  root, and they skip in a standalone code leg. The line-ending test
  reads the code root's own .gitattributes. This was measured by direct
  calls on main's code paths.
- Runs A and B (both rehearsal records): "only record" becomes "their
  logs survive only in that receipt". Run C's legs keep 558a12b and
  48d98cd as their heads' parents. The splits, 6d52162 and the remotes
  are gone. Checked in --no-local clones of run C's legs.
- root-guidance-2026-10-08.md: section 6's 85-token census joins the
  list of rows that exist only in the document.
- adopter-rehearsal-2026-10-08.md: the plan was replaced on main in
  #192's squash 1abee1d. 1a26463 is kept and named as the branch merge.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record #191's two Copilot rounds and the final head 8246983 in the installer evidence.

Round 1's two fixes and the one finding kept, with the reason; round 2's
body correction and the test that shows the behaviour; and a strict
composed run of the module's composed and local rows with root v3 and the
final head's code leg.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Date the "4 now pass only because they read the leg's own files" bullet: since #190 three of the four skip in a standalone leg (lane 3's 5481348731, item 8 leftover)

Lane 3's review 5481348731 on #196, item 8, left one [low]: root-guidance's
bullet "4 now pass only because they read the leg's own files" sat right
under the bullet that says, since #190, the 17 document-reading tests skip in
a standalone leg. The two contradicted each other back to back.

The four are the ones that passed after code-0001 at 837928a. At main
(f87fd5a) three of them (test_agents_md_is_short_enough_to_be_read,
test_readme_is_short_enough_to_be_read, test_claude_md_points_at_agents_md)
read through ROOTS.path_for and skip, with or without code-0001's own
AGENTS.md, CLAUDE.md and README.md. The fourth,
test_the_root_carries_the_line_ending_rule, reads the code root's own
.gitattributes: FileNotFoundError before code-0001, a pass after it.

Measured on 2026-10-11 in a scratch standalone leg built from f87fd5a (spec
trees and front-door files removed): before code-0001, 3 skipped and 1
failed; after it, 3 skipped and 1 passed.

The bullet now says what held at 837928a and carries one dated correction.
Nothing else in the file moves. Part of #186.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Date the T010 record against #192's and #190's landings, correct its summary of lane 3's #190 review, and give lane 3's review of the T010 patches a disposition per finding.

Lane 3's review of #196 (5481348731) raised three points.
- Finding 4. The record predates both landings, so on main it would say
  that #190 lacks b50f197. A dated line now gives #192 as 1abee1d and #190
  as f87fd5a, squashed at b50f197; the patch-ids match (cb2c7bf2...).
- Finding 5. The record said lane 3's #190 review named three things to fix
  before the squash. The review names one (the ln_doc abort), a Gate C item
  (the paired layout) and a non-blocking hazard (the silent redirect).
- Finding 6. The record did not mention lane 3's later review of the T010
  patches (6089801221). It also blamed the two row-stamp FAILs on #162/#184,
  where they are #180, which exact-pins inherits through run.sh.

Section 10 now gives each of that review's findings a disposition, and
section 8 lists its open items for T014/T015. The two patches are
byte-identical. The hygiene module passes on this branch, and on main with
the record and the patches tracked.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Gate B walk: lines 20, 29 and 31 are PARTLY by the walk's own legend, and the walk agrees with the candidate record at 16e6ea4 and slice B at 40f26f7, which #196 carries beside it (MET 27, PARTLY 18, OPEN 0, AFTER-GATE-C 1)

This answers lane 3's review 5481348731 of #196, findings 12 and 14.

Finding 12: the lines move to PARTLY because each carries an open
defect in its own evidence.
- Line 29: #193 items 2-7 and 15. It now also cites lane 3's verdict on
  b50f197 (6089801221 (a)).
- Line 31: #193 items 4 and 15, both fixed on #195, which is open.
- Line 20: the record's "23 in 9" count, which slice B leaves not
  established.
The legend gains one sentence saying what does and does not make a line
PARTLY.

Finding 14: the walk now agrees with the two records.
- The candidate record publishes r2 and classifies F1-F6. Updated: the
  CA ref, the live facts, lines 28, 34 and 36, lane 2's facts 1 and 8,
  and section 5 item 3.
- Lane 3 reviewed slice B as LAND (6092925484). Updated: the EF ref, the
  live facts, lines 3, 17 and 18, and the T006 row.
- Question (a) names the host for openspec 1.13.1.

One dated entry in the header says what changed and why. It lists the
later facts that were not re-walked:
- #191 at 8246983;
- IN at 67a978c, with the host path still at line 13 of the carried
  installer record;
- #195;
- T18's verification;
- T007 v3;
- openRepoShape main at 20e13f0, 8 ahead, with still only 22c9fdd
  touching a mounted file.

Part of #186
Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Point the three "no review verdict posted" rows of the T011 record at lane 3's verdict, 6089801221 (lane 3's 5481348731, item 7)

Lane 3's review 5481348731 on #196, item 7, left one [low]: gab:436, :443 and
:513 still say "no review verdict posted; lane 3's reviewer ... has resumed"
for the T010 patches and the follow-ups. Lane 3's verdict, #186 comment
6089801221, was posted at 21:47Z on 2026-10-09, after this record's origin
read of 20:29-20:30Z.

Each of the three rows now carries a dated correction that points at it and
says what it concluded for that row. The original text and the read time stay
as written.
- The `004-migrate-to-triad-ci` row and the T010 inventory row: `d28e2cc` is fit
  for evidence-3, both patches apply, and #180 reaches the composed job
  through `run.sh` (its finding 13).
- The `feat/test-roots-followups` row: `01ccc86` is fit for a PR onto main
  after one edit, the 340 to 347 assembly cap (its finding 7).

Part of #186.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Name the host behind "the bench's openspec is now 1.13.1" in run C's record, and give lane 3's 1.2.0 and 1.6.0 reading (lane 3's 5481348731, item 9)

Lane 3's review 5481348731 on #196, item 9, left one [low] on
adopter-rehearsal-2026-10-08.md:53: "The bench's `openspec` is now 1.13.1."
That depends on the container. On lane 3's host at 00:1xZ on 2026-10-11,
`which -a openspec` lists a user-level npm install first, at 1.2.0, and the
system install second, at 1.6.0. `validate migrate-to-triad --strict` then
exits 1 under 1.2.0 (the SHALL or MUST error) and 0 under 1.6.0.

The sentence now names the bench, lane 2's, and the dates of the two readings
(6088829392 item 8 on 2026-10-09; 6103482530 on 2026-10-10). A dated
correction records the first wording, says the version depends on the
container, and gives lane 3's reading with its exit codes. It points at lane
1's walk, section 4 (a), which records the same split.

Locations are described, not written as paths. gate-b-status-2026-10-09.md:266
repeats the claim and is lane 1's file, so it is left for its owner.

Part of #186.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Correct root-guidance:432 after lane 2's check of 1578f99: the fourth test reads the same file at 837928a and since #190, and the note records the line's first wording (lane 3's 5481348731, item 8)

Lane 2's checker verified 1578f99. Every measurement in it holds, but its
wording on line 432 needed a fix.
- test_the_root_carries_the_line_ending_rule reads the code root's own
  .gitattributes the same way at 837928a and at f87fd5a. #190 changed only its
  docstring. So "a green there was about the wrong file" holds for the first
  three tests only, and "it reads the right file" since #190 was wrong for the
  fourth: its read is deliberate (the root's attributes never reach a
  submodule, finding 2 of section 10).
- The line did not record what it first said. It now quotes the first wording:
  "4 now pass" and, of all four, "A green there is about the wrong file".
- "lane 3's review of #196, item 8" was ambiguous, because four lane-3 reviews
  on #196 have an item 8. The note now cites lane 3's 5481348731 by id.

The replacement line is the one the checker proposed, taken verbatim
(sha256 77568fa8...64d). Nothing else in the file moves. Part of #186.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Record r3 of the Gate B candidate: #191 at its final head 8246983 with T007's v3 entry point, the install modules green in composed mode, F1 gone, and the install still differing from main only in the implementation it hands over to.

Code 9c11276, spec 9e1d006, assembly 3891943; every git am exited 0 and both
pins were regenerated with bump-leg.py. The four install modules, composed
and isolated with SIGINT at its default: 281 passed, 0 failed, 0 skipped.
The intermediate r3a at 6568f08 (278 passed) is recorded beside it. The full
composed suite is redone at the frozen source for Gate C.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Close W22's findings 2-4 on 83bcec7 in the T011 record: date the section 7 leftover, restore "before the composed job is relied on", name the branch at :443 (lane 3's 5481348731, item 7)

W22's adversarial check of 83bcec7 (lane 2) found three leftovers of lane 3's
5481348731 item 7, in specs/004-migrate-to-triad/gate-ab-evidence-2026-10-09.md.

Finding 2 [low]: section 7 item 4 ("Refreshed 2026-10-09T20:30Z") still said
"The T010 patches: no verdict yet; lane 3's reviewer has resumed (#186 comment
6088703549)" with no pointer, while :436 and :513 already point at 6089801221.
It now carries the same dated correction, 2026-10-11 like its siblings: the
verdict, #186 comment 6089801221 (2026-10-09T21:47Z), came after the record's
origin read; fit for evidence-3, with #180 to settle before the composed job is
relied on (its finding 13). The original words stay as written.

Finding 3 [nit]: :436 and :513 said "One thing to settle first" and "with #180
to settle first". 6089801221 (c) says "One thing must be settled before the
composed job is relied on". Both rows now carry that condition, so it attaches
to relying on the composed job and not to "fit for evidence-3".

Finding 4 [nit]: :443 said "It answers the code-half part of #193 item 2" right
after a sentence that begins "It is #186 comment 6089801221", so "It" read as
the comment. It now says "The branch `01ccc86` answers ...".

One file, +7/-4. Part of #186.

Lane: openRepoTools-2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Correct the installer record for lane 3's DO NOT LAND on #196's installer files: no host path, a T008 table of the tests at the final head, and a disposition for every open item.

Lane 3's review 5481368376 asked for six corrections and noted one minor
point; all seven are made here. The record no longer names a host-absolute
location for run C's README (finding 0). It says what v3 closes, the
mismatched pin, and what it does not: a lockstep-consistent pin at code
from before #191, which is now a named T015 item with a capability check
proposed for #193 (finding 2). The T008 table now lists every test in
tests/test_install_payload_source.py at 8246983, which is main a74c7fc
(52 tests, 109 cases). Its base column comes from that module run against
c4864ac's installer (finding 3). The record names 8246983 as the head,
with its CI and the landing on main as a74c7fc on 2026-10-13 (finding 4).
The wip_shape_ref disposition gives the F2 fallbacks (finding 5). Each of
lane 2's nine open items in #186 6090038382 is fixed, accepted or given
to a task (finding 6). The minor point: "five things" and both reviews
of 2411f40 (finding 7).

The patch is unchanged.

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Correct the installer record against lane 2's check at 8246983: the walk residue's reason, the 200-garbage refusal, row 7's cases, CI on cddd1f7 and the measured holds.

- Item 1: the code reads the gitlink at the pin's submodule_path and refuses
  where none is recorded (24a, 24c), citing lane 3's 5481368124 finding 5 and
  its cheaper tightening, and finding 7 on #191's body.
- Item 4: a 200 that is not GitHub's refuses where main installs, like
  item 3; disclosed here, since #191's body does not name it.
- Item 2: the adopted missing-ref and missing-repository rows (19c-f, 19h).
- Item 7: the patch's header and commit message overstate; it stays.
- Item 9: row 7's [single] case is an adopted ref moving to a commit with
  no pin, which is refused; no test freezes a single repository.
- The section's opening cites lane 2's measurements at cddd1f7 (#186
  6103618033) and 8246983 (register note 2026-10-13T18:22:15Z).
- The runs table gains CI on cddd1f7 (run 38096352239).

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Add dated 2026-10-13 lines to the candidate record: #191 landed as a74c7fc with r3's installer blob, what r3 covered and what is still owed, where the answered gap is recorded, and F2's decision with #195's status.

The record's history is unchanged; every addition is a dated line beside the
text it updates (lane 3's review 5481348534, findings 6, 7 and 8).

Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Gate B walk, dated 2026-10-13: #191 landed as a74c7fc, what #196's final refold carries, and the hold for openRepoShape v1.0.0

These are dated lines only; no earlier text is rewritten. No status changes:
MET 27, PARTLY 18, OPEN 0, AFTER-GATE-C 1.

- #191 landed on main as a74c7fc, the squash of 8246983. Recorded at the
  header's #191 line and at question d's #191 CI line.
- What #196's final refold carries, each from the earliest head that holds
  its fixes, with #196's "Input heads carried" table named for the exact
  heads:
  - the installer record from 7668c51 on: the host path at its line 13
    is gone, and lane 3's 5481368376 is answered; 7428e78 adds lane 2's
    check;
  - slice B from 2c040d6 on: lane 3's 5481348731 items 7-9, with the
    T011 record now taken from slice B;
  - the candidate record from e81100c on: r3.
  Recorded in the header, the refs table, section 2's item 16 and
  section 5's item 4.
- The hold for openRepoShape v1.0.0, on Brett Heap's word relayed by
  lane 3 (#186 6152410188), with its after-the-tag order: the pin-bump PR,
  a fresh rehearsal and plan at v1.0.0, then the cutover. #196 is not
  held.

Part of #186
Lane: openRepoTools-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brett Heap <service@opensoft.dev>