feat(plans): plan 37 settling arc — I/J/K close

[darko-mijic]

Story

PR #21 published plan 37 as a mid-arc draft, not for merge as a close. Waves 0–2 had already adopted 46 of 58 leftover bindExample sites onto generated registrars and refused 12 with recorded reasons. The eight drift-alarm packets were prepared and unapplied. The context-bundle measurement was frozen before any plan-37 session was counted.

This PR is the close. It executes briefs I–K from plan 36 and marks plan 37 EXECUTED.

Did the frozen five-adapter registrar hold past the first tranche? Yes, with honest refusals. 46 sites adopted. 12 stayed on bindExample because the authored point is incomplete or the example needs a second product call. The interface was not loosened. Friction is in the close record, not in src/testing.

What happened to the eight implemented ∧ ¬ready alarms? The owner ratified five Specs to ready. Three stayed defined with a written blocker where the alarm still points. Nothing was demoted to silence the count.

Is a context bundle worth building? Not from this corpus. Eligible sessions: 12. Qualifying episodes: 0. Verdict is STAND-DOWN (unmet). The plan-35 re-entry trigger stays. No bundle, projection, verb, or reader accessor was built.

Adoption itself already landed in #21. Review here is the human statements, the K verdict, the independent review, the accepted commit-boundary deviations, and the status flip.

Tour

Close record

plans/37-adoption-tranches-drift-maturation-and-bundle-measurement.md now carries the I ledger, the eight-row J table, the K stand-down, and the re-derived close measurements. Operational tracking: .omo/plans/plan-37-settling-arc.md.

Independent review

.omo/evidence/task-19-plan-37-settling-arc.md is the plan-32-mold register. One finding, P37-R1, ACCEPT, CLOSED. The reviewer re-ran current-tree measurements rather than inheriting Todo-18 numbers.

Accepted historical deviations

.omo/evidence/plan-37-historical-boundary-disposition.md is the four-row set. F1 and F4 rejected commit placement, not the live tree. History was not rewritten. The 26-commit, 134-path census is the completeness proof.

Owner ratification

.omo/evidence/plan-37-j-packets/RATIFICATION-BUNDLE.md is the batch the owner filled. Todo 16 applied only the five READY rows.

Ratified decisions

Name	Ruling	Record
Brief J READY set	Owner-ratified ready on five already-implemented Specs. Three stay defined. No demotion.	spec:carrier.markdown-authoring, spec:extraction.claim-taxonomy, spec:model.pack-aggregate, spec:model.relations, spec:model.spec-sections
P37-R1 Brief-I exception	Brief I may enumerate the exact 56 tracked generated registrar siblings in contract-dependent-suites.mjs, and may disable @typescript-eslint/no-redundant-type-constituents only inside the existing contract-dependent-suite ESLint override, so clean-clone lint/check discovery works before generation. No other check or config change.	review register in plans/37
F1-COMMIT-BOUNDARY / F4-B1	Generated bounded-parity sibling landed in commission 1dae853; authored activation landed in e081071. Accept the split.	disposition
F4-B2	Anchor-pin batches 4a451e2, 0b098a3, 07098f3 landed standalone instead of riding adoption commits. Accept the batches.	same
F4-B3-K-CLOSE-RIDE	K definition/tooling landed in c265f2d / 9218be9 / dfb899b, not on close 8e6a86b. Accept the dedicated placement.	same
F4-B4-TODO8-Y	Todo-8 five-REFUSE evidence rode mixed Wave-1 checkpoint 065a18f instead of its own Commit:Y lane. Accept the checkpoint placement.	same

DEFINED rows that stay on recipe 2, with the blocker the owner kept:

Spec	Blocker
spec:consumers.projections-model	Impact-graph and measurement work remain unlanded. Plan 35 kept defined.
spec:extraction.regenerability	Quoted thresholds lack their measurement artifact.
spec:model.core-model	Blocking enrichment-lifecycle question remains open.

Try it

# Full repository gate. Todo 20 ran this twice consecutively at close, both exit 0
# (838 passed, 1 skipped), then flipped statuses in 85a338a.
npm run check

# Recipes 1, 2, 8. Lift each fenced body from docs/agent-surface/recipes.md
# and pipe it. Verified at 85a338a against this branch:
pnpm --silent sdp:q --json   # recipe 1 body
pnpm --silent sdp:q --json   # recipe 2 body
pnpm --silent sdp:q --json   # recipe 8 body

Expected at this HEAD, re-run in the F5 session rather than copied from Todo 18:

• Recipe 1: total: 0, excludedReadyExamples: 66, excludedReadyDecisions: 31, excludedWithoutVerifier: []
• Recipe 2: total: 3, the three DEFINED rows above, each floorReached: ready, firstUnmetClause: null
• Recipe 8: errors: 0, warnings: 5, byValidator: { "honesty/gaps": 5 }, subjects equal the five READY Specs

The five warnings are informative. Ready never requires a verifier. test/cli.test.ts pins them.

Feedback wanted. Are any of the 12 refusals evidence that the frozen registrar point is wrong, rather than useful selection pressure? For core-model, regenerability, and projections-model, do the recorded blockers still earn defined? Is zero qualifying K episodes a catalog-usage problem, or proof that sessions are not hand-assembling the same slice?

Numbers, re-derived at close

Re-run the recipes rather than inheriting these. Todo 18 measured them twice with identical output. This F5 session re-ran recipes 1, 2, and 8 at 85a338a and got the same JSON.

156 Specs · 1 Pack · 157 anchors → 314 nodes · 660 edges. Validate: 0 errors, 5 honesty/gaps warnings on the five READY Specs.

Measurement	Commission	Close	Delta
Recipe 1 backlog	empty, exclusions 66 / 31	empty, exclusions 66 / 31	unchanged
Recipe 2 drift alarm	8 Specs	3 Specs	−5, the READY set
Recipe 8	0 warnings	5 honesty/gaps	+5, same five subjects
bindExample census	58	12	−46 adopted
Tracked generated siblings	10	56	+46
Owed generated/registrars.json files	68	68	10 + 46 + 12 = 68

Upcoming work

After merge, resume the paused gen-1 skills parity plan (.omo/plans/sdp-skills-gen1-parity.md). It is blocked on this closure. This PR does not start that work.

Event-triggered deferrals still re-enter only on their recorded triggers: MCP (D6) on a concrete caller with its own verb boundary and contract; Spec Studio on the package-home ruling plus a reader Design Review does not serve; reference projection / structural-edge Mermaid on a named unserved reader. The context-bundle trigger is the same plan-35 rule K just failed to meet.

Durable refusals, unchanged from plan 36: default-carrier flip; Gherkin kind expansion, DocStrings/DataTables, Gherkin Packs; an implements slot; frozen-registrar-interface loosening; engine-side adopter execution; Scenario Outlines; re-specifying shipped projections; bySymbol / impact graph; the .sdp.gherkin suffix; new query verbs; the E2 placement ruling; the E3 MCP non-ruling.

The 12 adoption refusals are equally deliberate under the current freeze. Re-entry needs a future ruling on point completeness or multi-call examples.

Lineage: PR #21 · PR #20 · PR #19 · PR #16 · PR #10

Greptile Summary

This update records plan 37 completion, aligns five self-hosted Specs with their validated readiness, and updates graph and CLI expectations for their informative verifier-gap warnings. The checked corpus contains no extraction or readiness-floor errors; all five promoted Specs derive ready, and the checked CLI reports the expected five non-gating warnings.

Confidence Score: 5/5

No blocking failure remains.

The promoted Specs clear their readiness-floor requirements, and the exercised graph-validation and CLI paths agree with the updated expectations.

T-Rex Logs

What T-Rex did

• Ran the authored Node audit through the built extractor, graph validator, readiness-floor evaluator, and readiness derivation APIs for spec:carrier.markdown-authoring, spec:extraction.claim-taxonomy, spec:model.pack-aggregate, spec:model.relations, and spec:model.spec-sections, producing a 314-node, 660-edge graph with no extraction findings and five informative verifier-gap warnings, and confirming all promoted specifications derived readiness.
• Verified the five promotions show derivedReadiness: ready, an empty readiness-floor list, and five informative warnings that state there is no resolving verifier and are informative only.
• Checked the CLI view of the promoted corpus and ran the CLI tests, reporting zero validation errors and the same five warnings, with test/self-hosting-graph.test.ts passing 26 tests and test/cli.test.ts passing 80 tests.

_{Ran code and verified through T-Rex}

_{Reviews (2): Last reviewed commit: "docs(skills): tighten shipped SDP guidan..." | Re-trigger Greptile}