Scrub owner quotations from the new-WSL records: paraphrase the canonical inputs, regenerate, and carry a dated paraphrase in the relay contract - #1161

Scope

• What this PR changes, in one or two sentences: The owner-batch relay contract of the new-WSL definitive defaults now carries a dated paraphrase of the owner's decision, never the owner's words. The three relaying decision records and the layer-consensus record are scrubbed, and Paraphrase owner quotations in the new-WSL layer-consensus record and regenerate the manifest and handbook #1159's temporary exemption is removed. The command center ruled this P1 on 2026-10-15 (12:19Z: immediately after Paraphrase owner quotations in the new-WSL layer-consensus record and regenerate the manifest and handbook #1159; 12:38Z: this design).
• Base commit: 30542e705 (main; the branch was rebased once onto it on 2026-10-16, see Commit 10; the earlier base was 420452059). This PR lands the scrub in five commits. The first two are a3d53be4 (the paraphrase of the canonical inputs and the regenerated manifest and handbook, formerly Paraphrase owner quotations in the new-WSL layer-consensus record and regenerate the manifest and handbook #1159) and c95453a1 (the relay contract); the three after them follow the command center's reads, below. On the command center's direction of 2026-10-15T13:01Z, Paraphrase owner quotations in the new-WSL layer-consensus record and regenerate the manifest and handbook #1159 is closed as superseded by this PR, and Raise the Claude Code floor to 2.1.300 and record its source commit as the run pin #1128 rebases onto it.
• Lane: lane:foundation
• Owned paths touched:
  • evidence/artifacts/new-wsl-definitive-defaults-20261001/assemble_manifest.py, plus the regenerated definitive-manifest.json;
  • evidence/artifacts/new-wsl-layer-consensus-20261002/consensus.json;
  • three relaying decision records: docs/decisions/2026-10-04-token-full-stack-owner-default.md, 2026-10-04-2604-e2e-fix-wave.md and 2026-10-04-final-architecture-round2.md;
  • docs/new-wsl-handbook.json and .md (regenerated), plus the handbook receipt;
  • tests/test_new_wsl_handbook.py and tests/test_new_wsl_definitive_defaults.py;
  • manifests/evidence.json.

Commit 1 (a3d53be4, formerly #1159): the canonical inputs and the generated files

What changes:

• Canonical input: consensus.json had six owner quotation spans in five fields. Each becomes a dated, attributed paraphrase ("the owner, (paraphrased): …", they/them), dated from the record's own context. The agent-messaging amendment's decision, text, arms and experiment contract are unchanged.
• Regenerated, never hand-edited:
  • assemble_manifest.py wrote the manifest (5 spans; --check exit 0), and render_tables.py --check exits 0.
  • build_new_wsl_handbook.py --write, then --check, regenerated the handbook (5 spans). That covers the read's four at base :851, :933 and :934, plus one more at the context-supply record.
  • The handbook receipt has new current outputs, a regenerations entry and a previous_outputs entry.
• Regression: OwnerQuotationTests, in tests/test_new_wsl_handbook.py, uses invented prose only. It rejects any field that holds the owner's words, and any sentence that attributes a quoted span to the owner, in the canonical inputs and the generated handbook. Dated paraphrases and protocol or contract text pass. The existing owner-quote check in tests/test_new_wsl_definitive_defaults.py now uses the same rule.

Its temporary exemption (removed by the second commit): three owner-verbatim fields stay in consensus.json: the wave3, wave4 and wave5 authority.verbatim.

• They are the relay contract, not copies. assemble_manifest.py requires them (OWNER_AUTHORITY_FIELDS), and check_owner_batch refuses a batch whose hash-bound relaying decision record lacks them.
• They never reach the manifest or the handbook.
• The regression exempts exactly those three paths, and only while the assembler requires the field. The exemption's text names the follow-up branch claude/cc-native-practice-relay-contract-scrub-20261015.
• The command center ruled at 2026-10-15T12:19Z that this exemption is temporary. The follow-up comes immediately, at P1, as its own PR. It changes the relay contract to a dated paraphrase plus a hash binding to a private record, scrubs the three fields and the relaying decision records, and removes the exemption.

Commit 2 (c95453a1): the relay contract

The contract:

• Fields: OWNER_AUTHORITY_FIELDS swaps verbatim for two fields:
  • paraphrase: dated, they/them, beginning "the owner, <date_utc> (paraphrased): ";
  • relay_record: the command center's private record that holds the exact words, by name and location only.
• check_owner_batch keeps its sha256 binding to the committed relaying decision record and checks that the record states the paraphrase. Anything digest-like in relay_record is refused. A hash of the private record's bytes would break digest custody, and a short owner sentence is open to guess-and-check (the command center accepted this correction to its first wording). The exact words stay only in the private record; a local check outside CI can compare them.
• Scrubbed:
  • the wave3, wave4 and wave5 authority fields;
  • the three decision records' block quotes;
  • a second owner quotation in the round-2 record (a 2026-10-03 answer).
• Re-bound and regenerated: the three batches re-bind to the records' new sha256 values. The manifest and the handbook are regenerated through their builders, and the handbook receipt gains a regeneration entry.

Commit 3 (9371e2ad): repo-wide, from the command center's FULL read at c95453a

The CC ruled the split at 2026-10-15T13:25Z: this commit now, and burn-down PRs by area after it. Owner text is cited here by location only.

• P1: the named records. Each change is a dated, they/them paraphrase with a supersession note beside it:
  • wave2-records.json owner_decisions[1], where verbatim becomes paraphrase;
  • claude-request.md (the opening paragraph and item 3);
  • docs/decisions/2026-10-04-repository-quality-rule.md;
  • repository-quality-promptfoo.json:7 and verdicts.json:7, where owner_words becomes owner_paraphrase;
  • verdicts.json's Q11 correction, its 1067 reason and its 2541 claim;
  • docs/decisions/2026-10-04-upstream-surface-watch.md:19-20;
  • the consensus record's claim that the rule record quotes the rule verbatim (C:1662), regenerated into definitive-manifest.json and the handbook.
• Re-bound:
  • the consensus record's hashed records, copy-notes.json and integration-resolutions.json's measured_record;
  • the manifest and the handbook through their builders;
  • the handbook receipt, with a regeneration entry and a reason on every previous_outputs entry (P3, receipt.json:1308).
• Left at their own dates: these historical captures keep the hashes of their dates:
  • starred-protected-20261011 and starred-convergence-20261010;
  • the 2026-10-08 g6c-b record;
  • the 2026-10-06 fix-wave receipts;
  • the 2026-10-07 gateway relabel.
• P2: DIGEST_LIKE also refuses SRI and named-algorithm forms (sha512-…) and base64 runs of 20+ characters with a digit and both letter cases. A session name such as native-agent-stack-99 still passes.
• P3:
  • the relay paraphrases are in they/them;
  • an interim owner decision and a relayed owner_decisions entry refuse an owner-words key at runtime;
  • a paraphrase that is only its prefix is refused.
• Tree-wide scanner (tests/test_owner_quote_publication.py): it reads every tracked UTF-8 text file up to 8,000,000 bytes (MAX_BYTES). Two larger published JSON files are skipped; the GPT read at e21a7b6 checked both directly and found no hits.
  • What it flags: the owner, or the user (as these records call the owner), next to a speech word and a quote mark; a verbatim claim; an owner-words key; or a verbatim key in a file of owner decisions.
  • JSON: string values are read one at a time, so their delimiting quote marks do not count as quotation.
  • Ratchet: tests/fixtures/owner-quote-baseline.json holds per-file counts only: 290 hits in 169 files, the burn-down's input. An unlisted file must be clean, and a count changes only together with the baseline.
  • Fail-first (corrected by commit 4): with commit 3's detector, the run at c95453a flagged only the named records it could see. That detector missed the attribution without a speech word at claude-proposals.md:88 and the single-quoted owner words at verdicts.json:1070. Commit 4's detector catches both; its fail-first is below.
• Detector scope: the approved split said owner-only. The detector also reads "the user" as a speaker, because three of the named records call the owner that. An owner-only detector missed them at base.

Commit 4 (e21a7b61): final, from the command center's adjudication at 9371e2a

This is the third round, scope-locked. Findings are cited by path:line only.

• P1: evidence/artifacts/new-wsl-layer-consensus-20261002/claude-proposals.md:88 held an attributed owner sentence with no speech word.

  • It is now a dated they/them paraphrase.
  • The consensus record (consensus.json:30-32) and copy-notes.json re-bind to it.
  • The manifest and the handbook regenerate, and the handbook receipt gains its regeneration entry.

• P2-1: a relay_record case isolates DIGEST_LIKE's base64 branch. It fails at c95453a (unexpectedly None) and passes now.

  • The local review's P2 at c95453a is fixed in the same place: a dated record name was refused as a digest. A hex run now counts as a digest only when it mixes digits and letters a-f, and a hex-prefix case covers that.

• P2-2: the scanner also flags two more forms:

  • an attribution without a speech word (the owner, a colon, then a quote mark);
  • single-quoted owner words, opened by a typographic quote or by a straight single quote at the start of a word.

  Synthetic cases include plain apostrophes, which pass. verdicts.json:1070 no longer keeps single-quoted owner words. The ratchet baseline is regenerated: 477 hits in 242 files.

• Fail-first (this commit's scanner and baseline, in scratch exports):

  • at 9371e2a, claude-proposals.md is unlisted and verdicts.json is above its count;
  • at c95453a, claude-request.md, claude-proposals.md, repository-quality-promptfoo.json and upstream-surface-watch.md are unlisted, and verdicts.json, wave2-records.json and the rule record are above their counts.

• Results: 530 tests OK across the readers of these files; the generators' --check exit 0; validate rc 0.

Commit 5 (fd51f88d): final, from the command center's adjudication at e21a7b6

The command center ruled at 2026-10-15T15:49Z and completed the ruling at 16:20Z, with both families' reads at e21a7b6. Its scope is the 23 files this PR changes. Findings are cited by path:line only.

• The command center's items:
  • evidence/artifacts/new-wsl-layer-consensus-20261002/wave2-records.json:147 and :274, and evidence/artifacts/final-architecture-round2-20261004/verdicts.json:220 and :1160, are dated paraphrases.
  • A possessive attribution class: the owner's or the user's, then a quoted phrase of two or more words. A string literal's closing delimiter and a quoted dated-paraphrase prefix are not quotations.
  • A positive case for a dated relay-record name (assemble_manifest.py:395-397; also the GPT read's follow-up 1).
• A FULL read of the changed files:
  • wave2-records.json:103 held quoted owner wording and is now a dated paraphrase.
  • docs/decisions/2026-10-04-repository-quality-rule.md:7 and docs/decisions/2026-10-04-2604-e2e-fix-wave.md:51 now state the 2026-10-01 rule without quoting it.
  • Every other candidate line is a slot owner, code, the relay contract's own text, or an existing paraphrase.
• The GPT read's P2s and the local review's P2s (tests/test_owner_quote_publication.py):
  • Each attribution is matched within one sentence. Markdown paragraphs, list items and block quotes are read as wrapped prose, and a hit is located at the line where its attribution starts (GPT blocker 2; local review :73-75).
  • A typographic apostrophe inside a word is not a quote mark (GPT blocker 3; local review :32).
  • A single closing quote before the speaker is an attribution (local review :35).
• Re-bound:
  • consensus.json and integration-resolutions.json;
  • the manifest and the handbook, through their builders;
  • the handbook receipt, which gains this regeneration with its previous outputs and reason.
• Ratchet baseline: regenerated at 469 hits in 229 files.
• Fail-first:
  • The seven new detector cases all fail against e21a7b6's detector, and all hold now.
  • This commit's scanner and baseline over the e21a7b6 tree report verdicts.json and wave2-records.json above their counts.
  • The GPT reviewer's probe returns 3 of 3 OK. It failed 3 of 3 at e21a7b6.
• Results: 616 tests OK (3 skipped) across the readers of these files; the three generators' --check exit 0; validate rc 0.

Supersession: the no-owner-quote rule of 2026-10-14 supersedes, for every published file, the round-2 resolution to quote the owner exactly (evidence/artifacts/final-architecture-round2-20261004/integration-resolutions.json:543, :630, :801). #1159 applied it to the generated files; this PR applies it to the relay contract and its decision records.

No history rewrite: earlier commits keep the old text, and this diff necessarily shows the removed lines.

Commit 6 (9bcb8133): blocker, from the GPT fix3r read at fd51f88

The GPT fix3r read at fd51f88 (2026-10-15, CHANGES_REQUESTED, one P1; the command center released this commit at 2026-10-15T20:45Z). Findings are cited by locator only.

• evidence/artifacts/new-wsl-layer-consensus-20261002/wave2-records.json:274 (layers.custody.ruling.changes["12"]) kept earlier wording under a paraphrase label. It now states the owner's 2026-09-26 decision in other words, dated and attributed. Against the field at the PR's base, the attributed clause shares no run of two or more non-numeric words. The line count is unchanged.
• Re-bound: consensus.json (4) and integration-resolutions.json (1); the manifest and the handbook through their builders; the handbook receipt gains this regeneration with its previous outputs and reason; manifests/evidence.json.
• Fail-first: tests/test_new_wsl_handbook.py OwnerQuotationTests gains a check of the custody change's public text. It failed at fd51f88 (1 of 5) and holds now.
• Results: 741 tests OK (3 skipped) across the readers of these files; the three generators' --check exit 0; validate rc 0; tests.test_pre_push_gate OK on the committed tree; the reviewer's fix3r detector probe returns 7 of 7 on fd51f88 and on this head.

Commit 7 (e0ff2abb): the adjudicated fold at 9bcb813

The command center's adjudication of #1161 at 9bcb813 (2026-10-15T21:59Z) followed its delta read (PASS) and the local Codex
re-review after the bounded rounds (three P2s, no P0/P1). The GPT fix4 delta read of fd51f88..9bcb813 is an ACK. Findings are
cited by locator only.

• Fixed (claims false at the head): the detector's docstring (tests/test_owner_quote_publication.py:1, :7-8) now states its
  three bounds, each naming its Follow-up below. manifests/evidence.json is rewritten by scripts/evidence_manifest.py --write
  (:67), which restores 13 rows' \u escapes that the previous round's registration had written as raw UTF-8.
• Results: 228 tests OK across the detector's and the handbook's readers; validate rc 0; tests.test_pre_push_gate 29 OK on the committed tree.

Commit 8 (768d9cc0): the adjudicated fold at e0ff2ab

The local Codex re-review at e0ff2ab (two P2s, no P0/P1) and the command center's adjudication of it (2026-10-15, addendum 01 to
its adjudication at 9bcb813). The GPT fix5 delta read of 9bcb813..e0ff2ab (coordination/cc-reads-20261005/fix5-pr1161/GPT-VERDICT-e0ff2abb3.md)
is an ACK. Findings are cited by locator only.

• Fixed: commit 3 (9371e2a) had attached the relay-contract supersession reason to the historical ROW70 entry of the handbook
  receipt (previous_outputs[7]) as well as to its own entry ([20]). [7] is back to its bytes at the base, and the 19 entries
  the base holds are equal to it. test_each_supersession_reason_binds_only_the_entry_it_supersedes failed at e0ff2ab (entries 7
  and 20 shared a reason) and holds now.
• Results: 229 tests OK across the detector's and the handbook's readers; validate rc 0; tests.test_pre_push_gate 29 OK on the committed tree.

Commit 9 (1fb477b5): the P1 fold at 768d9cc

The command center's ruling of 2026-10-16: the co-op's GPT count read at 768d9cc found four real instances of the shape
that escaped the block-quote introduction check (tests/test_owner_quote_publication.py:66), a P1; the other shape (:61)
has none on real content and stays a Follow-up. The positions came from the GPT locator read (coordination/cc-reads-20261005/loc1-pr1161/).
Findings are cited by locator only.

• Two records carried directive wording under an owner introduction (docs/decisions/2026-09-25-top-rule-sota-sources.md:22,
  docs/decisions/2026-10-05-official-upstream-never-rebuild.md:17): each is now a dated paraphrase in other words.
• Two named a role or a lane as an owner (docs/decisions/2026-10-03-currency-wave-w1.md:227, a retained repair report, and
  evidence/artifacts/delegated-decisions-20260928/coordination.md:13): each now names it as a lane, with the change dated.
• Bindings: the starred-convergence captures record these files' sha256 at their own dates and stay unchanged;
  evidence/artifacts/engine-currency-20261013/context.json binds its file at a commit.
• Detector: a block quote is attributed by the paragraph or list item above it, read whole, when that names the owner and
  ends with a colon, with or without a speech word, and blank quote lines are skipped. This also closes C3. Four synthetic cases
  failed at 768d9cc and hold now.
• Baseline: one burn-down file goes from 2 to 3 (a block quote in a file already listed), and the detector's own fixtures
  from 30 to 35.
• Results: the detector's module OK; the binding readers (tests.test_ecosystem_manifest, tests.test_starred_convergence)
  124 OK; validate rc 0; tests.test_pre_push_gate 29 OK on the committed tree.

Commit 10 (d4e3b6ac): the fold at 1fb477b, the ratchet on the merge with main

This answers GPT's fix7 micro and the count2 re-check at 1fb477b. Both give the same P2 blocker
(coordination/cc-reads-20261005/fix7-pr1161/ and count2-pr1161/, cited by name and locator only): the ratchet failed on
the merged tree (tests/test_owner_quote_publication.py:217 at 1fb477b; a file that main added since, returns.json:902).
The command center ruled ONE fold at 2026-10-16T05:13Z.

• Rebase: once, onto main 30542e705. The range-diff shows nine = pairs, and git merge-tree rc 0. Commits 1 to 9
  are now 4cd31e8f, bbeb43a7, 81afff68, aa6e33a5, 7dde9061, ec766080, e556062c, ba0a7af8 and fd112c41.
  The headings above keep their earlier names. The earlier head is kept as a local backup ref.
• Measured baseline: measured_counts() and write_baseline() rebuild tests/fixtures/owner-quote-baseline.json from
  scan() of the tree that lands, with no hand-set number
  (python3 tests/test_owner_quote_publication.py --write-baseline).
  • On the rebased tree it measures 230 files and 478 hits.
  • The one change is the file main added, at 3.
• Merge test: test_the_ratchet_holds_on_the_merge_with_main runs the ratchet on git merge-tree --write-tree of the
  default branch and HEAD, extracted outside the checkout. It is skipped only where no default branch is fetched. With the
  earlier baseline it fails on this merge: the added file, 3 hits, unlisted.
• Results:
  • the detector module, 5 tests OK;
  • the reviewer probe, 7/7 (counts only);
  • tests.test_new_wsl_handbook, 94 OK;
  • test_new_wsl_definitive_defaults, test_validate, test_evidence_manifest and test_registry_rows together,
    248 OK;
  • validate rc 0, and build_new_wsl_handbook --check rc 0;
  • tests.test_pre_push_gate 29 OK on the committed tree.

SOTA sources

• The repository's own generators and contract, at Paraphrase owner quotations in the new-WSL layer-consensus record and regenerate the manifest and handbook #1159's head: assemble_manifest.py (OWNER_AUTHORITY_FIELDS, check_owner_batch), render_tables.py and scripts/build_new_wsl_handbook.py. This is a publication correction with no upstream mechanism.
• The rule: the 2026-10-14 no-owner-quote rule. The command center's rulings are of 2026-10-15T12:19Z and 12:38Z.

Evidence-class table

Local commands run

$ python3 -m unittest tests.test_new_wsl_handbook.OwnerQuotationTests tests.test_new_wsl_definitive_defaults.Manifest   # #1159 head a3d53be4 with these tests
FAIL: ...carry_no_owner_quotation (the three verbatim fields); FAIL x3: ...relaying_decision_records_carry_the_paraphrase_not_the_words; ERROR: ...refuses_an_owner_batch_outside_amendment_4
$ systemd-run --user --scope -p CPUWeight=idle --quiet -- python3 -m unittest tests.test_architecture_pin_source tests.test_harbor_currency_contracts tests.test_new_wsl_client_config tests.test_new_wsl_definitive_defaults tests.test_new_wsl_handbook tests.test_new_wsl_profile tests.test_starred_convergence tests.test_ecosystem_manifest
Ran 615 tests; OK (skipped=3)
$ python3 scripts/validate.py
rc 0; status passed
$ systemd-run --user --scope -p CPUWeight=idle --quiet -- python3 -m unittest tests.test_architecture_pin_source tests.test_codex_lane tests.test_harbor_currency_contracts tests.test_new_wsl_handbook tests.test_new_wsl_gateway_composition tests.test_research_entry tests.test_owner_quote_publication tests.test_new_wsl_definitive_defaults tests.test_new_wsl_client_config tests.test_new_wsl_profile   # fd51f88d
Ran 616 tests; OK (skipped=3)
$ python3 scripts/build_new_wsl_handbook.py --check; python3 .../assemble_manifest.py --check; python3 .../render_tables.py --check docs/decisions/2026-10-01-new-wsl-definitive-defaults.md
rc 0; rc 0; rc 0

Follow-ups

• From GPT's fix7 micro and count2 at 1fb477b, by locator. These are scanner-class: the locators were relayed by
  script, and their intent is not restated here. Lines are as at 1fb477b; the scanner file's lines move at d4e3b6ac
  (:62 to :66, :139 to :143, :171 to :175).
  • fix7 P3s:
    • tests/test_owner_quote_publication.py:62 with this body's :233;
    • tests/test_owner_quote_publication.py:139 with this body's :184;
    • docs/harness-defaults.md:123, which is already in the burn-down front's follow-up 2 (live locators into
      line-moving files).
  • count2 P2, rated as a follow-up: tests/test_new_wsl_handbook.py:96 with tests/test_owner_quote_publication.py:171.
    Both pass on the merged tree (the handbook module, 94 OK). Its interaction with WSL install: add no-start mode and host port allocation #1180's generator change (still open)
    is checked when WSL install: add no-start mode and host port allocation #1180 lands.
  • count2 P3s:
    • docs/harness-defaults.md:123 with tests/test_new_wsl_handbook.py:1397;
    • tests/test_owner_quote_publication.py:139 with this body's :239.
• First burn-down PR, at the front of the queue (the command center's ruling of 2026-10-15T16:20Z): these are the GPT read's items at e21a7b6 that fall outside the 23 files:
  • its P1 locators: evidence/artifacts/new-wsl-local-models-20261002/PREREGISTRATION-local-models.md:134-136, evidence/artifacts/prompt-audit-20260927/x9-round3/prereg-r3.draft.json:7 and prereg-r3.json:7. They are baselined at one hit each. They are preregistrations, so the burn-down states how a paraphrase keeps their hash bindings;
  • its follow-up 2, live locator references into line-moving files (for example docs/harness-defaults.md:123), and follow-up 3, the scanner's size bound;
  • the earlier queue: returns.json:5904, credential-inventory.json:213 and openhands-resolver-isolation.md:919.
  • From the command center's MICRO read at fd51f88:
    • P2: evidence/artifacts/new-wsl-layer-consensus-20261002/wave2-records.json:274: fixed in 9bcb813 (the GPT fix3r read made it its P1).
    • P3: the detector boundary at tests/test_owner_quote_publication.py:79 (below).
  • From the command center's adjudication at 9bcb813 (2026-10-15T21:59Z):
    • C1, the restructure: several attributions in one line or paragraph count once in the ratchet (rule 4: a recurring class is
      restructured, not given another pattern round);
    • C3: a block-quote introduction wrapped across lines: closed in 1fb477b; a CommonMark parser stays a PROPOSAL for
      introductions further from the quote;
    • the C2 bound: files over 8 MB are not scanned (the P2 was rejected: no failure on real content);
    • wave2-records.json:274 narrows the 2026-09-26 decision to paid search; set it against
      evidence/artifacts/final-architecture-round2-20261004/verdicts.json:1173 and :1186,
      catalogs/sota-convergence/manifest-20260929.json:4720 and evidence/artifacts/landscape-sweep-20260929/returns.json:17191
      and :17329.
  • From the local Codex re-review at e0ff2ab, P2-A (the command center's adjudication, addendum 01): the detector's JSON path
    keys on owner_decision tokens, so an attribution object without one is not counted. It is the detector-gap class sent to the
    restructure (rule 4); a scan of the tracked JSON finds no such object on real content (rule 5).
  • From the local Codex re-review at 768d9cc (pr1161-768d9cc0/review_output.json, cited by name): two more instances of the
    same detector-gap class, at tests/test_owner_quote_publication.py:61 and :66. They go to the detector's restructure with
    C1, C3 and P2-A (rule 4); no commit. The count-only check of real content for these two instances is run by the GPT family,
    which reads this class's review files.
  • From the GPT fix3r read at fd51f88 (coordination/cc-reads-20261005/fix3r-pr1161/GPT-VERDICT-fd51f88d2.md), three P3s:
    • the detector's disclosed sentence-boundary limitation;
    • an existing live locator, and the scanner's bounded coverage;
    • the pinned primary reference for the scanner corrections.
  • The final commit's wider detector reports 50 structural locators in 36 files outside the 23 that e21a7b6's detector did not. They are counted in the baseline and are not confirmed quotations; the classification pass decides each. The blueprints/us-equities/ one goes to the trading lane:
    • adoption/new-wsl/client-config-map.json:298,345,488
    • blueprints/convergence-practice/local-inference-latest-20260926/results/results.md:188
    • blueprints/us-equities/sim-paper-compare/README.md:522
    • catalogs/sota-convergence/manifest-20260926.json:16879,19225
    • docs/decisions/2026-09-22-github-automation-closure.md:385
    • docs/decisions/2026-09-23-max-effort-default.md:77
    • docs/decisions/2026-09-25-skills-trial-and-usage.md:3
    • docs/decisions/2026-09-25-workstation-sota-refresh.md:336
    • docs/decisions/2026-09-26-codex-worker-lane.md:89
    • docs/decisions/2026-09-28-community-sweep.md:219
    • docs/decisions/2026-09-28-delegated-decisions.md:107
    • docs/decisions/2026-09-28-openhands-resolver-isolation.md:157
    • docs/decisions/2026-10-01-definitive-sota-wsl-program.md:346
    • docs/decisions/2026-10-01-new-wsl-architecture-edition.md:175,182
    • docs/decisions/2026-10-01-new-wsl-definitive-defaults.md:344
    • docs/decisions/2026-10-03-compaction-window-ab-retirement.md:426
    • docs/decisions/2026-10-04-new-wsl-changelog-parity.md:44,57
    • docs/decisions/2026-10-05-macos-ci-advisory.md:13
    • docs/decisions/2026-10-10-landscape-scratch-ownership.md:115
    • evidence/artifacts/currency-wave-w1-20261003/judgments/move-hold-judges.json:33
    • evidence/artifacts/delegated-decisions-20260928/m3/packet.redacted.md:3
    • evidence/artifacts/final-architecture-round2-20261004/landscape-verdicts.json:24
    • evidence/artifacts/landscape-sweep-20260926/lanes.json:5774,6352,7129,10786,12120
    • evidence/artifacts/landscape-sweep-20260926/returns.json:3299,13223,16965,29959,30787
    • evidence/artifacts/mac-stage1-client-layer-20260927/README.md:254,557
    • evidence/artifacts/new-wsl-definitive-defaults-20261001/foundation-definitive.compact.json:387
    • evidence/artifacts/new-wsl-definitive-defaults-20261001/trading/trading-ownership.json:9
    • evidence/artifacts/new-wsl-install-plan-20261002/SOURCES.md:345
    • evidence/artifacts/omniroute-rebuild-20260930/PLAN.md:68
    • evidence/artifacts/omniroute-rebuild-20260930/receipt.json:112
    • evidence/artifacts/sdk-runtime-coverage-20260922/dispatch-frameworks.json:1678
    • evidence/artifacts/skills-trials-20261009/corpus/C10/case-optional-fields/SKILL.md:3
    • evidence/artifacts/sota-finalization-20260930/plan.json:57
    • manifests/stack.json:1940
    • tests/test_native_practice_retired_routes.py:389
    • tools/landing/README.md:124
• A known detector boundary (tests/test_owner_quote_publication.py:79, burn-down FRONT): a sentence ends only before a capital letter, so a paraphrase followed by a sentence that opens with a quote mark is read as one sentence. It errs toward a hit, never a miss.
• The baseline counts structural hits, not confirmed quotations. The burn-down starts with a classification pass, and trading files go to the trading lane.
• The starred-convergence captures (catalogs/convergence-practice/sources/starred-protected-20261011.json, starred-convergence-20261010.json) and docs/decisions/2026-10-08-source-based-records-g6c-b.md record the three decision records' sha256 at their own dates. They are dated historical captures that their tests do not recompute, so they are left unchanged.

Checklist

• No GitHub Actions changed.
• No workflows changed.
• No secrets are printed, logged or committed.
• No new paid hosting, subscription or billing surface.
• Peer-owned untracked files and worktrees were preserved.

🤖 Generated with Claude Code