Consolidation run — 2026-07-26

Third consolidation run (OPS-015; workflow v0d.24/25). Run branch: claude/consolidate-2026-07-26, off master (113c83c). Worksheets: consolidation-pregate-2026-07-26 · claim-retirement-queue-2026-07-26.

Run scope (fixed at C0)

Register at C0: 303 entries — 23 supported, 144 live, 130 candidate, 1 contested, 5 retired (unchanged through runs 1–2, which promoted nothing). normalize_claims_table.py --check: in sync at C0. First run on the user-decided run-3+ seat config: all subagent seats Opus 5 — adjudicators (claim-promotion-reviewer, claim-refuter) at high effort, citation-verifier at medium, Fable main-thread only (environment register, 2026-07-26).

  • C1 (subtraction): nothing to apply. Retirement queue at the continuity window (--staleness-days 60): 125 rows, all keep/staleness, 0 concrete-defect rows (downgrade/retire/merge) — third consecutive clean queue at register 303. keep/review rows route into batch selection (they are the same oldest-first population the pregate enumerates).
  • C2 (candidate triage): 30 oldest candidates (batch cursor = Updated:, per the run-1 rule). Following the run-1/run-2 discharge pattern: the 13 multi-source candidates get the two-agent gate (3-test claim-promotion-reviewer + rival-reading claim-refuter, peer-isolated); the 17 single-source candidates (incl. the two remaining Simmel entries, four Lawlor, three Nancy, two Gadamer, two Mounier-cluster) discharge mechanically as KEEP-with-condition — promotion is impossible under the single-source rule and the concrete-defect scan returned 0 rows. One gated candidate, post-husserlian-horizont-triangle, is additionally capped at candidate by construction (weave Pass 3 Latent-Adjacent, v0d.9 carve-out) — the cap binds regardless of gate verdicts.
  • C3 (supported jury): first 5 of 90 eligible, oldest-first — deformation-coherente-mp-coinage · language-as-missing-case-of-lateral-universal · ontologie-de-la-chair-misnomer · sedimentation-pluralizes-by-practice-type · phenomenal-invariants-replace-resemblance-and-denotation. Full jury per claim: citation-verifier on every Evidence anchor, independent 5-test review, unanimity default-deny refuter panel (evidence-chain / rival-reading / scope). Cap ≤5 promotions; no same-run substitution for blocks. (Run 2's five juried claims are correctly excluded by the pregate's 14-day settle window — their revision sets from the run-2 digest remain open.)
  • C4: skipped (C2+C3 consume the run's edit budget, run-2 precedent). Speculate backlog (~48 remaining survivors across ledger batches 1–2) and the ~82-row harvest reservoir stay queued.
  • Parked ledger: read at C0 — empty; nothing to unpark.
  • Adjudication workflow: wf_2de1ae1f-59d (consolidate-c2-c3-adjudication, 51 agents: 13×2 C2 + 5×5 C3), peer-isolated, launched before any register write. Run-3 scaffolding carry-forwards applied first: verdict-conditional refuter schema (REFUTED⇒objection, STANDS⇒tried, enforced at the StructuredOutput layer (erratum 2026-07-27: overstated — the conditional requirement lives in field descriptions only; neither field is in the schema's required list, so the layer cannot reject a bare verdict. See Post-run errata #6)) and C3 apply-block emission + clean-run provenance in gen_adjudications.py. The merge_rerun.py slot-enumerated coverage carry-forward stays queued (contingency tooling; no rerun needed this run).

Run mechanics (recorded deviations)

  1. Refuter schema false start. The verdict-conditional REFUTE schema (PR #112 round-3 carry-forward) was first implemented as a top-level allOf/if/then — the API rejects oneOf/allOf/anyOf at the top level of a tool input_schema, so all 28 refuters of the first launch errored at spawn while the 23 reviewer/verifier seats completed. Reverted to the run-2-proven shape (conditional requirement in the field descriptions; run 2 went 46/46 on it) and resumed wf_2de1ae1f-59d — completed verdicts replayed from cache, refuters ran live. Durable lesson recorded in the workflow script; a schema-layer redesign needs a nested discriminator (queued in NEXT.md).
  2. Session restart + checkpoint fork. The first resume was interrupted by a Claude Code process exit at 40/51 journaled verdicts; the interposed checkpoint fork re-ran 8 C2 reviewer seats under fresh job keys. 6/8 returned the same summary; 2/8 diverged — aufheben-… [PROMOTE, srcs=4] vs [REQUEST-REVISION, srcs=3] and touching-the-untouchable-… [REQUEST-REVISION] vs [HOLD]. Outcome-invariant: aufheben's rival-reading refuter REFUTED (gate blocks regardless of reviewer leg), and both touching verdicts are non-PROMOTE. Per default-deny, the split aufheben panel is itself recorded as unresolved doubt; the final relaunch completed the remaining panel from the durable journal + cache.
  3. Manual kill + fresh-session completion. The final relaunch of deviation 2 was in fact killed manually mid-run (2026-07-26 12:27, the session-backgrounding accident) after journaling 5 refuter verdicts — all duplicates of seats already journaled by the interrupted resume, so the kill cost no unique verdicts. Coverage at pause: 32/51 seats (23 reviewer/verifier + 9 refuters, all REFUTED; the 5 twice-run refuter seats concur REFUTED/REFUTED in every case, and the 6 same-summary fork reviewer re-runs likewise concur — divergence stayed confined to the two deviation-2 seats). The 19 never-run refuter seats (4 C2 rival-reading + 5 C3 × 3 lenses) were completed 2026-07-27 in a fresh session as wf_f4d3dab1-358 (.claude/workflows/rerun-refuters-missing-2026-07-26.js, prompts/schema/agent-types/seat-config verbatim; 19/19, 0 dropped), bringing the jury to 51/51. Assembly: reviewer/verifier verdicts from the completed instance's returned result; early refuter verdicts from the durable journals; alternates preserved in the verdicts JSON.

Adjudication outcomes (C5, applied 2026-07-27)

Jury: 51/51, decisive, 0 promotions. Verdicts: consolidation-2026-07-26-verdicts.json; per-claim apply-doc: consolidation-2026-07-26-adjudications.md (regenerated at 51/51; gate outcomes mechanically derived). All 28 refuters returned REFUTED (13 C2 rival-reading + 15 C3 across the three lenses) — the third consecutive all-REFUTED panel (74/74 across runs 1–3). C2 reviewers: 1 PROMOTE (chouraqui-two-2025-papers…, blocked by its refuter), 2 HOLD, 10 REQUEST-REVISION, plus the deviation-2 aufheben split (fork REQUEST-REVISION copy governs under default-deny). C3: no claim had a clean verifier sheet (1/2/0, 2/1/1, 3/1/0, 0/2/2, 4/1/0 PASS/UNCERTAIN/FAIL) and no 5-test reviewer said PROMOTE (3 REQUEST-REVISION, 2 HOLD) — every C3 claim was blocked on at least two independent legs before the refuter panel voted.

Absence-claim re-verification (main-thread, before apply). 242 absence assertions collected across the 51 seats; tiered re-verification (54 mechanical source-existence/quote-level checks re-run by grep + targeted raw reads on every objection-load-bearing item). Two catches, both worth the discipline: (a) the rival-reading refuter's side-assertion that Voix raw 4660 is not the conjoined locution is a false absence — the main-thread read confirms the definitional "la déformation cohérente est un moyen de création" verbatim at 4660, confirming the 5-test reviewer against its own panel peer; (b) the maintainer's own first-pass grep missed the same line (encoding artifact) and was corrected by reading the line. One expected self-hit: the "les-voix-du-silence referenced nowhere under wiki/" claim became false the moment this run's own artifacts referenced it.

Status changes

One status change (every other entry stays at its pre-run status, with objections and repairs applied). Standing veto covers every write below; each is revertible per entry within the single run commit.

Slug Change Rationale Veto
claims#deformation-coherente-mp-coinage live → contested The C3 reviewer found — and the main thread independently re-verified in raw — that Malraux's French Voix du silence uses déformation cohérente twice (raw 3762, Pléiade block {354}; definitionally at raw 4660), refuting the Johnson coinage attribution at the philological level (the 1951 volume precedes MP's 1952 Signs essay); the English 0-hit check that had sustained the entry is a Gilbert-translation artifact. Objections + finding in Counterpressure; corrective restatement filed to the Harvest reservoir. veto: say "veto deformation-coherente-mp-coinage" — restores live (pre-run status)
  • C2 gated (13): all stay candidate — every rival-reading refuter REFUTED under default-deny (the lone reviewer-PROMOTE was refuter-blocked). Each objection appended verbatim to the entry's Counterpressure; prose Status History line with the failed legs + an explicit would-promote condition; Updated: 2026-07-26 (batch-cursor rule). Reviewer-named mechanical repairs applied in the same edit (see Rule 16 / citation repairs); interpretive restatements queued, not applied (death-of-god headline; bergson problem-space reformulation; onto-theo exclusive-to-priority tightening; ontologie-de-la-chair philological-only narrowing; inkpin Payoff carry-in).
  • C2 single-source (17: two Simmel, four Lawlor, three Nancy, two Gadamer, two Mounier-cluster, militante, bete, no-common-world, klossowski): mechanical KEEP-with-condition discharge — promotion impossible under the single-source rule; each entry's Status History now names its specific second-source promotion path; Updated: bumped.
  • C3 blocked (4 further claims): language-as-missing-case…, ontologie-de-la-chair-misnomer, sedimentation-pluralizes…, phenomenal-invariants… stay live — verifier UNCERTAIN/FAIL + 5-test non-PROMOTE + 3× REFUTED each; 12 objections appended; verifier-named repairs applied.
  • Parked ledger: nothing parked this run; ledger remains empty.

Rule 16 findings (prominent — citation-fabrication-pattern class)

Five findings, all repaired in-run (entry-side and, where propagated, page-side); mixed provenance (compression artifacts + stale locators + one false nested attribution), same repair standard as run 2:

  1. nancy-haptico-transcendental-reduction — the "Corpus p. 40" shell. The Nancean "Sense IS touching" formula carried a nested attribution to Corpus p. 40 — a plausible book+page for a text not in the corpus (the named failure mode). Derrida's own nn. 16/24–25 attribute it to Nancy, "Elliptical Sense," A Finite Thinking pp. 109–110; Morin's extraction independently corroborates (FT 110 / OT 275). Corrected in the entry and on 4 propagated pages (erratum 2026-07-27: 3 pages — concept page, source page, jean-luc-nancy; the index.md Summary cell was a 4th, uncounted and uncorrected site, repaired 2026-07-27 — see Post-run errata below); the 5th location (the Derrida extraction note's hedged line) is queued — extraction notes sit outside this run's write scope. All four Derrida page locators in the entry were also +10 (raw-marker vs printed folio); corrected, matching Morin's independent cites.
  2. language-as-missing-case-of-lateral-universal — two pseudo-quotes. Evidence bullet 1's headline Kee "quotation" and its quoted Lau sentence occur nowhere in the raw article (repo-wide grep: claims.md only) — same class as the two 2026-07-21 sibling repairs. Replaced with Kee's actual sentences (raw 84/86/87); raw-line locators replace the unverifiable page cites (the Kee raw is unpaginated). The entry's Counterpressure hedge ("page correspondence cannot be verified") was itself a false absence — discharged via the SW&WE folio-to-page concordance (constant −54 offset, index-confirmed).
  3. sedimentation-pluralizes-by-practice-type — untraceable French quotation + garbled IP quote + fn mis-attribution + wrong typology memberships. The PdM-142 French sentence exists nowhere in the repository outside the entry; replaced with the raw-verified English anchor (PoW pp. 139–140). The "IP 117" quotation was garbled and French-paginated; corrected to English p. 85, raw 1208, verbatim. fn 14 was credited with Origin-of-Geometry content it does not contain (its actual content: the painting-thematization concession). The typology memberships ("mathematics… literature, music… history, philosophy") were not Inkpin's — §3 names tools and language as the intermediate cases.
  4. circulus-vitiosus-two-receptions — misattributed quotation. "Being that represents itself" was presented as MP's formulation; it is Klossowski's (quoting a Nietzsche note, fn. 103, raw 1112) with no MP-side attestation anywhere in the corpus. Corrected; the bullet now does Klossowski-side work only.
  5. deformation-coherente-mp-coinage — the decisive one. Johnson's fn-29 locator was mis-chaptered (Ch 6, not Ch 3; Ch 3's own fn 29 is a Lyotard citation), Johnson's Pléiade p. 322 page-cite does not contain the phrase — and the phrase itself IS in the volume, twice (see Status changes). The uningested French raw file had never been referenced by any wiki artifact, which is how the English-side 0-hit corroboration survived four review passes.

Also repaired, non-fabrication class: stale motifs.md line-number locators in three entries (aufheben 327→§heading at 351/355; death-of-god 698→821–826 with the BRIDGE-not-HUB weight correction; onto-theo 441→534; ontological- rehab 778→1086–1088 with the THEME-not-HUB correction); the nonexistent GA 11 "Randbemerkung (89)" locator (the phrase is main-body text at raw 1322, fn (110)); marker-offset corrections in touching… (p. 58→47, p. 204→194); Kee-slogan re-attribution in post-husserlian-horizont-triangle (the "horizons, not the limits" formula is Kee p. 80's, not a PbP quotation); chouraqui-two-2025… quote-hygiene (two wiki compressions de-quoted; two Counterpressure statements corrected against Healing fn 16/18); unattested "pragmatist" descriptor removed from coherent-deformation-inkpin…; bergson-fourth-corner… mis-addressed anchor re-pointed; evidence-anchor additions where reviewers supplied verified loci (GM III.27/II.20 primary anchor; Morin 2022 second source; V&I + Phaedo legs for touching…; HT extraction anchors; WdL-1813 loci; Healing fn-16 Huizinga bridge).

Caps consumed / backlog remaining

C1 0/15 (0 concrete-defect rows, third consecutive clean queue) · C2 30/30 triaged (13 gated + 17 single-source) · C3 5/5 juried, 0/5 promotions (cap unconsumed; 1 downgrade live→contested) · C4 skipped. Register after run: 303 entries — 23 supported, 143 live, 130 candidate, 2 contested, 5 retired. Remaining backlog: 40 candidates untriaged by C2 batches (130 − 30×3), 85 supported-eligible live claims (90 at this run's pregate − 5 juried), ~54 speculate survivors, the ~82-row harvest reservoir + 1 new reservoir item (the Malraux-coinage corrective restatement). The register is not yet dry: next trigger is another /consolidate run (run 4).

Files touched

wiki/claims.md (35 entries: 13 gated + 17 single-source + 5 C3; 28 verbatim refuter objections into Counterpressure; ~40 reviewer/verifier-named citation repairs; 1 restatus) · label/citation sync on 7 pages (wiki/concepts/coherent-deformation.md, wiki/concepts/haptico-transcendental-reduction.md, wiki/concepts/nothingness-as-pseudo-idea.md, wiki/entities/andre-malraux.md, wiki/entities/jean-luc-nancy.md, wiki/sources/derrida-2000-on-touching-nancy.md, wiki/questions/taddio-2025-on-science-secrete-and-gestalt.md) · derived surfaces (claims.md Summary table; index.md Claims section + counts + page tables) · baselines (Paper A manifest claim-status fixture 144 live/1 contested → 143/2; wiki/overview.md stats; graph SHA fixture; the Paper-A dashboard_contested_claim_slugs pin in paper_a_provenance_expected.json

  • test_paper_a_a2.py, hand-patched — outside the refresh_baselines patcher list, carry-forward filed in NEXT.md) · run artifacts (consolidation-2026-07-26-verdicts.json, consolidation-2026-07-26-adjudications.md, this digest, pregate + queue worksheets) · wiki/log.md · NEXT.md (OPS-015 progress) · run tooling under .claude/workflows/ (consolidate-c2-c3-adjudication.js refuter- schema fix + parameterization; gen_adjudications.py C3 apply-block emission
  • clean-run provenance; rerun-refuters-missing-2026-07-26.js new).

Post-run errata & repairs (2026-07-27, PR #113 review follow-up)

Surfaced by the Opus 5 review of PR #113; applied on the run branch before merge. Each carries the standard post-hoc veto path.

  1. index.md Summary cell — the uncounted 4th propagation site. The haptico-transcendental-reduction row's editorial Summary cell still carried both repaired defects ("§13" and "Corpus p. 40") — Summary cells are editorial, so normalize_index_tables.py correctly preserves them and no script would ever fix it. Repaired by hand; the Rule 16 finding #1 text above overcounted "4 propagated pages" (3 were corrected; inline erratum added). Veto: revert the cell edit.
  2. Concept-page locator harmonization. The run's §12/+10 correction on haptico-transcendental-reduction.md landed only in the Sources line; the body kept raw-marker locators (pp. 279/285/303, "§13" at the haptocentrism Stakes bullet) beside the corrected printed folios with nothing distinguishing the conventions. Harmonized to printed folios throughout (verified against the raw TOC: §12 = printed 265–280, markers {275}–{290}; §13 starts printed 281; the quasi-hyper- variant at printed 292–293 IS in §13, so "§§12-13" stands). updated: bumped. Veto: revert the page edit.
  3. deformation-coherente-mp-coinage confidence medium → low. The C3 5-test reviewer explicitly recommended the confidence downgrade with the restatus; the run applied live -> contested but neither applied nor recorded declining the confidence change. Applied 2026-07-27 with a Status History prose line; Counterpressure bullet 4's leg-(a) promotion condition (proved unsatisfiable by the French attestation) annotated as discharged, matching bullet 1. Veto: restore Confidence: medium, strike the 2026-07-27 Status History line, and drop the bullet-4 annotation — the contested status is untouched either way.
  4. Survivor-count discrepancy. This digest says both "~48 remaining survivors across ledger batches 1–2" (C4 note) and "~54 speculate survivors" (Close). The run dirs' SPECULATIONS files contain rival-pair duplicate headings, so neither figure is verifiably derived; recount queued to the next ops-ledger pass (NEXT.md OPS-015).
  5. Generator divergences queued, not patched. The gen_adjudications.py findings (C3 blocked-apply hard-codes "status stays live"; clean-run prov_override leaves coverage_line/"salvage" strings hard-coded; the embedded-newline coverage sentence; stale evidence_bullet_counts docstring; the uncommitted workflow-meta JSON behind "regenerable at any time") are queued in NEXT.md OPS-015 rather than patched here — the committed run-3 artifacts were hand-corrected at apply time and re-generating them now would churn bytes without changing any applied outcome.
  6. "Enforced at the StructuredOutput layer" was overstated (Codex PR #113 comment on consolidate-c2-c3-adjudication.js:73). The REFUTED⇒objection / STANDS⇒tried requirement is expressed in field descriptions only — the API rejects top-level allOf, and neither field is in required, so the schema accepts a bare verdict. Exposure this run: zero (28/28 REFUTED, every one carrying its objection). Residual risk is asymmetric: a bare REFUTED still blocks under default-deny (objection text is lost, not the block); a bare STANDS would count toward consent unrecorded. Downstream gate validation (reject a STANDS without tried / REFUTED without objection at gen_adjudications.py) queued in NEXT.md OPS-015 alongside the nested-discriminator redesign. Inline erratum added at the scaffolding bullet.
  7. Cite-back sync completed beyond Wiki homes (Codex comment on claims.md:897). The run's label sync covered the two Wiki homes but missed non-home cite-backs: galen-johnson.md ("live, 2026-04-29" ×2 — also a stale "Ch. 3"), the Taddio question page ("live" ×2, one calling the phrase an MP "coinage" flatly), h-synth-reaudit ("live claim" ×1), plus the obsolete promotion guidance on both home pages ("would require … verified absence" — the condition run 3 discharged adversely). All synced 2026-07-27, label/guidance-only (no updated: bumps, OPS-013 precedent). Codex's further "rollup counts" leg was a false positive — the other live labels on the home pages reference different, genuinely live claims. Veto: revert the label edits.