Citation Integrity Spot-Check — 2026-05-09 (Phase 6)

Scope. Audit-produced and freshly-ingested content since the 2026-05-07 tenth-run close: today's Heinbokel ingest (5 new pages + 11 updates), today's deferred-queue discharge (36 pages including Saint Aubert Priority A integration + ultrareview discharge), and the 2026-05-08 weave outputs + 2026-05-07 tenth-run stubs. Three parallel citation-verifier subagents executed against ~37 sampled citations across 19 distinct pages, biased toward audit-produced content per S2.

Headline. No fabrication patterns detected. The audit-produced content meets the citation-traceability standard. Five corrective items flagged for follow-up (mechanical, low-stakes). One critical schema-level finding: wiki/claims.md was duplicated in its entirety (introduced by today's Heinbokel ingest commit d87c8f7); diagnosed and surgically repaired during this audit phase. Reported separately under §"Critical Finding" below.

Summary by batch

Batch Pages audited Citations sampled PASS UNCERTAIN FAIL
6A — Heinbokel ingest 6 (1 source + 2 new concepts + 2 updated concepts + 1 claims.md) 31 31 0 0
6B — Deferred-queue discharge 9 concept pages 17 16 1 0
6C — Tenth-run + weave 5 concept pages 37 + slug refs 28 7 2 (FAIL-adjacent)
Total 19 ~85 ~75 ~8 0 fabrications

Per-batch findings

Batch 6A — Heinbokel ingest content (clean)

  • All 31 citations on the Heinbokel-related artifacts trace to either (a) the extraction-note Pass 2c Evidence section, (b) verified raw-line anchors, or (c) Heinbokel's own quoted material from non-raw/ sources (Heelan, Ricœur, Goldenberg).
  • The Heelan-mediation discipline is consistently maintained: Heelan 2001 is not in raw/; the wiki nowhere makes a primary claim on Heelan and consistently routes Heelan citations through Heinbokel's quotation. Verified at science-as-coherent-deformation, philosophical-praxis-of-medicine, coherent-deformation.
  • The phenomenology-of-medicine canon (Leder, Toombs, Zaner, Young) is correctly framed as Heinbokel-mediated (Heinbokel note 5). The pages do not make primary claims on these authors.
    • Minor flag: science-as-coherent-deformation provides specific titles (Leder The Absent Body; Toombs Handbook; Zaner The Context of Self; Young Presence in the Flesh) that are not visible in Heinbokel's text — these titles are external general-knowledge attributions. Pages carry confidence: medium and the canon framing is appropriate; acceptable as-is, but a future audit could either remove specific titles or mark them as external attributions.
  • The "crease of speech" silent-key citation is positionally precise and consistently cross-referenced across (a) source page Key Passages, (b) extraction note Pass 3 Part C, (c) coherent-deformation Application Register subsection, (d) schneider-case Methodological-Epistemic Exhibit subsection, (e) science-as-coherent-deformation Key Points and Details, and (f) the case-report-as-coherent-deformation claim entry.
  • Date correction "Heinbokel 2024 → 2021" is consistently propagated. Legacy "2024" string survives only in audit reports and claims.md Meta section under the un-ingested-period framing (correct historical preservation).

Verdict: Heinbokel ingest is a clean ingest from the citation-traceability standpoint.

Batch 6B — Deferred-queue discharge audit-produced content (mostly clean, 1 minor flag)

All audit-produced subsections, Positions, and Sources bullets on 9 pages traced cleanly to extraction notes or source pages. Specific verifications:

  • hyper-reflection.md Position 4 (Saint Aubert) — multi-anchor PASS: EM2 [189]v(2), EM2 [157]v, S(Mont) 252/259, RC55 68, Sorb(SCCE) 228, NMS [106]α, SA-2006 Ch III §2a — all verified verbatim in extraction notes. Most evidence-dense Position on the page.
  • lebenswelt.md Husserl-via-Gurwitsch subsection — PASS on Gurwitsch causal-genealogical thesis (Ch IV §§2–3) and tourbillon substitution at NT 297–298 April 1960.
  • ontological-difference.md Position 3 NLVIaf2 [149] — PASS verbatim.
  • nonphilosophy.md "Foi interrogative" subsection — PASS on three NPVIf [162–163] equations + Claudel "où suis-je" 13-occurrence count + Chouraqui-distinction.
  • schneider-case.md "Schneider's Loss" subsection (Saint Aubert) — PASS on EM3 [247]v(32) + PbPassiv 247/259.
  • freud-without-demonology.md, shadow-philosophy.md, eternal-recurrence.md — Sources bullets all PASS against T&D and Heidegger Nietzsche II extraction notes.
  • seinsgeschichte.md Knight 2024 aquatic-withdrawal bullet — UNCERTAIN. The substantive claim (Heidegger=Hesiod / Levinas=Orphic / MP=Homer cosmogonic mapping; aquatic withdrawal as MP's polemical target against horizonal ontology) is solidly anchored across Knight Intro §2 + Ch 4 §5. The specific citation says "Ch 4 §5"; the physis kryptesthai philei and écart-as-elemental-non-being readings are at Ch 4 §5 in the extraction note, but the explicit comparative claim ("more radical than Heidegger's revealing-concealing pair") is closer to Intro §2 plus Ch 4's general programmatic line than a single §5 sentence. Recommendation: accept; tighten section attribution to "Ch 4 §5 / Intro §2" at next opportunity. Not a fabrication concern.

Verdict: Saint Aubert Priority A integration and ultrareview discharge are well-anchored. The single UNCERTAIN is a section-level granularity issue, not a fabrication risk.

Batch 6C — Tenth-run + weave audit-produced content (mostly clean, 5 corrective flags)

The 2026-05-07 tenth-run stub creation and 2026-05-08 weave hub/register split are well-anchored overall, with five concrete corrective items concentrated in two pages where the consolidating Sources sections sweep across many works:

Priority Page Issue Fix
1 pensee-de-survol.md Sources line 95 "Aspects de l'homme" mislabel — should be Inédits I §Aspects politiques (already cited line 88) or T&D Silverman/Barry editorial framing p. 18 per motifs.md line 190 Re-route to Silverman/Barry T&D editorial framing (the canonical attestation per motifs.md)
2 pensee-de-survol.md Sources line 92 + Motif Weight line 77 PoP 1961 Kosmotheoros polemic unanchored in PoP extraction note (term solidly attested in V&I Working Notes p. 287) Re-route to V&I p. 287 anchor or mark PoP as confidence: speculative
3 capital-as-phenomenology.md Sources line 53 + Key Points line 21 "Marx hégélien d'un bout à l'autre" misattributed — this is Inédits II editor's gloss (raw line 1295), not MP's PoP text. PoP extraction note line 538 has the closest MP-original ("passage from the Phenomenology to the Logic") Re-anchor or restate; current wording implies MP himself wrote the phrase in PoP
4 imaginary-mp.md Key Points line 19 + Sources line 49 La Nature p. 371 "Appearances are the canon" not in Nature extraction note or source page; phrase has Dufourcq/Halák register Targeted raw-source check of La Nature p. 371; if not there, mark speculative or re-route via MC2026 Dufourcq chapter
5 philosophy-of-biology.md (multiple) Nature p. 152 traces only via Halák (MC2026 extraction line 500); attribution chain runs via Halák's quotation, not direct MP-side extraction. Maturana-Varela / Mossio-Bich / Weber-Varela citations correctly framed as "Per Halák's reading in M-C 2026 Ch 5" Acceptable as-is; tighten with direct Nature extraction-note anchor at next audit opportunity

All [[claims#slug]] references on Batch 6C pages match actual claim slugs in wiki/claims.md (verified for 11 slugs on politics-mp.md, 3 on imaginary-mp.md, 4 on philosophy-of-biology.md).

Disposition. Items 1, 2, 3 are mechanical attribution fixes — apply during Phase 8 Step 8.5 batch with citing rationale. Item 4 needs targeted raw-source check; defer to next ingest cycle if raw/ Nature volume is loaded. Item 5 is acceptable; flag for next audit.

Critical finding — wiki/claims.md whole-file duplication (REPAIRED)

Symptom. Phase 6C subagent flagged that claim slugs in wiki/claims.md appear in two contiguous bands (lines ~2964–4578 and ~7587–9494). Investigation confirmed: the entire claim register body was duplicated, including a corrupted middle band where the body of mp-institution-as-stiftung-meets-french-social-thought was truncated mid-prose (around line 4683) by a summary-table-style dump fragment, with the duplicate-block bloch + diacritical entries (lines 4876–4954 of the post-Block-1-deletion file) sitting in the second band that should have been a clean continuation.

Diagnosis.

  • File at commit 68e4b0a (deferred-queue discharge, 2026-05-09 morning): 4886 lines, 113 H2s, 1 summary table — clean.
  • File at commit d87c8f7 (Heinbokel ingest, 2026-05-09 afternoon): 9715 lines, 226 H2s, 1 summary table — duplicated.
  • The Heinbokel ingest commit introduced the duplication. Mechanism likely the ingest workflow's Edit operations on a file at the size threshold, where a non-unique old_string matched at multiple positions or where a previous in-memory version got concatenated into the write. Same-class root-cause hypothesis as the embedded Status History garbling within Block 2 (a separate, smaller-scale instance of the same bug, likely introduced during commit 5917dc4 "Phase 8 latent-claim discharge — 2 candidates from new lint check" when bloch and diacritical entries were duplicated within the same Phase 8 write batch).

Verification before surgery.

  • 110 slugs appeared in BOTH copies; 6 slugs appeared in only one (the second/canonical block).
  • Spot-check of kaushik-stiftung-literary-frame: line 324 (first copy) had Status: live, Updated: 2026-05-05 (STALE pre-deferred-discharge state); line 4942 (second copy) had Status: supported, Updated: 2026-05-09 (CURRENT post-deferred-discharge state).
  • Python audit confirmed: 0 slugs had a later Updated: date in Block 1 than Block 2; 0 slugs were unique to Block 1; 8 slugs were unique to Block 2 (all preserved by the surgery).
  • [[claims#...]] anchors on concept pages would have resolved to the FIRST H2 occurrence per standard Markdown anchor behavior — meaning readers clicking through to claims would have seen STALE statuses (e.g., kaushik as live instead of supported). The surgery fixed this latent rendering bug.

Surgery.

  1. Backup copy preserved at /tmp/claims.md.pre-repair-20260509-023638 (1.4 MB).
  2. Deleted lines 131–4748 of the 9715-line file (stale Block 1 + corrupted middle band): result 5097 lines, 118 H2s, 116 unique slugs (still 2 internal duplicates within the kept block).
  3. Repaired the embedded Status-History corruption inside the kept block: truncated the corrupted ending of coherent-deformation-and-ecart-twin-operators Status History line (which had been spliced with a fragment from mp-institution-as-stiftung-meets-french-social-thought's Payoff section) to its canonical "...via chiasm)." ending — verified against the original at commit b3860e1. Deleted the orphaned fragment + duplicate bloch + duplicate diacritical entries (lines 4869–4954). Result: 5014 lines, 116 H2s, 116 unique slugs, 0 duplicates.
  4. Added 3 missing rows to the summary table for entries that had been added 2026-05-07 (tenth run) and 2026-05-08 (weave) but never indexed: coherent-deformation-and-ecart-twin-operators, dyadic-encounter-presupposes-institution, geological-institution-as-anthropocene-paradigm. Result: 5017 lines.
  5. Final state: summary table 116 entries, H2 entries 116, perfect 1:1 correspondence, no duplicates.

Recommended follow-up beyond this audit.

  • Workflow guard. The ingest workflow should run a post-write integrity check on wiki/claims.md (e.g., grep -c "^## " | uniq -c) before commit. A new lint item: "every claim slug appears exactly once as an H2 heading and exactly once in the summary table." Add to scripts/lint-mechanical.py and to .claude/skills/ingest/SKILL.md self-review checklist.
  • Edit safety. When ingest workflows perform large Edit operations on wiki/claims.md, the old_string parameter must be unique enough that a substring match cannot land at two different positions. The skill should prefer Edit-with-replace_all-disabled and explicit unique-context strings; consider adding a pre-write check that locates old_string in the file and warns if multiple matches are found.
  • Schema-changelog entry. Consider a v0d.6 schema entry explicitly stating the lint invariant for claims.md slug uniqueness, so future maintainers reading the schema cold encounter the rule.

This is a workflow-failure finding, not a routine audit finding. The corruption was latent (claim anchors silently rendered the stale state for hours between commits d87c8f7 and the audit) and could have been propagated further if Phase 8 had continued without diagnosis.

Outcomes

  • 0 fabricated citations detected (Faul "Painter's Courage" pattern not present).
  • 5 mechanical attribution fixes flagged on pensee-de-survol, capital-as-phenomenology, imaginary-mp, philosophy-of-biology — to be addressed during Phase 8 Step 8.5 batch.
  • 1 critical schema-level repair completed: wiki/claims.md whole-file duplication (introduced today, 2026-05-09 afternoon) surgically repaired in this audit phase. Backup at /tmp/claims.md.pre-repair-20260509-023638.
  • Post-repair claims.md state confirmed: 5017 lines | 116 unique slugs | 13 supported / 53 live / 49 candidate / 1 contested / 0 retired (matches the existing index.md count line).
  • Phase 6 verdict: PASS overall, with one critical infrastructure repair plus five mechanical follow-ups queued for Phase 8.

Note on biased sampling (R3)

Per AUDIT_PLAN R3: a clean biased sample (today's audit-produced + freshly-ingested content) is evidence that fabrications aren't concentrated in the just-landed material; it is not certification that the unsampled ~95% of citations are fabrication-free. The whole-file duplication finding shows that biased sampling can also surface critical infrastructure bugs that pre-date the sample window — the duplication was introduced today, but the same-class bug (embedded Status History garbling within Block 2) was introduced 2 days earlier (commit 5917dc4, 2026-05-08). The unsampled corpus may carry analogous infrastructure issues that this audit did not exercise.