🧬 Aging Wiki

2026-06-02

46 min read

2026-06-02 — ad-hoc daily log

[2026-06-02] repo-hygiene | retired ROADMAP.md + surfaced memory-only conventions (user-prompted)

Two governance cleanups prompted by the public/collaborative nature of the wiki:

  1. Surfaced AI-extraction failure-mode conventions that had lived only in private auto-memory into verifying-extraction § “Known AI-extraction failure modes” (repo-facing, read by any agent): verify outcomes not just targets; abstract-only verification can invert an effect-size sign (propagate the fix to MOCs); a verified:true page can still carry the verifier’s own error (adjudicate against the PDF); don’t trust canonical IDs/DOIs asserted from training memory. (Audit found most other conventions were already in schema-history R40/R41, the tracked .claude/agents/ defs, and CLAUDE.md; personal/protocol + private-tooling memories correctly stay out per the leak-gate.)

  2. Retired ROADMAP.md (R50; user agreed it was no longer relevant). It had decayed to ~80% closed-campaign archive (duplicating log/) + a forward queue not touched since ~2026-05-20 despite heavy seeding since; it disclaimed its own authority and carried stale [ ] entries. Seeding is now driven by inbound-count discovery (canonical), ad-hoc, and distributed #stub markers (the missing-page lint).

    • Folded the residual zero-inbound proactive intent (telomere DNA-structure, melanocortin/neurotrophin/cAMP/SPM/melanogenesis clusters, pathways, HSP90/UPR/TAK1 proteostasis cluster, Tyshkovskiy long-tail) + the still-open propagation backlog into new planned-coverage (type: index, excluded from the unsourced-claims lint like ROADMAP was).
    • Repointed references: .claude/agents/wiki-seeder.md description, sops/lint-pass.md (§ Step 3 destination + EXCLUDE_FILES), index.md header, gaps/README.md. Fixed stale per ROADMAP round N pointers on foxo-transcription-factors (→ R7) and mitochondrial-dysfunction (→ R9), and the modality zero-bucket note. Historical ROADMAP mentions in log/ + the R27/R33 lessons left as audit record (they’re the precedent this formalizes).

[2026-06-02] build | organ-system MOC framework + endocrine/vitamin-K atomic pages (user-requested)

Trigger: user asked to build out a *-system MOC for every body system, connect them to existing organs/tissues/cell-types, and stub obvious missing pages — plus seed the specific items thyroid, parathyroid, and vitamin-k-cycle.

New schema (R49): introduced type: organ-system — navigational overlays in organ-systems/ (the anatomical counterpart to type: framework; no verified block). The directory + the parent-system: field already implied this entity category, but it was never defined and the [[cardiovascular-system]]-style links (referenced from tissue parent-system: fields) were all broken. Filename = the parent-system: slug, so those links now resolve. Documented in CLAUDE.md (type: organ-system block + glands-live-in-tissues note) and schema-history R49.

11 organ-system MOCs created (in organ-systems/): cardiovascular-system, musculoskeletal-system, integumentary-system, nervous-system (rich); hematopoietic-system, immune-system (subsumes lymphatic via alias); endocrine-system (developing); digestive-system (microbiome-led); urinary-system, respiratory-system, reproductive-system (planned/mineral-axis-anchored). Each links existing atomic pages + a Dataview rollup of its tissues by parent-system:, and lists marked #stub planned pages. Top-level index: by-organ-system (type: framework, single Dataview cross-system table). Wired into index (anatomical-overlay pointer + new “Organ systems, tissues & cell types” section).

6 high-value organ stubs seeded (type: tissue, #stub, parent-system set, so the Dataview rollups surface them): kidney (urinary; FGF23/Klotho/phosphate hub), liver (digestive), lung (respiratory), pancreas (endocrine/digestive), thymus (immune — thymic involution), spleen (immune/hematopoietic).

3 atomic pages seeded + verified same-day (parallel wiki-seeder → wiki-verifier):

  • tissues/thyroid.md (type: tissue, parent-system endocrine) — TRUST RCT (Stott 2017, full-PDF-verified: SCH levothyroxine null), NHANES III prevalence, age-rising TSH set-point. Verifier fixed a nodule-malignancy figure (4–5% → 23.6% resected-denominator).
  • tissues/parathyroid.md (type: tissue, parent-system endocrine) — PTH/CASR, secondary hyperparathyroidism in aging, FGF23–Klotho–PTH axis. Verifier caught an inverted PTH-paradox mechanism (Jilka 2007: cAMP-PKA is the intermittent-PTH anti-apoptotic pathway, not the continuous-PTH one) + Farrell 2018 study-design/quantitative corrections.
  • pathways/vitamin-k-cycle.md (type: pathway) — brings the K1/K2/MK-7/menadione forms together on the GGCX↔VKORC1 redox cycle; Reactome R-HSA-6806664; UBIAD1 endogenous MK-4 route (Nakagawa 2010, full-PDF-verified; verifier corrected a UBIAD1 tissue-expression claim “arterial wall” → pancreas/brain/testis). Propagated: cleared the stale vitamin-k-cycle #stub note on ggcx.

Leak-gate: the vitamin-k-cycle verifier wrote an absolute local filesystem path (username + mount + private-project name) into its log entry — scrubbed to “local full-text PDF read end-to-end”. Gate re-run clean over all new + tracked files. (The path appeared only in the log note, never in a wiki page.)

Remaining (low-demand) stubs surfaced by the new MOCs: glands (adrenal, pituitary, hypothalamus, gonads/ovary/testis), lymphoid (lymph-nodes, tonsils), respiratory/renal/repro organs, and many cell types (T/B-cells, hepatocytes, etc.) — all marked #stub in the MOC “Missing/planned” sections; tracked by the missing-page lint as the next seeding queue.

[2026-06-02] ingest+verify | vitamin-K-cycle enzymes (ggcx, vkorc1) — bone/vascular cluster close-out

Trigger: user “continue” — the final tier of the lint-sweep ranking. The two vitamin-K-cycle enzymes referenced from matrix-gla-protein and vitamin-k. Parallel wiki-seeder ×2 → wiki-verifier ×2 → propagation. Both flipped verified: true (individual verify entries below).

Pages seeded + verified:

  • molecules/proteins/ggcx.md — type: protein. γ-glutamyl carboxylase (UniProt P38435). Carboxylates all VKDPs incl. MGP; consumes reduced vitamin K. druggability-tier 4 (undruggable for aging; axis is drugged upstream via vitamin K substrate); mr-causal-evidence not-tested.
  • molecules/proteins/vkorc1.md — type: protein. Vitamin K epoxide reductase (UniProt Q9BQB6); filename is the HGNC symbol, with vkor/VKOR aliases so existing [[vkor]] links resolve. Warfarin target. druggability-tier 3 with the warfarin paradox documented (warfarin engages it but is harmful to the vascular-aging axis → no aging-beneficial drug); mr-causal-evidence partial.

Verifier-caught fabrications (corrected; none propagated — sibling pages grep-checked clean):

  • ggcx: De Vilder 2017 patient count (36 → actual 47); two wrong N-glycosylation residues; an unsourced PDB ID “9BUM” removed; carbamylation targets corrected (arginine added); TM-helix count reconciled (5 biochemical vs 9 cryo-EM per Wu 2025).
  • vkorc1: TM topology (4 → 3 helices per Rost 2004); wrong SNP (rs2884737 → rs8050894 in Holden 2014); D’Andrea 2005 n (297 → 147) + SNP clarified (1173C>T, in LD with rs9923231, not the promoter SNP directly); Nuotio 2021 ORs given full CIs (ICA 18.27 with a real near-separation CI; CCA 2.64).

Propagation (gap markers cleared — both enzymes now exist; vkor resolves to vkorc1 via alias):

Leak-gate: clean (also fixed a self-referential hit — the previous batch’s log note had quoted the forbidden CLI verb verbatim while describing its own fix; reworded).

Cluster status: the bone/vascular axis (osteoblasts/osteoclasts/osteocytes/bone/osteoporosis/runx2/sost/dkk1/lrp5-lrp6/romosozumab/mgp/fgf23/klotho/vascular-calcification/hyperphosphatemia/vitamin-k/ggcx/vkorc1) is now fully seeded + verified. Remaining stubs in this cluster: vitamin-k-cycle (pathway page), parathyroid, pit-1-slc20a1/pit-2-slc20a2, adventitial-fibroblasts, aortic-aneurysm, cardiovascular-system MOC — all low inbound demand.

[2026-06-02] ingest+verify | gap-fill batch (mcclung-2019-loft, hyperphosphatemia, vitamin-k, splicing-dysregulation)

Trigger: user-requested ad-hoc batch, continuing the gap-ranking from the 2026-06-02 lint sweep (lint). Parallel wiki-seeder ×4 → wiki-verifier ×4 → propagation. All four flipped verified: true same day (individual verify entries below).

Pages seeded + verified:

  • studies/mcclung-2019-loft.md — type: study. LOFT (odanacatib Phase 3, n=16,071; Lancet Diabetes Endocrinol 2019; doi:10.1016/S2213-8587(19)30346-8). Canonical home for the trial the osteoclasts page cites. The LOFT-alone stroke HR 1.32 (95% CI 1.02–1.70) was confirmed present in the structured abstract (verifier) — so the osteoclasts.md citation needs no correction. verified:true at abstract scope (closed-access full text).
  • phenotypes/hyperphosphatemia.md — type: phenotype. Causal-graph-node framing (intermediate phosphate-driver, not a classic aging phenotype). Fills the [[hyperphosphatemia]] gap on vascular-calcification/fgf23/klotho.
  • molecules/compounds/vitamin-k.md — type: compound (family page: K1 vs K2/MK-7). Fills the matrix-gla-protein substrate gap; honest human-evidence-level: limited (biomarker/surrogate-level for vascular/bone). literature-checked-through 2026-06-02.
  • processes/splicing-dysregulation.md — type: process (alias alternative-splicing). Canonical home for age-related spliceosome biology contextualizing sf3b1; framed as hallmark-adjacent (NOT a López-Otín 12 hallmark).

Schema fix (de-facto practice undocumented): type: phenotype carries verification discipline — 27/28 phenotype pages have a verified: block — but CLAUDE.md’s “AI-extracted vs human-verified” type list and the type: phenotype frontmatter block both omitted it. The seeder followed the literal (incomplete) schema and shipped hyperphosphatemia without the block. Fixed: added phenotype to the required-types list AND added the four-field verified block to the type: phenotype schema template in CLAUDE.md; added the block to hyperphosphatemia.md (verifier then flipped it).

Verifier-caught fabrications (all corrected; none propagated — each verifier grep-checked sibling pages):

  • vitamin-k: sign error on Diederichsen 2022 aortic-valve-calcification (“+17 AU” → actual treatment effect −17 AU, non-sig, p=0.64) + missing 720 µg/day dose. Confined to vitamin-k.md.
  • splicing-dysregulation: fabricated compound “H3BS-10000” (Latorre 2017 actually tested resveratrol + numbered stilbene analogues) and quercetin/fisetin misattributed to Latorre 2017 — both removed; splicing-factor panel corrected to the real 20-factor list; Pabis 2024 mouse intron-retention over-claim corrected; “~one-third” → 38% (InCHIANTI)/29% (SAFHS). Scaffidi 2006 found to be PMC-OA (gap tag removed).
  • hyperphosphatemia: fabricated Block 2004 RR (1.27, CI 1.18–1.37, comparator 2.4–6.5 mg/dL — none in source) → replaced with the paper’s actual dose-response RR ladder (1.07→2.02); “patient-years”→“patients”; McGovern 2013 comparator band + Larsson 2010 non-CV-mortality claim corrected.
  • mcclung-loft: stroke-HR gap-note corrected (the seeder wrongly said 1.32 was unconfirmable; it IS in the abstract).

Propagation (gap markers cleared now that pages exist):

Leak-gate: a verifier wrote a paper-archive CLI verb into this log (matched a forbidden token) — reworded to “local PDF unavailable”. Gate now clean.

Remaining bone/vascular-cluster gaps (next tier): ggcx, vkor (vitamin-K-cycle enzymes, 1 inbound each); osteoblasts-cluster fully closed.

[2026-06-02] verify | phenotypes/hyperphosphatemia.md

Page verified: phenotypes/hyperphosphatemia.md · type: phenotype

Sources checked:

  • Kuro-o 2021 (doi:10.5551/jat.RV17045) — full PDF verified end-to-end (11 pages; JAT 28:203–213)
  • McGovern 2013 (doi:10.1371/journal.pone.0074996) — full PDF verified end-to-end (PLoS One OA; PMC3769279)
  • Larsson 2010 (doi:10.1161/ATVBAHA.109.196675) — full PDF verified end-to-end (ATVB 30:333–339; downloaded via AHA journals)
  • Block 2004 (doi:10.1097/01.ASN.0000133041.27682.A2) — abstract-only; paper is not_oa no-fulltext-access
  • Kuro-o 2010 (doi:10.1016/j.mad.2010.02.008) — abstract-only; paper is not_oa no-fulltext-access
  • Isakova 2011 (doi:10.1038/ki.2011.47) — abstract-only; PMC PDF fetch failed (1817 bytes redirect); abstract verified via PubMed efetch (PMID 21389978); PMC3134393

Corrections made:

  1. Block 2004 body text — fabricated single-RR comparison removed (critical): “serum phosphate >6.5 mg/dL (>2.1 mmol/L) was associated with a relative risk of death of 1.27 (95% CI 1.18–1.37) compared to 2.4–6.5 mg/dL” → replaced with the dose-response gradient that the abstract actually states: adjusted RR 1.07, 1.25, 1.43, 1.67, 2.02 for bands 5.0–6.0, 6.0–7.0, 7.0–8.0, 8.0–9.0, ≥9.0 mg/dL above the >5.0 threshold. The specific 1.27 RR, the CI 1.18–1.37, and the “2.4–6.5 mg/dL” comparator range do not appear in the abstract and could not be verified. Added no-fulltext-access for full CI table.
  2. Block 2004 footnote — “patient-years” → “patients”: “n=40,538 patient-years” → “n=40,538 maintenance hemodialysis patients.” The paper describes 40,538 patients, not patient-years.
  3. Block 2004 footnote — dose-response RRs added: footnote now reflects the full dose-response gradient from the abstract instead of the single unverifiable 1.27 value; population-attributable risk (17.5%) added from abstract.
  4. McGovern 2013 body text — comparator range corrected: “compared to 0.87–1.13 mmol/L” → “compared to the 0.75–1.00 mmol/L reference band.” The paper uses bands <0.75, 0.75–1.00, 1.00–1.25, 1.25–1.50, >1.50 mmol/L; the reference group is 0.75–1.00. The “0.87–1.13” range does not appear in the paper.
  5. McGovern 2013 body text — ORs added: added the actual OR values (1.36, 1.40) and CIs from Table 4 for the 1.25–1.50 vs 0.75–1.00 comparison in normal and CKD 1–2 subgroups.
  6. McGovern 2013 footnote — comparator and subgroup n corrected and enriched: same comparator correction plus subgroup Ns added (normal=24,184; CKD 1–2=20,356; CKD 3–5=13,292) and the CKD 3–5 hyperphosphatemia OR added (2.34, 1.64–3.32).
  7. Larsson 2010 footnote — non-cardiovascular mortality claim corrected: removed implication that Pi independently predicted non-cardiovascular mortality. In the fully adjusted model B, Pi was NOT a significant predictor of non-cardiovascular mortality (HR 1.02; p=ns). Added confirmed HRs for total and cardiovascular mortality with CIs and p-values.
  8. Isakova 2011 footnote — n added, stage caveat added: n=3,879 added; stage claim (FGF23 rises at stage 1–2) qualified as extrapolated beyond the study’s CKD stage 2–4 population.
  9. Kuro-o 2021 footnote — CPP framing sharpened: added clarification that CPPs (not free phosphate ions) are described as the proximate pathological mediator per Kuro-o 2021.
  10. Kuro-o 2010 footnote — phenotype details added from abstract: glucose metabolism/insulin sensitivity/oxidative stress effects noted; no-fulltext-access clarified to specify what requires full text.
  11. Removed ⚠️ auto-extraction banner; flipped verified: true with scope description.

Unverifiable claims:

  • Block 2004: specific CI table for each phosphate band; Ca×P product threshold (>72 mg²/dL²) and its specific risk; full dose-response above >5.0 mg/dL endpoint detail — require full text
  • Kuro-o 2010: specific phosphate restriction protocol (level, duration, strain data) — require full text
  • Isakova 2011: eGFR threshold slopes for FGF23 vs PTH divergence (described in abstract qualitatively but not numerically) — require full text

Supersession check: Not required for type: phenotype per CLAUDE.md (canonical-data-stable type). No literature-checked-through: field on phenotype pages.

Downstream propagation needed:

  • None required. The body-text errors in this page were self-contained. The corrections to Block 2004 (RR description) and McGovern 2013 (comparator band) do not appear to be cited verbatim on vascular-calcification, fgf23, or klotho — those pages cite the studies via their own footnotes. If those pages carry the “1.27 (1.18–1.37)” or “0.87–1.13 mmol/L” values, they should be corrected. Main agent should grep for these strings across those pages.

[2026-06-02] ingest | hyperphosphatemia phenotype page

Trigger: user-requested ad-hoc seed; explicit brief. Fills #gap/needs-page for [[hyperphosphatemia]] tagged on processes/vascular-calcification.md, molecules/proteins/fgf23.md, and molecules/proteins/klotho.md.

Page seeded: phenotypes/hyperphosphatemia.md — type: phenotype, ICD-10-CM E83.39 (confirmed via NLM ICD-10-CM API). ICD-11 pending (WHO API auth required; tagged #gap/needs-canonical-id).

Framing: page positioned as a causal-graph node and intermediate biochemical driver rather than classical symptom-defined phenotype; opening paragraph makes this explicit. Owns phosphate homeostasis physiology, FGF23-Klotho-PTH-vitamin D network, clinical definition/thresholds, causes, downstream pathology pointers, and management. Defers VSMC osteogenic mechanism entirely to vascular-calcification.

Hallmarks chosen: [[deregulated-nutrient-sensing]] (phosphate is a nutrient-sensing signal; FGF23-Klotho-PTH-vitamin D is the dedicated mineral endocrine axis; age-related axis degradation is nutrient-sensing deregulation) + [[altered-intercellular-communication]] (FGF23 is an osteocyte-to-kidney endocrine signal; age-related Klotho decline breaks this signaling fidelity). Both hallmark pages confirmed to exist in hallmarks/.

Key citations (all confirmed via PubMed esummary / Crossref):

  • Kuro-o 2021 JAT doi:10.5551/jat.RV17045 — phosphate as pro-aging factor; CPP mechanism; FGF23-Klotho framing (~67 citations)
  • Kuro-o 2010 Mech Ageing Dev doi:10.1016/j.mad.2010.02.008 — Klotho-null mouse phosphate-rescue experiment
  • Block 2004 JASN doi:10.1097/01.ASN.0000133041.27682.A2 — n=40,538 hemodialysis; serum phosphate–mortality dose-response (~2,076 citations; landmark)
  • Isakova 2011 Kidney Int doi:10.1038/ki.2011.47 — FGF23 rises before PTH and phosphate in CKD progression (~1,067 citations)
  • McGovern 2013 PLoS One doi:10.1371/journal.pone.0074996 — n=57,832 community cohort; phosphate-CV-event association in CKD and non-CKD subgroups (~114 citations)
  • Larsson 2010 ATVB doi:10.1161/ATVBAHA.109.196675 — n=2,176 Swedish men; 30-year follow-up; serum phosphate predicts CV + total mortality in community (~119 citations)
  • Calvo 2023 Nutrients doi:10.3390/nu15030736 — cited via phosphate-additive-reduction (verified 2026-05-23); bioavailability differential framing

Archive status: Kuro-o 2010 = not_oa; Block 2004 = not_oa; others = download pending. All DOIs confirmed via PubMed esummary + Crossref; no fabricated DOIs.

Implicit stubs created (new wikilinks to non-existent pages): [[arteries]] (tissue), [[parathyroid]] (tissue/cell-type), [[pit-1-slc20a1]] (protein), [[pit-2-slc20a2]] (protein).

Gap markers cleared: #gap/needs-page for [[hyperphosphatemia]] on vascular-calcification.md (line 79 tag), fgf23.md caused-by: field, and klotho.md implicit references.

Verifier priority: Block 2004 not_oa + Kuro-o 2010 not_oa → verify key RR values against abstracts; Isakova 2011 and McGovern 2013 are gold/pending OA.

[2026-06-02] ingest | bone-cluster seeding (osteoblasts, osteoclasts, romosozumab)

Trigger: follow-on to the 2026-06-02 stale-gap lint sweep (see lint). That sweep ranked the remaining real page-existence gaps by inbound demand; user requested seeding “osteoblasts and the rest of the bone cluster, then verify.” Parallel wiki-seeder batch of 3 + per-page wiki-verifier pass.

Pages seeded (all verified: false on draft → flipped verified: true same day by verifier subagents; see verify entries below):

  • cell-types/osteoblasts.md — type: cell-type, CL:0000062. Highest inbound demand (9 distinct pages). Bone-forming mesenchymal cell; RUNX2/SP7 commitment, RANKL/OPG production, Wnt/sclerostin regulation, aging decline + senescence.
  • cell-types/osteoclasts.md — type: cell-type, CL:0000092. Hematopoietic monocyte/macrophage-derived bone-resorbing cell; RANKL/RANK/OPG axis, resorption machinery, estrogen-withdrawal hyperactivity.
  • molecules/compounds/romosozumab.md — type: compound (biologic mAb), ChEMBL2107874. Anti-sclerostin; completes the sost cluster (FRAME/ARCH/BRIDGE, CV boxed warning). Seeder added a sclerostin-inhibition / wnt-pathway-activation mechanism class to intervention-classes.

bmsc resolved by aliasing, NOT a new page: the ranking listed bmsc (2 inbound) as a gap, but mesenchymal-stem-cells already covers BM-MSCs as its primary source — including the exact osteoblast/adipocyte fate-skewing fact [[bmsc]] was referenced for (§ Osteoblast/adipocyte fate skewing). Per the atomic-entity model (one canonical home), added bmsc/BMSC/BMSCs/bone marrow stromal cells/bone marrow mesenchymal stem cells to the MSC page aliases and repointed the [[bmsc]] links (bone.md, osteoporosis.md) to [[mesenchymal-stem-cells|BMSCs]]. No duplicate page created.

Gap-marker cleanup (osteoblasts/osteoclasts now exist): removed/rewrote #gap/needs-page for osteoblasts/osteoclasts/bmsc across bone.md, sost.md, dkk1.md, lrp5-lrp6.md, runx2.md, osteocytes.md, osteoporosis.md.

Downstream propagation (resolved):

  • The osteoblasts verifier flagged a possible Farr 2016 SASP-misattribution on osteocytes.md / cellular-senescence.md / senolytics — checked, none cite Farr 2016 (jbmr.2892); the misattribution was confined to the newly-seeded osteoblasts page and is fixed there. No further action.
  • The osteoclasts odanacatib “23%” fabrication was confined to osteoclasts.md (grep-confirmed absent from osteoporosis.md/bone.md); corrected to 32% (HR 1.32, LOFT/McClung 2019).
  • The romosozumab verifier flagged that already-verified sost.md (2026-05-23) carried the same wrong Zheng 2023 MR DOI (10.1002/art.42592) and an imprecise FRAME T-score footnote. Independently confirmed PMID 37096546 → 10.1002/art.42538 via PubMed efetch + Crossref; fixed the DOI on sost.md and enriched the Cosman/FRAME footnote to specify “total hip or femoral neck” eligibility. (A verified page carrying a verifier-era error — the feedback_verifier_introduced_error pattern.)

Follow-up candidate (not done): osteoclasts verifier suggested seeding studies/mcclung-2019-loft.md (LOFT, n=16,071) as a dedicated study page given its weight; left for a future batch.

[2026-06-02] verify | cell-types/osteoblasts.md

Page verified: cell-types/osteoblasts.md · type: cell-type

Sources checked:

  • Farr 2016 (doi:10.1002/jbmr.2892, PMCID PMC5289710) — full text verified via eutils efetch XML end-to-end
  • Kusumbe 2014 (doi:10.1038/nature13145, PMCID PMC4943525) — full PDF verified (PMC author manuscript downloaded)
  • Xu 2018 (doi:10.1111/acel.12794, PMCID PMC6052401) — full PDF verified (downloaded)
  • Komori 1997 (doi:10.1016/s0092-8674(00)80258-5) — full PDF verified (local archive); consistent with runx2-page pass 2026-05-23
  • Simonet 1997 (doi:10.1016/s0092-8674(00)80209-3) — full PDF verified (local archive)
  • Boyle 2003 (doi:10.1038/nature01658) — full PDF verified (local archive)
  • Manolagas 2010 (doi:10.1210/er.2009-0024) — abstract only; full PDF unavailable (HTTP 403/500)
  • Seeman 2003 (doi:10.1007/s00198-002-1340-9) — abstract only; closed-access (not_oa) no-fulltext-access
  • Farr 2019 (doi:10.1016/j.bone.2019.01.015) — not re-read; review not directly cited for quantitative claims

Corrections made:

  1. SASP attribution (critical): The page claimed senescent osteoblast progenitors and mature osteoblasts “express a SASP that includes RANKL, IL-6, IL-8, MMP-3, and MMP-13.” Farr 2016 full text explicitly states the opposite: “relatively few SASP factors were altered significantly with aging in osteoblast progenitors and osteoblasts (GSEA p values > 0.05).” The SASP in the bone niche comes predominantly from osteocytes and myeloid cells (23/36 SASP genes elevated in osteocytes, GSEA p<0.001; 26/36 in myeloid cells). All three locations (body §3, hallmark table, chronic-inflammation hallmark row) corrected to reflect osteocytes as the dominant SASP source.
  2. n values (Farr 2016): “n=5–10/group” → n=12 young (6 mo) / n=10 old (24 mo) males for main rt-qPCR cohort; n=4/group for SADS assay; n=3/group for cultured osteocytes.
  3. Sorting strategy (Farr 2016): “FACS-sorted” → “MACS-sorted” (magnetic-activated cell sorting, not fluorescence-activated); osteoblast progenitors = Lin−/Lepr+; mature osteoblasts = AP+/CD31/34/45/54−.
  4. Kusumbe 2014 vessel marker notation: “EMCN+CD31+” → “CD31^hi/Emcn^hi” (the paper defines Type H as the high-expressing subset; low-expressing sinusoidal Type L are EMCN+CD31+ at lower levels; the distinction is quantitative not binary).
  5. Kusumbe 2014 pharmacological restoration: “PDGF-BB or Notch signalling” → deferoxamine mesylate (DFM) — the actual intervention tested (PHD inhibitor stabilising HIF-1α; 15 mg/ml i.p. every alternate day for 4–5 weeks). PDGF-BB is not an intervention in this paper (PDGFRβ is a marker of mesenchymal cells around Type H vessels). Notch is cited from a companion manuscript as another regulator, not a drug tested in this paper.
  6. Xu 2018 organism: “mouse” → rat as primary model; human BMSC data also presented.
  7. Seeman 2003 “fourth decade” claim: qualified to “young adulthood (commonly cited as beginning in the third–fourth decade)” with no-fulltext-access tag; the abstract says “young adult women and men” but does not specify the decade.
  8. Removed ⚠️ auto-extraction banner; flipped verified: true with detailed verified-scope.

Unverifiable claims:

  • Manolagas 2010 full review body text (FoxO/β-catenin mechanism accurately represented per abstract; specific figure-level claims not verifiable without full PDF)
  • Seeman 2003 “fourth decade” onset (closed-access)

Supersession check (R25): Cell-type page — supersession check not required per CLAUDE.md. Light literature check confirmed post-2021 bone senescence papers (PMID 40698069, 38521178) continue to support the osteocyte-centric SASP framework; no supersession detected.

Downstream propagation needed:

  • osteocytes.md — may need updating if it cites “osteoblast SASP” based on Farr 2016; verify that osteocytes page correctly identifies osteocytes as dominant SASP source
  • hallmarks/cellular-senescence.md — if it summarizes Farr 2016 as “osteoblast SASP,” correct to osteocyte-dominant
  • interventions/pharmacological/senolytics.md — if it attributes bone SASP to osteoblasts, update

[2026-06-02] verify | cell-types/osteoclasts.md

Page verified: cell-types/osteoclasts.md · type: cell-type

Sources checked:

  • McClung 2019 (PMID 31676222, LOFT trial primary report) — full abstract verified via PubMed
  • Farr 2023 JCI (doi:10.1172/JCI162519) — full PDF verified end-to-end
  • Cheng 2022 (doi:10.3390/ijms23031376) — full PDF verified
  • Wang 2023 (doi:10.3390/ijms24065814) — full PDF verified
  • Luo 2025 (doi:10.3389/fimmu.2025.1611932) — full PDF verified
  • Miyamoto 2006 DC-STAMP (doi:10.1007/s10165-006-0524-0) — abstract only (not_oa)
  • Drake 2017 (doi:10.1210/er.2015-1114) — abstract only (PMC full text publisher-blocked)
  • Kuno 2018 (doi:10.1007/s00424-018-2137-9) — abstract only (not_oa)
  • Yang 2012 (doi:10.1002/jbmr.1623) — unavailable (download failed, no OA URL)
  • Rogers 2011 (doi:10.1016/j.bone.2010.11.008) — unavailable (not_oa)

Corrections made:

  1. Odanacatib stroke risk: “23% increased risk” → “32% increased risk (HR 1.32, 95% CI 1.02–1.70, p=0.034)” — sourced from primary LOFT trial (McClung 2019). The “23%” figure was fabricated by the seeder.
  2. Odanacatib mechanism attribution: “atrial fibrillation / thromboembolic mechanism” removed — AF was NOT significantly elevated in LOFT (HR 1.18, p=0.24); mechanism remains unresolved.
  3. LOFT trial details added: trial name spelled out, n=16,071, citation updated to include primary report [^mcclung2019loft] alongside Drake 2017 review.
  4. DC-STAMP KO severity: “severely impaired bone-resorbing activity” → “reduced bone-resorbing activity and an osteopetrotic bone phenotype” — Miyamoto 2006 abstract says “reduced” not “severely impaired”; adds the osteopetrosis phenotype per abstract.
  5. Resorption pit pH: “pH ~4.5” → “pH ~4–4.5” with gap tag added; Kuno 2018 abstract confirms acid-inducible H+-leak activates at pH <5.5 (consistent) but does not explicitly state ~4.5 in abstract; exact value tagged no-fulltext-access.
  6. Farr 2023 JCI footnote: expanded with verified specifics from full PDF (n per group, Figure references, transplant protocol, exact Rankl finding per model arm).
  7. Luo 2025 first author corrected: “Luo Y” → “Luo J” (per PDF cover page).
  8. Added McClung 2019 as new primary citation [^mcclung2019loft] with full trial data.
  9. Removed ⚠️ auto-extraction banner; flipped verified: true.

Cell Ontology: CL:0000092 confirmed = “osteoclast” via OLS4 API.

Supersession check (R25): Cell-type page type — supersession search not required per CLAUDE.md (canonical-data-stable type). No literature-checked-through field required.

Unverifiable claims: resorption pit pH ~4.5 (Kuno 2018 full text not accessible); V-ATPase subunit specifics (Yang 2012 not accessible); bisphosphonate farnesyl-PP-synthase mechanism (Rogers 2011 not accessible; mechanism is consensus pharmacology); RANKL/OPG functional review (Boyce 2008, Yasuda 2021, Udagawa 2021, Ono 2018 not accessible).

Downstream propagation candidates: osteoporosis.md (odanacatib stroke percentage may be cited there); osteoblasts.md (not directly affected). Main agent should check osteoporosis.md for any “23%” odanacatib stroke claim.

ingest+verify | interventions/procedural/microneedling.md (user-requested)

Trigger: user asked whether the wiki covers micro-needling as a skin intervention (“interested in physical skin interventions, but also skeptical”). No dedicated page existed — microneedling was covered only as one of seven bundled modalities on dermatologic-resurfacing, and granular sub-modality pages were explicitly deferred at R44 (ROADMAP line 471–472). Seeded the dedicated page per user request.

Page: interventions/procedural/microneedling.md · type: intervention · mode: procedural (first content page in interventions/procedural/ besides venous-valve-reconstruction; uses the R44-formalized procedural mode rather than the legacy mode: pharmacological workaround the sister resurfacing page still carries).

Evidence appraisal: the page foregrounds the mechanism-vs-efficacy distinction and grades each claim on its own evidence (neutral voice — the user noted personal skepticism, but the page reads like any other intervention page). Core findings:

  • Mechanism (wound-healing → neocollagenesis) is histologically real but rests on El-Domyati 2015, n=10, single-arm, no control/sham (collagen I/III/VII + tropoelastin ↑, total elastin ↓ — the elastin drop = solar-elastosis clearance, desirable).
  • Headline aging evidence (Foppiani 2025) is a proportion meta-analysis of patient satisfaction (83%, no control arm, Level of Evidence IV); authors themselves flag absence of standardized esthetic outcome measures.
  • Evidence-hierarchy inversion: the only RCT-grade evidence (Shen 2022, 12 RCTs/414 pts, LoE III) is for atrophic acne scars, not aging; even there subjective satisfaction was non-significant vs comparators and RF-microneedling underperformed non-RF.
  • Central confound: most rejuvenation studies bundle MN with PRP / topical vit-C / growth factors / exosomes; needling enhances transdermal delivery of those agents, so monotherapy effect size for photoaging is essentially unknown.
  • No blinded sham-controlled RCT; no biological-age/epigenetic-clock endpoint ever measured; durability >few months unknown.
  • Genuine strong point = safety: chromophore-independent → low PIH risk in Fitzpatrick IV–VI (Chao 2023; Shen 2022 reported no induced PIH); low cost/downtime. Home <0.5 mm dermarollers flagged as not reaching dermis + infection risk.

Verification: abstract scope. All cited primaries confirmed not_oa in local archive. El-Domyati 2015 (PMID 26096653), Foppiani 2025 (40542236), Shen 2022 (35426044), Jaiswal 2024 Cureus (39449889, PMC OA) abstracts read directly from PubMed efetch. Robati 2020 / Vassão 2022 / Chao 2023 footnotes inherited from the verified (2026-05-19) resurfacing page. Per-isotype % effect sizes in El-Domyati below abstract granularity → flagged #gap/needs-fulltext. verified: true with explicit abstract-scope note.

Data pulled live: ClinicalTrials.gov v2 — microneedling AND (rejuvenation/photoaging/wrinkles/skin aging), RECRUITING (4) + ACTIVE_NOT_RECRUITING (1) = clinical-trials-active: 5 (2026-06-02; query-term-sensitive — many registered MN trials are drug-delivery/alopecia/acne-scar not aging).

Propagation: backlink added from dermatologic-resurfacing Cross-References. Not added to ROADMAP (ad-hoc user seed per standing convention).

Link targets confirmed present: col1a1, col3a1, eln, wound-healing, skin-aging, dermis, epidermis, dermal-fibroblasts, chemical-peels, retinoids, uv-protection, loss-of-proteostasis, cellular-senescence, dermatologic-resurfacing, sops/integrating-clinical-trials.

Gaps surfaced: no blinded sham-controlled monotherapy RCT for aging (#gap/needs-replication); no biological-age endpoint (#gap/no-mechanism at clock level); protocol non-standardization (#gap/dose-response-unclear); durability (#gap/long-term-unknown); combination-therapy confound (#gap/contradictory-evidence).

[2026-06-02] verify | studies/mcclung-2019-loft.md

Page verified: studies/mcclung-2019-loft.md (type: study; McClung 2019 LOFT trial, Lancet Diabetes Endocrinol, PMID 31676222).

Source status: Closed-access (Lancet Diabetes Endocrinol; archive status: not_oa). Full-text PDF unavailable. Verification performed against PubMed structured abstract via efetch (PMID 31676222).

Critical finding — LOFT-alone stroke HR confirmed in abstract: The seeder believed HR 1.32 (95% CI 1.02–1.70, p=0.034) was body/table-only and not in the abstract. This is incorrect — the PubMed efetch structured abstract explicitly states “1·7% [136/8043] vs 1·3% [104/8028], HR 1·32, 1·02–1·70; p=0·034” for LOFT-alone stroke. The figure is abstract-confirmed. The no-fulltext-access note on this specific number is removed.

Corrections made:

  • Stroke table: LOFT-alone row restructured with confirmed event counts (136/8,043 vs 104/8,028); row ordering fixed (LOFT-alone first, combined second); the ambiguous merged-cell formatting in the seeder’s table replaced.
  • AF HR: CI 0.90–1.55 added from abstract (seeder recorded HR 1.18 p=0.24 but omitted the CI).
  • Gaps section: no-fulltext-access narrowed — no longer covers the load-bearing stroke HR or fracture HRs (all abstract-confirmed); retained only for adjudication methodology and Extension survivor-bias handling (not in abstract).
  • Footnote updated: removed “(not confirmed abstract-only; see verifier note)”; now states both stroke HRs and AF HR are abstract-confirmed.
  • Banner removed; verified: true set.
  • literature-checked-through: 2026-06-02 added.

Supersession check (R25): Post-2019 PubMed search on odanacatib/cathepsin K inhibitors found no new Phase 3 RCT or meta-analysis overturning LOFT conclusions. PMID 33200257 (2021 men’s RCT) found no CV harm in men but does not supersede the women’s LOFT findings. PMID 42181204 (2026 NMA in men) does not address stroke safety. No supersession.

Downstream propagation: The LOFT-alone stroke HR 1.32 on osteoclasts footnote [^mcclung2019loft] is confirmed correct — no correction needed there. The osteoclasts page already has the correct HR, CI, and p-value.

[2026-06-02] verify | molecules/compounds/vitamin-k.md

Page verified: molecules/compounds/vitamin-k.md (type: compound, family page)

Sources checked (4 local PDFs + 1 abstract):

  • Diederichsen 2022 (Circulation 145:1387–1397) — local PDF verified
  • Li/Wang 2023 (Front Nutr 10:1115069) — local PDF verified
  • Murali 2023 (JAHA 12:e031676) — local PDF verified
  • de Vries 2025 (Nutrients 17:815) — local PDF verified
  • Knapen 2015 (Thromb Haemost 113:1135) — PubMed abstract only (#gap/no-fulltext-access)

Corrections made:

  1. Diederichsen 2022 sign error (critical): Wiki body said “mean difference +17 AU” — corrected to “treatment effect −17 AU” (MK-7 progression +275 AU vs placebo +292 AU; CI −86 to +53; p=0.64). The “+17” implied treatment was worse; the actual (non-significant) direction slightly favours MK-7. Propagated to footnote and human evidence summary table.
  2. Diederichsen 2022 dose added: Wiki omitted the MK-7 dose. Added “720 µg MK-7/day + 25 µg vitamin D” to body and footnote — important because 720 µg is 4× the Knapen/de Vries dose, making under-dosing an implausible null explanation.
  3. Diederichsen footnote page range added: 1387–1397 added.
  4. Li 2023 dp-ucMGP CI added to footnote: was “p=0.0001” alone; now “(95% CI −366.08 to −120.53; p=0.0001; I²=71%)”.
  5. Li 2023 total participants clarified in footnote: 1,533 in analysis (1,842 enrolled initially).
  6. Murali 2023 footnote enriched: median participants/duration range added (154, range 68–388; median 12 mo); “1 RCT achieved primary endpoint” phrasing precise.
  7. de Vries 2025 footnote precision: full trial n (243) vs post-hoc cohort (165) distinction added; Young’s modulus values confirmed from Table 2 (placebo +49.1%±77.4, MK-7 +9.4%±67.1); n=26 confirmed as MK-7 arm of post-menopausal high-SI subgroup.
  8. Knapen 2015 footnote precision: author names corrected to match abstract (Braam LA not LAJLM, Hoeks AP not APG); above-median stiffness threshold confirmed (SI β median 10.8); inflammatory markers enumerated (IL-6, hsCRP, TNF-α, VCAM, E-selectin, AGEs).
  9. clinical-trials-active methodology: Added documentation note explaining why a broad “menaquinone” query returns 60 trials (majority NOAC-vs-VKA management) vs ~7 genuinely K2/MK-7 aging/CV/bone trials. Count of 7 retained as defensible strict-filter result (DANCODE, InterVitaminK, CKD K2, osteoporosis/bone trials).
  10. Banner removed; verified: true set.
  11. Unverifiable footnotes flagged: Schurgers 2007, Schurgers 2013, Bladbjerg 2024 = not_oa; Lewy 2023 = pending download; all marked with appropriate archive notes.

PubChem CIDs confirmed: 5284607 (K1, C31H46O2, 450.7 Da) and 5287554 (MK-7, C46H64O2, 649.0 Da) — live REST API. ChEMBL IDs not verified (API 500 error at verification time; not populated).

Supersession check (R25, compound — required): PubMed search (2022–2026) for menaquinone/MK-7/K2 + vascular calcification/arterial stiffness/BMD + meta-analysis/RCT returned 12 hits. New findings:

  • Hasific 2025 (Atherosclerosis, PMID 41100911) — AVADEC sub-study (same Diederichsen cohort): K2+D3 720 µg/day did NOT reduce epicardial adipose tissue or systemic inflammation markers (hs-CRP, IL-6, TNF-α) despite dp-ucMGP reduction. Consistent with existing framing (null on inflammation, null on structural endpoints). Not a supersession.
  • Zhang 2025 (Front Endocrinol, PMID 41268154) — meta-analysis of K2 on bone turnover markers (OC, ucOC, BAP) in postmenopausal osteoporosis (9 RCTs, n=2,570): positive biomarker effects (OC ↑, ucOC ↓, BAP ↑). Bone biomarker evidence is strengthened but hard BMD/fracture endpoints not reported. Not a supersession; consistent with existing cautionary framing for hard bone endpoints.
  • Meer 2023 (Atherosclerosis, PMID 37852868) — post-hoc K1 trial in T2DM (MK-7 360 µg × 6 mo, T50 calcification propensity); not a supersession on vascular calcification. No supersession found. literature-checked-through: 2026-06-02 confirmed.

Downstream propagation needed:

  • vascular-calcification — may cite Diederichsen 2022 with the “+17 AU” sign error; check and correct if so
  • matrix-gla-protein — may carry the same Diederichsen sign error in any summary of the trial; check
  • Any page referencing the Diederichsen 2022 mean difference should use “treatment effect −17 AU (95% CI −86 to +53; p=0.64; non-significant trend favouring MK-7)“

[2026-06-02] verify | processes/splicing-dysregulation.md

Page verified: processes/splicing-dysregulation.md · type: process

Sources checked:

  • Scaffidi & Misteli 2006 (doi:10.1126/science.1127168, PMC1855250) — verified against PMC full text (local PDF unavailable; PMC OA copy used)
  • Holly et al. 2013 (doi:10.1016/j.mad.2013.05.006, PMC5863542) — verified against PMC full text
  • Latorre et al. 2017 (doi:10.1186/s12860-017-0147-7) — verified against full PDF (gold OA; downloaded to archive)
  • Pabis et al. 2024 (doi:10.7554/eLife.87811) — verified against full PDF (gold OA; downloaded to archive)
  • Hayman & Simons 2026 (doi:10.1007/s11357-025-02079-6) — verified against PubMed abstract + Crossref metadata (Springer OA PDF not downloadable)
  • Remaining 9 footnotes (Scaffidi 2005, Latorre 2019, Adusumalli 2019, Yao 2020, Baralle 2023, Okada 2021, Bhadra 2020, Deschenes 2017, Angarola 2021) — not re-read; footnote metadata sourced from Crossref/PubMed and seeder extraction

Corrections made:

  1. “~one-third of splicing factors” → “38% (27/71 in InCHIANTI; 29% in SAFHS replication)” — primary InCHIANTI result is 38%, not one-third; two-cohort design and exact n values added throughout
  2. Holly 2013 footnote: added InCHIANTI n=695 (age 30–104), SAFHS n=1,238 (age 15–94), exact proportions, 9 cross-cohort replicating genes, ATM coefficient and q-value; removed “n not listed in abstract” (incorrect)
  3. Fabricated compound “H3BS-10000” removed entirely. Latorre 2017 tested resveratrol + 5 novel synthetic stilbene analogues (compounds 1–6, unnamed); no compound called “H3BS-10000” exists in this paper. Also removed incorrect claim that “quercetin” and “fisetin” were tested (they were not).
  4. Latorre 2017 splicing factor list corrected: “SRSF1, SRSF2, SRSF3, SRSF6, hnRNPB1, hnRNPA1” → actual 20-factor panel including SRSF7, SRSF18, HNRNPA0, HNRNPA2B1, HNRNPD, HNRNPH3, HNRNPK, HNRNPM, HNRNPUL2, SF3B1, LSM2, LSM14A, others; “hnRNPB1” does not appear in the paper.
  5. Latorre 2017 “dose-dependent” qualifier removed (all experiments used 5 µM; dose-dependence was shown only for the ERK inhibitor trametinib, not the resveralogues themselves in the primary assay).
  6. Latorre 2017 footnote: updated author list (Birar, Sheerin, Jeynes et al.), journal title, compound characterization, quantitative SA-β-Gal and Ki67 results, statistical details; removed incorrect “#gap/no-fulltext-access” (paper is gold OA and was downloaded).
  7. Pabis 2024: added species specificity caveat — mouse liver IR trend non-significant (p=0.14) and readthrough unchanged (p=0.22); human effects do NOT robustly replicate in murine liver. Body text and model-organism table both updated with this correction. Added human n=143, age 0–96.
  8. Scaffidi 2006: added quantitative nuclear abnormality data (Tri-Me-K9H3: 61±23% vs 25±3.8%; phospho-H2AX: 43±6% vs 12±3%; ASO BrdU+30%); removed incorrect “#gap/no-fulltext-access” tag (paper available via PMC1855250).

Unverifiable claims:

  • Scaffidi & Misteli 2005 (Nature Medicine, closed-access): claims about HGPS ASO mechanism verified only from abstract; no-fulltext-access retained
  • Latorre 2019 (FASEB Journal), Adusumalli 2019, Yao 2020, Baralle 2023, Okada 2021, Bhadra 2020, Deschenes 2017, Angarola 2021: not re-read; footnote metadata from Crossref/PubMed

Supersession candidates: None. PubMed search (2022–2026) for meta-analyses or RCTs on splicing factor aging returned 0 results — no superseding evidence found. literature-checked-through: 2026-06-02 confirmed.

Downstream propagation needed:

  • sf3b1 — the Holly 2013 correction (SF3B1 is one of the 9 cross-cohort genes) does not require a change on the sf3b1 page (which focuses on CHIP mutations, not expression aging); no propagation needed
  • The compound “H3BS-10000” fabrication was contained to this page only (no other page cites this phantom compound per grep)
  • The “mouse tissues” over-claim from Pabis 2024 was contained to this page; no other wiki page currently cites Pabis 2024

[2026-06-02] verify | molecules/proteins/vkorc1.md

(Re-homed here 2026-06-02 — the verifier subagent originally misfiled this entry into the inactive log/R9.md round-log; moved to the correct ad-hoc daily log.)

  • verified: molecules/proteins/vkorc1.md (type: protein)
  • primary sources checked:
    • Rost 2004 (10.1038/nature02214) — local PDF; confirmed 163 aa protein length, ER localisation; corrected topology 4→3 transmembrane helices (PSORT II predicted 3, per Li et al. companion paper in same Nature issue; seeder wrote “4 helices”)
    • Nuotio 2021 (10.3389/fneur.2021.696244) — gold OA PDF; OR values confirmed (ICA OR 18.27 CI 2.53–2323; CCA OR 2.64 CI 1.51–4.63); histological effect size added (8.46, CI 3.36–13.56, P=0.0018); warfarin user n clarified (82 of 500); CI note added explaining Firth’s method
    • Holden 2014 (10.1161/ATVBAHA.114.303211) — bronze OA PDF; SNP corrected rs2884737→rs8050894; n clarified (167 total, 86 CAC-progression subset); full ORs/HRs with CIs confirmed
    • D’Andrea 2005 (10.1182/blood-2004-06-2111) — bronze OA PDF; n corrected 297→147; SNP clarified as 1173C>T (intron 1, LD with rs9923231; not the promoter -1639G>A variant directly)
  • unverifiable (closed-access): Di Lullo 2019 (10.1016/j.ijcard.2018.11.119), Wang 2006 (10.1161/CIRCULATIONAHA.105.580167) — no-fulltext-access; abstract-level claims retained
  • canonical IDs (UniProt Q9BQB6, NCBI Gene 79001, HGNC 23663, Ensembl ENSG00000167397) accepted from seeder live-REST confirmation
  • mr-causal-evidence: partial — confirmed; druggability-tier: 3 — confirmed per aging-context convention
  • supersession (R25): PMID 41401404 (2025 meta-analysis, 3 RCTs n=272, DOACs vs warfarin plaque progression) and PMID 41465069 (2025 cross-sectional n=302 CKD, VKORC1 + arterial stiffness/MGP) — both directionally consistent; no supersession
  • corrections total: 5 factual (topology, Nuotio CIs/n, Holden SNP+n, D’Andrea n, D’Andrea SNP identity)
  • downstream propagation: matrix-gla-protein, vitamin-k, vascular-calcification — none cited the corrected values directly; low priority

[2026-06-02] verify | molecules/proteins/ggcx.md

Page verified: molecules/proteins/ggcx.md · type: protein

Sources checked (5 PDFs read, 3 unverifiable):

  • Tie & Stafford 2016 (doi:10.1182/blood-2015-10-677633, Blood 127:1847–1855) — full PDF verified end-to-end
  • De Vilder 2017 (doi:10.3390/ijms18020240, IJMS 18:240) — full PDF verified (pp. 1–20 / systematic review + patient tables)
  • Berkner & Runge 2022 (doi:10.3390/ijms23105759, IJMS 23:5759) — full PDF verified (pp. 1–12)
  • Wu et al. 2025 (doi:10.1038/s41467-025-65488-3, Nat Commun 16:10480) — full PDF verified (pp. 1–8)
  • Kaesler et al. 2025 (doi:10.1111/apha.70040, Acta Physiol 241:e70040) — full PDF verified (pp. 1–6)
  • Stafford 2005 (doi:10.1111/j.1538-7836.2005.01419.x) — not_oa; no-fulltext-access retained
  • Watzka 2014 (doi:10.1016/j.thromres.2014.07.004) — not_oa; no-fulltext-access retained
  • Ghosh 2022 (doi:10.1002/humu.24300) — download failed (0 candidate URLs); no-fulltext-access retained

Canonical IDs confirmed via live APIs:

  • UniProt P38435 = VKGC_HUMAN, gene GGCX (synonym GC), EC 4.1.1.90, 758 aa, ER membrane, 9 TM helices (UniProt REST API 2026-06-02)
  • HGNC 4247 = GGCX, NCBI Gene 2677, Ensembl ENSG00000115486, UniProt P38435 (HGNC REST API 2026-06-02)

Corrections made:

  1. De Vilder 2017 patient count (critical): “36 GGCX-mutation patients” → “47 GGCX-mutation patients (28 individual probands).” Paper is explicit: 47 patients P1–P47 withheld from the analysis; three with ABCC6 co-mutations were excluded from the starting 50, leaving 47. The “36” was a seeder fabrication.
  2. N-glycosylation sites (structural): Asn-459, Asn-550, Asn-570, Asn-605 → Asn-459, Asn-550, Asn-605, Asn-627. De Vilder 2017 (page 3 topology section) explicitly lists “AA 459, 550, 605 and 627.” Asn-570 does not appear; Asn-627 was missing.
  3. TM helix count reconciled: Wiki stated “nine transmembrane helices” without noting the older topology literature predicted 5 TMDs. Wu 2025 cryo-EM confirmed 9 TM helices; De Vilder 2017 (citing Tie 2000 topological study) described 5 TMDs; Tie 2016 Discussion also cited 5. Updated body text and Identity table to cite both models and note that cryo-EM supersedes the biochemical prediction.
  4. Unsourced PDB 9BUM identifier removed: “PDB 9BUM” was in the body but is not confirmed in the Wu 2025 paper (no PDB accession visible in the pages read). Removed to avoid citing an unsourced identifier.
  5. Carbamylation targets broadened (Kaesler 2025): Wiki said “carbamylation of GGCX lysine residues” — paper identifies carbamylation primarily at arginine residues (Arg 436, 672/673 confirmed in multiple CKD models; additionally Arg 9, 351, 436, 673, 687 in rats; Arg 436, 672 in mice), with lysine also affected (Lys 325, 480). Body text and footnote corrected.
  6. Kaesler 2025 last author corrected: “Jankowski V” → “Vera Jankowski” (corresponding authors Rafael Kramann and Vera Jankowski per the paper cover page).
  7. Footnote enrichment: De Vilder 2017, Wu 2025, and Kaesler 2025 footnotes substantially expanded with verified details (cohort composition, mutations identified, cryo-EM resolution, carbamylation sites, compound screening results).

Unverifiable claims:

  • Stafford 2005: vitamin K cycle mechanism framing is consensus biochemistry and matches multiple other verified sources; retained without alteration
  • Watzka 2014: n=9/10 mutations claim in the footnote; retained with no-fulltext-access
  • Ghosh 2022: skin laxity / PXEL-MCFD claims; De Vilder 2017 covers the PXE-like phenotype independently; Ghosh 2022 body claims retained with no-fulltext-access

Verified claims confirmed correct (seeder accurate):

  • D153G mutation: ~60% FIXgla-ProT, ~15% MGP carboxylation at physiological K1 (Tie 2016 Fig 5, Discussion confirmed)
  • Bicarbonate-mediated CO2 capture: confirmed as Wu 2025 abstract + text central claim
  • druggability-tier: 4 — confirmed correct; no clinical drug engages GGCX for aging indication; vitamin K axis drugged upstream via VKOR or via substrate supplementation, not GGCX directly
  • mr-causal-evidence: not-tested — no GGCX MR study identified (consistent with rare-enzyme framing)
  • Hallmark altered-intercellular-communication — defensible via MGP/vascular paracrine axis; reasoning in body is sound

Supersession check (R25, protein): GGCX has no active drug-dev pipeline (druggability-tier 4) and no fast-moving aging-specific literature — literature-checked-through field is optional and left null per CLAUDE.md cadence table. PubMed search 2022–2026 returned only Wu 2025 and Kaesler 2025 — both already cited. No supersession found.

Downstream propagation needed:

  • matrix-gla-protein — the De Vilder 2017 citation on that page should not state “36 patients”; if it repeats that number, correct to 47. Main agent should check.
  • No other wiki page is known to cite the glycosylation residue numbers directly.

[2026-06-02] verify | pathways/vitamin-k-cycle.md → verified:true

Trigger: explicit verifier invocation on freshly-seeded pathway page.

Source verified: Nakagawa 2010 (Nature doi:10.1038/nature09464, PMID 20953171) — local full-text PDF read end-to-end. All other sources no-fulltext-access (Stafford 2005, Tie 2016, Shearer 2014, Chatrou 2012, Danziger 2008).

Corrections made:

  • UBIAD1 tissue-expression bullet corrected: “highest in brain, testis, and arterial wall” → “highest endogenous MK-4 concentrations in brain, testis, and pancreas; Ubiad1 mRNA highest in heart but MK-4 concentrations in heart paradoxically low.” The original claim (“arterial wall”) is not supported by Nakagawa 2010. Nakagawa shows high MK-4 in brain, testis, and by the tissue distribution data also pancreas; heart has highest mRNA but lowest MK-4 protein (Supplementary Fig. 6). “Arterial wall” as a high-UBIAD1-expression tissue is not stated in the paper.
  • Nakagawa 2010 footnote expanded: added K3-d8 and MK-4-d12 as confirmed substrates (the paper explicitly shows UBIAD1 converts MK-4-d12 → MK-4-d7 as well as K3-d8 and PK-d7); added ER localisation confirmation; added 5-week C57BL/6 mouse model detail; added cell-line models (MG-63 + Sf9 baculovirus).
  • ⚠️ auto-extraction banner removed.
  • Frontmatter: verified: true, verified-date: 2026-06-02, verified-by: claude, verified-scope populated, literature-checked-through: 2026-06-02.

Confirmed correct (no change):

  • Reactome R-HSA-6806664 = “Metabolism of vitamin K” pathway — confirmed via live ContentService API.
  • Reactome R-HSA-6806674 = “UBIAD1 prenylates menadione to form MK4” reaction — confirmed via live API.
  • KEGG null / hsa00130 secondary characterisation — upheld (hsa00130 is terpenoid-quinone biosynthesis, contains VKORC1/GGCX/NQO1 nodes but is not a dedicated VK cycle map; this is a defensible mapping).
  • UBIAD1 identification as MK-4 biosynthetic enzyme — confirmed.
  • GGPP as isoprene donor — confirmed (Fig. 4c/d, proportional activity).
  • Warfarin-insensitive conversion — confirmed (Sf9 cell + MG-63 experiments; Fig. 3d/e).
  • ER localisation — confirmed (Fig. 4a; GFP co-localises with ER-tracker Red, not Golgi BODIPY-TR ceramide).
  • druggability-tier 3 rationale — upheld; warfarin blocks the cycle (harmful direction); no aging-beneficial VKORC1 activator exists.
  • altered-intercellular-communication hallmark — consistent with sibling verified pages ggcx, vkorc1, vitamin-k, matrix-gla-protein.
  • arterial-stiffening page EXISTS (confirmed — phenotypes/arterial-stiffening.md, verified:true).
  • Sibling pages consistent: no cross-page contradictions with ggcx, vkorc1, vitamin-k, matrix-gla-protein.

Stale note propagated to ggcx.md: Removed the now-stale “#gap/needs-page — vitamin-k-cycle pathway page is a stub” note from ggcx.md body + pathway membership section (page now exists and is verified).

Supersession check (R25, pathway): CLAUDE.md designates pathway pages as “canonical-data-stable” — supersession check is explicitly skipped per verifier SOP.

Downstream propagation needed:

  • vitamin-k line 147: currently says MK-4 conversion is in “(brain, testis)” — this is an acceptable abbreviation and does not name arterial wall, so no correction needed there.

[2026-06-02] verify | tissues/thyroid.md

Source(s) checked:

  • doi:10.1056/NEJMoa1603825 (TRUST trial, Stott 2017) — full PDF read (local archive, downloaded today)
  • doi:10.1210/jcem.87.2.8182 (NHANES III, Hollowell 2002) — abstract-only (closed-access, not_oa)
  • doi:10.1093/gerona/glw132 (Yeap 2016) — abstract-only (bronze OA; download failed — UWA repository blocks scraping)
  • doi:10.1089/thy.2017.0414 (Moon 2018 meta-analysis) — abstract-only (closed-access, not_oa)
  • doi:10.1089/thy.2005.15.708 (Lin 2005 nodule malignancy) — abstract-only (closed-access, not_oa)

Corrections made:

  1. Body text: nodule malignancy rate “approximately 4–5% of surgically resected nodules across all ages” → “approximately 23.6% of surgically resected nodules overall (858/3,629)” — matches Lin 2005. Added note distinguishing 3.9% of all evaluated subjects vs 23.6% of resected. The 4–5% figure was unsourced and wrong for this denominator; unsourced tag added for unselected-population malignancy rate.
  2. TRUST trial body description expanded: added TSH inclusion range (4.60–19.99 mIU/L), SD (±2.01), group allocation (369/368), starting dose, TSH target (0.40–4.59 mIU/L), exact p-values (P=0.99 and P=0.77), 95% CIs for both primary outcomes, instrument name (ThyPRO), MCID (9 points), and secondary outcome confirmation.
  3. TRUST footnote expanded: added mean age (74.4 yr), sex (53.7% women), median follow-up (17.3 months placebo), dose details, precise SD on TSH outcomes, adverse events of special interest (confirmed null).
  4. Lin 2005 footnote clarified: surgical resected denominator (3,629 of 21,748) made explicit; 37.2% elderly figure correctly framed as resected-nodule rate.
  5. Gaps section: “only one year” softened to “primary endpoint at one year (median follow-up 17.3 months; maximum 3 years)” to match actual trial design.

No corrections needed:

  • NHANES III prevalence figures (4.3% SCH, 0.3% overt HypoT, 11.3% anti-TPO) — confirmed.
  • Yeap 2016 TSH range (0.64–5.9 mIU/L), ~8% reclassification, FT4-not-TSH mortality association — confirmed.
  • Moon 2018 RR 1.33 overall, null in ≥65 years, all-cause mortality RR 1.20 — confirmed.

Supersession check (R25 — type: tissue): CLAUDE.md designates tissue pages as “canonical-data-stable” — supersession check is explicitly skipped. However, as a tissue page with active clinical-trial evidence, a targeted check was run: post-2021 RCT/meta-analysis search found TRUST ancillary studies (bone geometry 2022, antibody subgroup 2022, muscle function 2023) — all null for levothyroxine benefit, consistent with wiki framing. No supersession of the primary TRUST or Moon 2018 findings found.

Downstream propagation needed: None identified — thyroid.md is a newly seeded page with no downstream pages yet citing its specific quantitative claims.

Final state: verified: true (partial scope — TRUST full PDF; 4 other sources abstract-only per closed-access; secondary review sources not re-verified)

[2026-06-02] verify | tissues/parathyroid.md → verified:true (partial scope)

Page: tissues/parathyroid.md

Sources checked:

  • Minisola 2022 (doi:10.1002/jbmr.4665) — full PDF downloaded and read (OA via PMC). verified
  • Hannan 2018 (doi:10.1038/s41574-018-0115-0) — full PDF downloaded and read (OA via Europe PMC, PMC6535143). verified
  • Jilka 2007 (doi:10.1016/j.bone.2007.03.017) — read via PMC submitted-version (PMC1995599; not OA but accessible). verified
  • Farrell 2018 (doi:10.1111/cen.13486) — PubMed abstract only (closed-access, not_oa). abstract-verified; no-fulltext-access retained
  • Müller-Höcker 2014 (doi:10.1016/j.ajpath.2014.07.015) — closed-access (not_oa); abstract-only. #gap/no-fulltext-access retained
  • Mao 2024 (doi:10.1016/j.kint.2023.11.027) — closed-access (not_oa); abstract-only. #gap/no-fulltext-access retained

Corrections made (6):

  1. Farrell 2018 study design (WRONG → CORRECTED): Wiki said “data mining across 17 reference-interval studies (n=several thousand).” Actual paper is a single-institution retrospective study of 33,652 PTH test results from one private pathology laboratory. The “17 studies” characterization is incorrect — this paper does not meta-analyze 17 other studies.

  2. Farrell 2018 quantitative finding (MISSING → ADDED): Wiki omitted the paper’s key quantitative finding. Added: 63% increase in upper and lower PTH reference limits between youngest (18–29 years) and oldest (≥80 years) age partitions.

  3. Jilka 2007 — continuous PTH apoptosis mechanism (WRONG → CORRECTED): Wiki said continuous PTH promotes osteoblast apoptosis “via cAMP-PKA pathway.” Per Jilka, cAMP-PKA is the ANTI-apoptotic pathway used by intermittent PTH. For continuous PTH, the mechanism is sustained suppression of Runx2 below the osteoblast survival-signaling threshold. The table row was corrected.

  4. Jilka 2007 — intermittent PTH anti-apoptotic mechanism (WRONG → CORRECTED): Wiki said “inhibited via PTHrP/PP2A; anti-apoptotic Bcl-2 upregulation.” Per Jilka, the actual pathway is cAMP→PKA→Bad phosphorylation + Bcl-2 transcriptional upregulation + transient Runx2 elevation. PTHrP/PP2A is not in Jilka’s mechanism description for this pathway. Table row corrected.

  5. Minisola 2022 prevalence figures (IMPRECISE → PRECISE): Wiki said “~0.1–0.3% (approximately 1 in 400–1000 adults).” Paper gives specific US figures: 233/100,000 women (~0.23%) and 85/100,000 men (~0.085%); Scandinavian postmenopausal women 2–5%. Updated to reflect the paper’s actual reported numbers.

  6. 85–90% single-adenoma claim (UNATTRIBUTED → TAGGED): The claim that PHP is due to a single adenoma in ~85–90% of cases is not from Minisola 2022 (which says “clonally dysregulated overgrowth of one or more parathyroid glands” without a proportion). Tagged #gap/unsourced.

Non-corrections confirmed:

  • Year 2022 for Minisola is correct (paper masthead/copyright = November 2022; Crossref published-print 2020 is a metadata error).
  • CASR G-protein coupling updated from “(Gq/Gi11)” to full Gq/11, Gi/o, G12/13 per Hannan (completeness, not a factual error).
  • Klotho–CASR interaction in parathyroid confirmed by Hannan p.5.
  • FGF23–Klotho–PTH axis description consistent with Agoro 2023 (abstract level, closed-access).
  • Hallmark connections (altered-intercellular-communication, deregulated-nutrient-sensing, mitochondrial-dysfunction) consistent with available evidence.
  • Jilka vol/issue (40(6):1434–1446) confirmed.

Supersession check (R25 — type: tissue): Type: tissue is “canonical-data-stable” per CLAUDE.md R25 skip list. Targeted post-2021 search run regardless given the active clinical evidence: (a) PTH reference interval literature 2022–2025 — three large studies (n=22,662; n=2,794; n=1.5M) all confirm age-stratified PTH increase, consistent with Farrell 2018 framing; one 2025 paper (Clinica Chimica Acta, n=22,662) notes 61% reclassification of >50-year-old “hyperparathyroid” patients using age-adjusted intervals — this strengthens the wiki’s argument for age-stratified reference ranges and does not contradict it. (b) PHP epidemiology 2022–2026 — no global meta-analysis supersedes Minisola 2022. African meta-analysis (PMID 41850293, n=2,807) is regional only. No supersession of load-bearing claims found.

Downstream propagation needed: None identified — parathyroid.md is newly seeded; the corrections are contained within this page. The 85-90% adenoma figure may appear on a future PHP phenotype page if seeded; flag at that time.

[2026-06-02] ingest | multivitamins intervention page (ad-hoc)

Trigger: ad-hoc user question on whether multivitamin supplementation is worthwhile, anchored on Loftfield 2024 (JAMA Netw Open). No dedicated page existed (multivitamins were only mentioned incidentally on flavan-3-ols/glycolic-acid). Seeded directly (research already gathered); verified: false + banner, queued for verifier.

Page: interventions/dietary/multivitamins.md · type: intervention · mode: dietary · human-evidence-level: limited-negative (COSMOS cancer/CVD null + Loftfield mortality null + PHS-II modest-only cancer) with the COSMOS cognition exception foregrounded.

Evidence base (all DOIs Crossref-confirmed, not memory-asserted):

  • Loftfield 2024 — 10.1001/jamanetworkopen.2024.18729 (n=390,124 pooled cohort; all-cause mortality HR 1.04, null/slight-positive attributed by authors to sick-user confounding).
  • COSMOS / Sesso 2022 — 10.1093/ajcn/nqac056 (RCT n=21,442; total cancer + CVD null).
  • COSMOS-Web / Yeung 2023 — 10.1016/j.ajcnut.2023.05.011; COSMOS-Mind / Baker 2022 — 10.1002/alz.12767; meta Vyas 2024 PMID 38244989 (cognition benefit in ≥60).
  • PHS-II / Gaziano 2012 — 10.1001/jama.2012.14641 (total cancer HR 0.92, modest).
  • USPSTF 2022 — 10.1001/jama.2022.8970 (insufficient evidence; against beta-carotene + vitamin E).
  • HOPE-2 / Lonn 2006 — 10.1056/NEJMoa060900 (B-vitamin homocysteine-lowering null on cardiac events).

Framing the page foregrounds: multivitamin = deficiency insurance, not a geroprotector; cohort mortality nulls are confounded both directions (healthy-user vs sick-user); formulation matters (iron / beta-carotene / high-dose vitamin E are the harm subset a clean product omits); benefit concentrates where a deficiency gap exists.

Implicit stubs / link targets: folate, mthfr, by-intervention-type (some may not yet resolve — flagged for lint).

Leak-gate: run clean over new + modified tracked files.

[2026-06-02] verify | interventions/dietary/multivitamins.md

Page verified: interventions/dietary/multivitamins.md · type: intervention

Sources checked:

  • Loftfield 2024 (doi:10.1001/jamanetworkopen.2024.18729) — full PDF read end-to-end (9 pp. + supplement)
  • Sesso 2022 COSMOS (doi:10.1093/ajcn/nqac056) — full PDF read end-to-end (10 pp.)
  • Yeung 2023 COSMOS-Web (doi:10.1016/j.ajcnut.2023.05.011) — full PDF read end-to-end (10 pp.)
  • Baker 2022 COSMOS-Mind (doi:10.1002/alz.12767) — PMC abstract + PMC body text (PMID 36102337); PDF download failed
  • Vyas 2024 (doi:10.1016/j.ajcnut.2023.12.011, PMID 38244989) — PubMed structured abstract
  • Gaziano 2012 PHS-II (doi:10.1001/jama.2012.14641, PMID 23162860) — PubMed structured abstract; not_oa for full PDF
  • USPSTF 2022 (doi:10.1001/jama.2022.8970, PMID 35727271) — PubMed structured abstract; download failed
  • HOPE-2 Lonn 2006 (doi:10.1056/NEJMoa060900, PMID 16531613) — full PDF read end-to-end

Corrections made:

  1. Vyas 2024 DOI (critical): footnote had wrong DOI 10.1016/j.ajcnut.2023.05.011 (= Yeung 2023 COSMOS-Web) → corrected to 10.1016/j.ajcnut.2023.12.011 (confirmed Crossref). PMID 38244989 was always correct.
  2. COSMOS table age criterion: “ages ≥60” → “women ≥65/men ≥60” (confirmed from full PDF Table 1).
  3. COSMOS table result row enriched: “no significant reduction” → added actual HRs: cancer 0.97 (0.86–1.09), CVD 0.98 (0.86–1.12).
  4. COSMOS footnote enriched: added cancer HR, CVD HR, mean age 72.1, confirmed both p-values (cancer p=0.57, CVD HR in Figure 4).
  5. COSMOS-Web n corrected: “n≈3,500” → “n=3,562” (ITT sample from CONSORT diagram; randomized n was 3,960, ITT n=3,562).
  6. COSMOS-Web effect size made exact: “~3 years” → “3.1 years” (paper states “equivalent to ~3.1 y of age-related memory change”; p=0.025 at 1 yr, p=0.011 over 3 yr added).
  7. Baker 2022 footnote year corrected: “2022” → “2023” (published online Sept 2022, journal issue Apr 2023; DOI-confirmed journal year 2023). Age criterion corrected from “≥65 yr” → “women ≥65, men ≥60” (parent COSMOS trial eligibility); mean age 73, 60% women added.
  8. Baker 2022 global cognition z-score added: mean z=0.07 (95% CI 0.02–0.12) p=0.007 — primary quantitative result now explicit in footnote.
  9. HOPE-2 footnote enriched: added n=5,522, primary composite RR 0.95 (95% CI 0.84–1.07) p=0.41, stroke RR 0.75 (95% CI 0.59–0.97) p=0.03, and PMID 16531613.
  10. USPSTF footnote enriched: added PMID 35727271, I/D recommendation grading, and “moderate certainty” language.
  11. Loftfield footnote enriched: added per-cohort n breakdown, per-cohort maximum follow-up years, “decline to claim harm” authors’ interpretation.
  12. Removed ⚠️ auto-extraction banner; flipped verified: true.

Unverifiable claims:

  • SELECT trial (high-dose vitamin E → prostate cancer): characterized via USPSTF synthesis, not independently verified against the SELECT primary paper. USPSTF 2022 is authoritative for this characterization.
  • ATBC/CARET beta-carotene signals: characterized via USPSTF synthesis, not independently re-read.
  • SEARCH and VITATOPS concordance with HOPE-2: cited in HOPE-2 paper’s discussion section; those two PDFs not independently read.

Supersession check (R25 — required for intervention pages): PubMed search 2024–2026, term “multivitamin supplement aging meta-analysis.” Found: Wang et al. 2026 (Ageing Res Rev 114:102965, PMID 41308839) — rapid review of 19 meta-analyses, 5,535,426 participants. Key findings: MVM improved global cognition, episodic memory, and immediate recall in older adults (consistent with COSMOS cognition framing); no benefit for all-cause mortality (consistent); higher risk of AMD progression (a new signal not currently on the page — warrants a caveat note if the page is ever expanded). No supersession of the page’s core claims detected. literature-checked-through: 2026-06-02 confirmed.

AMD signal note: Wang 2026 rapid review reports a higher risk of AMD progression with MVM in some meta-analyses. This is not addressed on the current page. Not a supersession of the existing claims (the page doesn’t claim MVM is safe for AMD), but worth flagging as a potential future addition to the “formulation matters” or “harm subset” section.

Post-verify addition (main agent, 2026-06-02): integrated the Wang 2026 finding as a new ”## Recent synthesis (2026)” section + [^wang2026] footnote — corroborates the cognition + mortality framing and records the AMD-progression signal with a contradictory-evidence flag (framed as likely confounding-by-indication / formulation heterogeneity, not a randomized-trial harm; AMD signal not independently verified against the included primaries). Downstream DOI check run: the corrected Vyas DOI does not appear mis-attributed on any other page (grep clean).

Downstream propagation needed:

  • None critical. The Vyas 2024 DOI correction is isolated to this page (no other wiki page cites the Vyas footnote directly by DOI). The COSMOS-Mind age criterion correction (≥65 → women ≥65/men ≥60) is not cited elsewhere by that number. Main agent should confirm no downstream page inherited the wrong Vyas DOI.

[2026-06-02] schema | R51 — experiments/ relocated to the private protocols/ tree (user-requested)

Trigger: user judged the type: experiment pages to have become user-authored working drafts (speculative proposals) more akin to the private brainstorming tier than to public research, and asked to move experiments/ under protocols/.

Move: the 6 type: experiment pages + README moved out of the public top-level experiments/ into the private nested repo at protocols/experiments/ (7 files: deletions staged in the public repo, additions staged in the private repo). .gitignore already excludes all of protocols/, so no ignore change was needed.

type: experiment schema unchanged — lifecycle + resolves-edges:/resolves-nodes:/adds-nodes: retained; only the home dir + privacy status changed. Still documented in CLAUDE.md § Page types alongside the other private-tree types. Published results still graduate to a public type: study page via published-as:.

Link discipline — now one-way. The experiment↔causal-graph integration was bidirectional; it’s now outbound-only (experiment pages → public research; never the reverse). Four public inbound references were de-linked, keeping the prose:

  • frameworks/causal-graph-data.md — worked-example column, scale table, edge-matrix cell, and “Resolving experiments” bullets: experiments now named in plain text as proposed / not-yet-run.
  • index.md — the ”## Experiments” Dataview section removed entirely.
  • phenotypes/chronic-venous-disease.md — two refs reworded to plain literature-gap statements (no “private tracker” mention, no dangling “proposed” antecedent — they read as field-level gaps).

Leak-gate extended: the bare experiments/-prefixed wikilink alias still resolves into the private tree (Obsidian suffix match), so experiments was added to the forbidden-link alternation (CLAUDE.local.md) and to the public-repo invariant text (CLAUDE.md, both spots). Bare prose mentions of “experiments” and the directory name remain allowed (architecture documentation). NB: documentation must describe the forbidden link in prose with bare backticked paths, never the literal bracket form, or it self-matches the gate (caught + fixed this pass).

Docs updated: CLAUDE.md (directory map, methods descriptions ×2, type: experiment block, protocols structure block, privacy invariant ×2), schema-history.md R51, and the relocated README (private banner + one-way note + Dataview FROM paths repointed to protocols/experiments).

Leak-gate: re-run clean over all tracked public files after the de-linking.

Graph View

Backlinks