🧬 Aging Wiki

2026-06-04

21 min read

[2026-06-04] log

[2026-06-04] ingest | androgen / testosterone-longevity mechanism — gap-fill (user-prompted)

Triggered by a user question on TRT from a longevity perspective, which escalated into the mechanistic question of whether androgens up-regulate pro-aging signaling (mTOR/IGF-1) and whether early/cumulative androgen activity causally drives late-life testosterone decline (a damage-based, not depletion-based, directional model). Existing TRT/andropause/Leydig coverage (seeded + verified 2026-06-03) answered the clinical question; the comparative/evolutionary and limiting-factor angles were gaps.

Seeded (wiki-seeder, parallel batch; all verified: false, banners on)

  • added: hypotheses/androgen-deprivation-longevity-hypothesis.md — Mode A evidence-aggregating. Claim: reduced lifetime androgen exposure extends male lifespan. Aggregates Hamilton & Mestler 1969 + Min 2012 (eunuch cohorts), Garratt 2026 Nature (cross-vertebrate sterilization meta-analysis; pre-pubertal effect stronger; female survival slightly decreased — implicates testosterone specifically), Garratt 2021 (castration extends maximal not median lifespan, mice), Sugrue 2021 eLife (castration decelerates the epigenetic clock, sheep/mice). Balanced by Le Bourg 2015 critique of Min 2012. R25 recency run; literature-checked-through: 2026-06-04.
  • added: studies/min-2012-korean-eunuchs-lifespan.md, studies/hamilton-mestler-1969-eunuchs-mortality.md — both gap/no-fulltext-access (closed access; effect sizes from secondary accounts pending PDF verify).
  • propagation: links added from sex-differences-in-aging and female-longevity-advantage.
  • Correction logged: the popularly-cited “castrati singers = null counter-evidence” claim has no confirmed primary study behind it (seeder literature search returned none beyond Nieschlag 2003, a voice-category proxy study). Earlier conversational answer that cited it as an established null was wrong; the real counterweight on the hypothesis page is Le Bourg 2015.

Limiting-factor mechanism → cell-types/leydig-cells.md (verified page; new section verified)

Added § “Steroidogenesis as a self-damaging process — and its reversibility” answering “what is the limiting factor in late-life testosterone decline.” Synthesis: the per-cell steroidogenic decline is best modelled as reversible, activity-dependent oxidative wear of the steroidogenic machinery (steroidogenic P450s leak electrons → ROS), NOT depletion of a finite reserve.

  • [^chenzirkin1999] — Chen, Hardy & Zirkin 1999 PNAS (doi:10.1073/pnas.96.26.14877; PMC24741) — full text verified. 8-mo testosterone-implant suppression of LH/steroidogenesis in Brown Norway rats (13→21 mo) then released → testosterone + all steroidogenic enzymes/proteins restored to 13-mo-control levels, while un-suppressed 23-mo controls declined. Reversible recovery, not finite-reserve preservation. Authors floated steroidogenesis-ROS damage but noted “no direct evidence” (1999).
  • [^beattie2013] — Beattie et al. 2013 Biol Reprod (doi:10.1095/biolreprod.112.107052) — abstract-verified only (closed access, OUP, no PMC OA; full PDF pending gap/no-fulltext-access). LH stimulation → transient ROS in primary Leydig cells; aged cells peak later / clear slower; LH → more DNA damage in aged cells; vitamin E prevents it. Supplies the ROS evidence Chen & Zirkin 1999 lacked. Specific clearance-time figures (full-text only) deliberately omitted.
  • frontmatter: verified: true retained, verified-date → 2026-06-04, scope addendum describing the new section’s verification level.
  • Caveats embedded: rat-only reversibility experiment (no human equivalent); in aging men most low-T is secondary/comorbidity-driven (EMAS), so this describes the intrinsic-testicular component of andropause, not the dominant population cause.

Gaps surfaced / pending

  • Study pages not yet seeded as standalone files: [[studies/chen-zirkin-1999-suppression-prevents-leydig-aging]], [[studies/beattie-2013-lh-ros-dna-damage-leydig]], [[studies/garratt-2021-castration-maximal-lifespan-mice]], [[studies/garratt-2026-sterilization-vertebrate-lifespan]].
  • PENDING (next pass): testosterone Mendelian-randomization gap-fill (Ruth 2020 / Mohammadi-Shemirani 2022) — second wiki-seeder running; the directionality-paradox synthesis (T engages mTOR yet aging is low-T/low-mTOR; muscle-local vs systemic mTOR) is to be written onto molecules/compounds/testosterone.md in one combined verified pass once MR study pages land, so the verified compound page is edited once.
  • Verification priorities for wiki-verifier: Hamilton & Mestler 1969, Min 2012, Garratt 2026, Sugrue 2021 (per seeder report); Beattie 2013 full PDF if it becomes fetchable.

[2026-06-04] ingest+update | testosterone Mendelian-randomization + directionality synthesis

Second strand of the androgen/testosterone-longevity work (first MR seeder hung at ~4 min with no output → TaskStop’d, left no files → re-dispatched fresh seeder with a timebox guardrail, which completed cleanly in ~4 min).

Seeded (wiki-seeder; verified: false, banners on)

  • added: studies/ruth-2020-testosterone-mr-disease.md — Ruth 2020 Nat Med (doi:10.1038/s41591-020-0751-5; PMC7025895).
  • added: studies/mohammadi-shemirani-2020-testosterone-mr-elife.md — Mohammadi-Shemirani 2020 eLife (doi:10.7554/eLife.58914; PMC7591257).
  • Seeder corrected a training-memory error in my brief: Mohammadi-Shemirani is eLife 2020 (461 male outcomes, null CV), NOT a JAMA VTE/heart-failure paper — that framing was fabricated noise; discarded, not seeded. (Reinforces seeder-brief-DOI/identity-unreliable memory.)

molecules/compounds/testosterone.md (verified page) — two sections added, both verified

  • § Mendelian randomization — causal evidence beyond trials. Ruth 2020 + Mohammadi-Shemirani 2020, both full-OA-text verified 2026-06-04 (all ORs/CIs cross-checked against PMC):
    • Ruth: T2D OR 0.86 (0.76–0.98) men / 1.37 (1.22–1.53) women; PCOS 1.51; prostate cancer 1.23 (1.13–1.33) men.
    • Mohammadi-Shemirani (93 SHBG-filtered instruments): prostate cancer OR 1.51 (1.21–1.88); hypertension 1.17; hematocrit +1.37%; BMD +0.40 SD; body fat −1.88%; MI borderline 1.23 (1.00–1.53, p=0.05); null T2D/CV/cognitive.
    • T2D-discordance reconciliation (analytic, not raw contradiction): Mohammadi filtered SHBG-acting variants; SHBG is itself protective for T2D, so testosterone’s apparent male metabolic benefit is likely SHBG-mediated. Framed as such; gap/contradictory-evidence retained for the formal-reconciliation gap.
    • MR nuances the saturation model: lifetime genetically-higher exposure causally raises prostate cancer (≠ supra-saturation TRT dosing argument; not in conflict, but lifetime exposure is not prostate-neutral). Longevity/frailty/mortality untested by MR.
  • § Testosterone, mTOR, and the rate of aging — a resolved tension (the user’s directionality question; framed as resolved, not a paradox, per user). Three-part resolution: (1) T’s mTORC1 activation is muscle-local, not the systemic hyperfunction rapamycin targets; (2) aging’s low-T state is downstream damage — Leydig reversible-wear (leydig-cells) + comorbidity-driven functional suppression (Huhtaniemi 2011, home andropause); (3) net lifetime-androgen-cost is partly supported (castration/MR) but the early-T→early-Leydig-failure directional claim is plausible-but-untested in humans. Cross-links mtor, sarcopenia, androgen-deprivation-longevity-hypothesis, antagonistic-pleiotropy.
  • frontmatter: verified: true retained, date → 2026-06-04, scope addendum added. No mr-causal-evidence frontmatter field added (not in compound schema — kept in body; schema discipline).

Propagation

  • hypotheses/androgen-deprivation-longevity-hypothesis.md — MR “what would update” item updated: disease-outcome MR now exists (links both study pages), longevity-endpoint MR still gap/not-yet-done; pointer to testosterone.md synthesis sections. (New study pages now inbound-linked from testosterone.md footnotes + hypothesis page — not orphans.)

Pending

  • wiki-verifier standalone pass on the 2 MR study pages (priorities per seeder: Ruth 2020 cancer ORs + instrument F-stats; Mohammadi-Shemirani full beta/OR table + FDR threshold).
  • Standalone study pages still unseeded for chen-zirkin-1999, beattie-2013, garratt-2021, garratt-2026 (footnote-embedded for now).

[2026-06-04] verify | androgen/testosterone-longevity study pages (4× wiki-verifier, parallel)

  • verify — studies/ruth-2020-testosterone-mr-disease.md → verified: true. Full OA PDF (PMC7025895) read end-to-end. Prostate cancer OR 1.23 (1.13–1.33) confirmed. Correction propagated to testosterone.md: breast/endometrial cancer ORs are NOT in the narrative (Extended Data forest plots only) — “exact ORs in full text” phrasing was wrong on both testosterone.md body + footnote; fixed. Added ovarian-cancer-protective finding. Supersession-surfaced (R25): PMID 35218343 (2022 IJE — higher T → lower T2D in women, contra Ruth’s 1.37), PMID 36653534 (2023 phenome-wide MR — minimal causal impact). Added a provisional-female-T2D caveat to testosterone.md (#gap/contradictory-evidence; PMIDs recency-surfaced, unverified, not seeded).
  • verify — studies/mohammadi-shemirani-2020-testosterone-mr-elife.md → verified: true. Full OA PDF (PMC7591257) read. All testosterone.md-cited ORs confirmed correct (no figure errors). Study page corrected: multiple-testing is Bonferroni not FDR; instrument = 93 SHBG-filtered SNPs of calculated free testosterone; units per 0.1 nmol/L CFT. Propagated to testosterone.md (omissions, not errors): added HDL −0.074 SD and spinal stenosis OR 2.03; noted Bonferroni in footnote. Supersession: PMID 41131989 (2026 JCEM, higher T → higher CAD in men, BP-mediated) — corroborating, not superseding.
  • verify — studies/min-2012-korean-eunuchs-lifespan.md → verified: true (Cell Press served OA PDF via fallback; 2-page letter read in full). Major study-page correction: comparator group was wrongly “~329–749 aristocrats” → actual three clans Mok 1,126 / Shin 1,414 / Seo 49. Precise figures recovered: mean eunuch lifespan 70.0 ± 1.76 yr; advantage 14.4–19.1 yr; centenarians aged 100/101/109; kings 47.0, royal males 45.0; dynasty 1392–1910; 81 of 385 registry eunuchs with lifespans. Propagated to hypothesis page (precise figures replace “~14–19 yr”). Le Bourg 2015 full text closed-access (Karger) — critique attribution accurate but arguments not full-text-verified.
  • verify — studies/hamilton-mestler-1969-eunuchs-mortality.md → verified: false retained (permanent blocker: closed access, no OA route, no PMC). Numbers cross-checked via Min 2012, which quotes the primary: mean age at death 69.3 (castrated) vs 55.7 (intact men) = 13.6 yr (was “~13–14 yr”). Propagated to hypothesis page. Verifier also fixed a wrong Le Bourg DOI on this page (10.1159/000381651 → 10.1159/000435854). Cohort n, women-comparison, age-at-castration dose-response remain unconfirmed.
  • Hypothesis page androgen-deprivation-longevity-hypothesis.md stays verified: false: human eunuch arm now verifier-checked, but Garratt 2021/2026 + Sugrue 2021 are footnote-embedded and unverified (Garratt 2026 closed-access). Banner updated to reflect partial-verification state.
  • Follow-ups surfaced: consider seeding PMID 35218343 / 36653534 / 41131989 (testosterone MR recency); verify Sugrue 2021 (gold OA) + attempt Garratt 2026 to enable flipping the hypothesis page.

[2026-06-04] verify | androgen-deprivation hypothesis — OA animal/epigenetic sources + page flip

Re-verification of the hypothesis page’s remaining OA sources (3× verifier agents, read-only; corrections applied centrally to avoid same-file write races).

  • Sugrue 2021 (eLife) — full text verified. All 3 claims confirmed. Corrections to hypothesis page: clock-deceleration result is sheep-specific (mouse/bat data show conservation of androgen-sensitive CpGs only, NOT a clock-rate result — page previously implied sheep+mice); added magnitude 3.1 mo deceleration, p=0.018, n=61/63, ≥18 mo; clock is a purpose-built 185-CpG sheep clock, not Horvath 353; strengthened author caveat (“may not promote aging per se”). Supersession: Sugrue 2025 PNAS (androgen clock) extends, not supersedes.
  • Garratt 2021 (GeroScience) — full text verified. All 4 claims confirmed. Corrections: “90th percentile survival” was ambiguous/invertible → restated as 90%-mortality timepoint (top-decile survivorship); added effect sizes castration +87 d (Fisher P=0.026), female access +84 d (P=0.039), non-additive; median null (Cox P=0.632/0.883); clarified the female effect is on overall/median survival (log-rank P=0.034), opposite the male maximal-lifespan pattern; design 2×2 factorial C57BL/6Jausb n=32/27/36/27.
  • Garratt 2026 (Nature) — abstract-verified only; full text permanently closed. Agent exhausted archive/PMC/bioRxiv/medRxiv/Research Square/institutional/SharedIt — no OA copy exists. All 4 claims abstract-confirmed (incl. the load-bearing pre-pubertal gradient + women-slightly-decreased). Corrections: added that male benefit is restricted to surgical castration, not hormonal contraception (females benefit from both); study design is hybrid zoo-cohort + meta-analysis, not pure meta-analysis; effect sizes remain gap/no-fulltext-access.
  • Hypothesis page flipped to verified: true with a documented verified-scope (full-text: Min/Garratt2021/Sugrue2021; abstract-only permanent-closed: Garratt 2026; cross-checked: Hamilton-Mestler). Banner switched from ⚠️ to a verification note.
  • Residual gaps: key-evidence-against frontmatter still [] (Le Bourg 2015 + Nieschlag 2003 closed-access, documented at body level, lack study pages) — Mode-A schema gap noted in scope. Implicit-stub study pages studies/garratt-2021-… and studies/garratt-2026-… still unseeded (now full-text/abstract verified respectively, ready to seed as standalone pages if desired).

[2026-06-04] ingest+update | secular testosterone-decline debate + measurement-artifact (video-prompted)

User asked whether a science-summary YouTube video’s transcript + description reference list could be scraped, then to ingest the relevant data. Two outputs: a new SOP for the scrape procedure, and a new contested-topic section on testosterone.md. The video itself is not cited anywhere — used strictly as a discovery lead list; every claim traced to and cited from the primary source.

New SOP

  • added: sops/scraping-youtube-references.md — raw-HTML shortDescription extraction for the reference list; yt-dlp for the transcript (direct api/timedtext is token/IP-gated → empty); discipline reminder that the video is a tertiary/promotional lead, never a citation; resolve→map→ingest→cross-check-framing workflow. Indexed in CLAUDE.md SOP list.

molecules/compounds/testosterone.md (verified page) — one section added

  • § The secular (population-level) decline — real trend, or a measurement artifact? Distinguishes the (undisputed) individual age-related decline from the (contested) cross-generational decline. Key correction vs the video’s framing: the cohorts disagree on whether obesity explains the trend — Danish decline vanished after BMI adjustment (Andersson 2007), Finnish + MMAS + young-US persisted; the video lumped them all as obesity-independent.
  • Linchpin source full-text verified: Arun 2025 Clin Chem (doi:10.1093/clinchem/hvaf025) — user supplied the PDF; read end-to-end. NHANES median TT 495/500/418/409/433 ng/dL across 5 cycles (−16.4% across the 2003-04→2011-12 immunoassay→LC-MS/MS switch); <300 ng/dL prevalence 11/12% → 22/21/18%; re-scoring later cycles at the Endocrine Society 264 ng/dL LC-MS/MS cutoff returns 13/14/11% → the “low-T epidemic” is largely an assay/threshold artifact. It’s a Letter to the Editor and NHANES-specific — does not dissolve single-assay cohorts (MMAS immunoassay-throughout; Israeli Maccabi clinically-ordered tests w/ referral-selection caveat).
  • added: studies/arun-2025-testosterone-threshold.md (verified: true, full-text). Inbound-linked from testosterone.md body + footnote (not an orphan).
  • Supporting cohorts (Travison 2007, Andersson 2007, Perheentupa 2013, Lokeshwar 2021, Chodick 2020) + obesity/weight-loss reviews (Ng Tang Fui 2014, Okobi 2024) cited abstract-level (structured abstracts read; full PDFs not retrieved). All DOIs/PMIDs/authors resolved via PubMed/Crossref (not memory); obesity-review PMID 24407187 memory-asserted then DOI-verified.
  • frontmatter: verified: true retained; literature-checked-through 2026-06-03 → 2026-06-04; verified-scope addendum carving out the new section’s per-source verification level.

Gaps surfaced / pending

  • Abstract-level cohort papers above could be promoted to standalone studies/ pages + full-PDF-verified if the secular-decline topic is revisited; the obesity→testosterone and exercise/sleep/TMG intervention threads from the video’s description (PMC3955331, PMC7739287, PMC11519272, PMC7934563, PMC9116406) are resolved but not yet ingested (lower-priority, partly promotional).

[2026-06-04] update | NewLimit commercial-landscape refresh (user-prompted)

Triggered by a user-shared BioSpace article on NewLimit’s $435M Series C (“age reversal in human liver cells”). Coverage check: the underlying science (partial epigenetic reprogramming) is already deep and verified — process page, info-theory-of-aging hypothesis, OSK/OSKM factor pages (Oct4/Sox2/Klf4/c-Myc/Nanog), TET1/2/3 cofactors, landmark studies (Ocampo 2016, Lu 2020, Yang 2023 ICE), and the in-vivo-partial-reprogramming-therapy intervention page with an industrial-pipeline table. No new science to seed; the article carries no genes/mechanisms.

Two stale facts in the pipeline table corrected (table is press-release-sourced gap/unsourced, refreshed each pass):

  • updated: interventions/stem-cell-therapy/in-vivo-partial-reprogramming-therapy.md — NewLimit row. Fixed founders (was “Brian Armstrong, Jake Hwang” → correct: Brian Armstrong, Blake Byers, Jacob Kimmel/CEO); focus T-cell → liver lead program; added 3.1B valuation / first-in-human-planned. New [^newlimit2026] footnote flagged company-reported / non-peer-reviewed, no registered NCT located.
  • updated: processes/partial-reprogramming.md — NewLimit industry-landscape bullet, same corrections; cross-linked to pipeline table.

Gaps surfaced

  • Lin28 (LIN28A/B) still has no protein page (referenced in Lapasset 2011 OSKMNL six-factor; flagged on nanog.md).
  • tissues/liver.md remains a stub; no hepatocyte-reprogramming coverage despite NewLimit’s liver lead — candidate seed if revisited.
  • No NewLimit peer-reviewed liver result exists yet; revisit [^newlimit2026] when a publication or ClinicalTrials.gov registration appears.
  • No standalone company pages (NewLimit/Altos/Retro/Life Bio/Turn) — pipeline data lives only in intervention tables, by design.

[2026-06-04] ingest | liver tissue page + Lin28 protein page (user-prompted, parallel seed)

Follow-on from the NewLimit refresh: user asked to seed tissues/liver.md out of stub status (liver aging as an organ/tissue, with partial reprogramming as one thread among several) and to fill in the missing Lin28 protein page. Two wiki-seeder subagents dispatched in parallel.

added

  • tissues/liver.md — full type: tissue page (~300 lines), verified:false + banner. Sections: zonation/anatomy, 6 hepatic cell types, normal physiology (glucose/lipid, CYP450 clearance, IGF-1/somatotropic axis), hallmarks-in-liver (senescence, regeneration decline, mito dysfunction, proteostasis, epigenetic/inflammaging, nutrient sensing), age phenotypes (MASLD, fibrosis, cholestasis, HCC), pathways table, partial-reprogramming section (LNP hepatotropism as a real delivery advantage; OSK-mRNA-LNP preclinical; NewLimit flagged company-reported/non-peer-reviewed; Yamanaka dysplasia caveat), other interventions (CR/rapamycin/senolytics/NAD+). Weaves in existing yang-2023-primate-liver-aging-snrna-srebp2. 14 footnotes, DOIs resolved via PubMed/Crossref.
  • molecules/proteins/lin28.md — new type: protein page (~275 lines), verified:false + banner, two-paralog treatment (LIN28A primary uniprot Q9H9Z2 + LIN28B Q6ZN17 in body). IDs resolved fresh via UniProt/NCBI/HGNC/Ensembl REST (LIN28A: gene 79727, HGNC 15986, ENSG00000131914, mouse Lin28a). GenAge entry not found → genage-id null. Covers let-7/TUT4-7 bistable switch, heterochronic origin, Thomson 5-factor + Lapasset OSKMNL reprogramming role, Shyh-Chang regeneration biology, let-7 tumor-suppressor / oncogenic-reactivation tension. literature-checked-through 2026-06-04 (R25 recency search run; Maklad 2023 + Krsnik 2022 integrated). 13 footnotes.

corrections during integration

  • Stripped seeder-introduced private paper-archive status annotations (the local store’s download/OA-status fields) plus citation-count clutter from all footnotes in BOTH pages + the liver banner — would have tripped leak discipline. Public pages cite by DOI only; local-store download/OA status is not part of the footnote schema.
  • Liver: seeder left a dangling [^xu2026inflammaging] ref (Kupffer M1/inflammaging claim) with no definition and implying an unverified “Xu 2026” paper — re-pointed to the already-cited Hunt & Cogger 2019 review + gap/needs-replication rather than assert a paper I couldn’t verify.
  • Lin28: seeder could not confirm “ZCCHC2” alias for LIN28B (NCBI lists only CSDD2) — correctly omitted.

gaps surfaced / new seed targets (implicit-stub wikilinks now created)

[2026-06-05] verify + ingest | liver-cluster: 2 pages verified, 10 new atomic pages seeded (user-prompted)

Continuation of the liver/reprogramming thread. “Verify then seed the new cell types and phenotypes.”

verified (wiki-verifier, full-PDF cross-check)

  • tissues/liver.md → verified:true (2026-06-05, claude). Corrections: LSEC fenestrae 100–200nm → 50–250nm (+ sieve-plate ~5% surface); Kupffer “M1 shift” framing walked back (Hunt/Cogger: ↑CD68/IL-6 only, no change in TNFα/Mrc1/Arg1/IL-10); Bird 2018 TGFβ1 “Kupffer-derived” → macrophage-derived (87% reduction on macrophage ablation); progenitor-blunting citation reassigned Bird→Hunt/Cogger; Duan 2023 art# 8392→8151 + ALOX15/catalase-KD detail; Yin 2024 “mouse+human”→mouse-only + pages; Mukherjee 2025 = Letter, mito-NAD+ pool; journal pages fixed. verified-scope notes 4 closed-access sources (KAT7/FOXO1/Klotz/Zhang-OSK) verified at abstract level only.
  • molecules/proteins/lin28.md → verified:true (2026-06-05). Corrections: rotenone → antimycin-A (in-vivo OxPhos inhibitor); adult digit-repair claim was an overstatement — Shyh-Chang 2013 found NO adult digit effect, regeneration is neonatal; added the let-7-independent arm (LIN28 directly enhances translation of Pfkp/Pdha1/Idh3b/Sdha metabolic mRNAs; glycolysis AND OxPhos both required); removed unverifiable “~14 U’s”; added OSNL-failure negative result to the OSKMNL section; “wound healing”→pinnal/ear tissue. Canonical IDs spot-confirmed (UniProt Q9H9Z2/Q6ZN17, mouse Lin28a). literature-checked-through 2026-06-05.
  • Propagation: greps confirm NO other page cites Bird 2018 or Shyh-Chang 2013 — corrections did not need to propagate.

seeded (10 new atomic pages, all verified:false + banner, queued for verifier)

  • cell-types: hepatocytes (CL:0000182), hepatic-stellate-cells (CL:0000632), kupffer-cells (CL:0000091), liver-sinusoidal-endothelial-cells (CL:1000398), cholangiocytes (CL:1000488), hepatic-progenitor-cells (CL:0002196). All CL IDs resolved fresh via EBI OLS4; markers + DOIs resolved fresh. Kupffer + HPC + LSEC pages mirror the verifier-corrected liver.md facts (no M1 overstatement; niche-driven HPC blunting; 50–250nm fenestrae).
  • phenotypes: masld (K76.0; built around 2023 NAFLD→MASLD/MASH nomenclature), liver-fibrosis (K74.0), age-related-cholestasis (K71.0; deliberately scoped as NOT-a-discrete-entity, heavy gap use), hepatocellular-carcinoma (C22.0). ICD-11 codes unresolved on all 4 (WHO browser not programmatically accessible) → gap/needs-canonical-id.

cleanup / discipline

  • Seeders + verifiers repeatedly wrote private paper-archive status annotations (the local store’s download/OA-status fields) plus citation-count clutter into footnotes. Stripped all such refs (clean across all 12 pages). Per user (2026-06-05): OA open/closed flags are fine to keep — only private-store references must go; gap/no-fulltext-access carries the load-bearing access signal. New memory: feedback_oa_flag_vs_tooling_ref.
  • Removed an invented icd-10-note: frontmatter field on age-related-cholestasis (schema invention; explanation retained in body Scope note).
  • Footnote ref/def integrity verified intact on all 12 pages post-cleanup.

deferred / gaps surfaced

  • Yang SREBP-2 study slug mismatch: studies/yang-2023-primate-liver-aging-snrna-srebp2.md has year: 2024 (Protein Cell 15(2), 2024). Slug says 2023 (received-date artifact). 6 inbound content links (pcsk9, srebp-2, ldlr, lipoprotein-metabolism, digestive-system, liver). Rename → yang-2024-… deferred (link churn; link resolves fine as-is).
  • hepatocytes.md [^adekunbi2024] DOI is tentative (PMID 38607532 confirmed, DOI inferred) — flagged for verifier.
  • 10 new pages are verified:false — next verifier batch. New stub wikilinks spawned: sasp, tgf-beta, notch-pathway, primary-biliary-cholangitis, primary-sclerosing-cholangitis, ursodeoxycholic-acid, telomere-attrition, csf1r, tlr4, nitric-oxide-signaling, pluripotency-network, oct4-sox2-nanog.
  • Pre-existing soft-leak noticed (NOT fixed at the time — out of session scope): phenotypes/androgenetic-alopecia.md L210 referenced the private store in a footnote; should be reworded to a neutral OA flag on a future lint pass. [UPDATE 2026-06-05: fixed — see incident entry below.]

[2026-06-05] maintenance | private-store footnote-annotation leak: fix + gate patch + liver.md incident

Triggered by user “fix that local paper archive mention” (androgenetic-alopecia.md).

fixed

  • phenotypes/androgenetic-alopecia.md — 6 footnote/verified-scope references to the private paper store reworded to neutral OA flags (“closed access”, “full-text PDF verified”); bronze-OA flag + gap/no-fulltext-access retained. The page is verified:true; edits cosmetic/privacy-only, no claim change.

systemic finding (surfaced, NOT mass-fixed per user decision)

  • The pattern is repo-wide and already in the initial public-release commit: ~134 content files carry footnote status annotations from the private store; ~432 reference the local-store full-text phrasing; ~448 carry bare OA-status jargon. These evade the hard leak-gate because that gate only matches the CLI-subcommand form, not the colon/status form or the full-text phrasing.
  • Severity moderate, not critical: none of these contain the hard-forbidden tokens (no project name, no CLI, no paths, no username) — they only soft-hint at a local store; the OA-status half is the part the user wants kept.
  • User decision: grandfather the ~450 existing pages (no mass rewrite); clean them in passing when a page is touched for other reasons.

leak-gate patched (CLAUDE.local.md — git-ignored, safe to edit)

  • Added **check 3: scans the working-tree diff vs HEAD (NEW/changed lines only, NOT all tracked files — so the grandfathered pages don’t permanently block commits) for the colon-form store annotations + the local-store full-text phrasing. OA flags themselves are intentionally NOT matched. Positive-control tested (correctly flags an injected line); negative-tested clean. Two of my own prior log lines that quoted the literal tokens as examples tripped it → reworded them generically (this entry avoids the literal strings too).

INCIDENT — liver.md destroyed and recovered

  • While testing the patched gate I ran git checkout -- tissues/liver.md to revert an injected test line. liver.md had a full verified ~300-line body that was working-tree-only (never committed), so checkout reverted it to the committed 22-line stub and the verified content was unrecoverable from git. Self-inflicted; careless use of a destructive command for a test cleanup.
  • Blast radius confined to liver.md: the other 11 new pages are untracked (checkout can’t touch them) and intact; other tracked edits preserved.
  • Recovery: re-seeded liver.md via wiki-seeder with all prior verifier corrections pre-baked, then re-ran wiki-verifier → verified:true (2026-06-05). Re-verification full-PDF-confirmed 5/10 sources, abstract-level 3, metadata 2. Net result is actually MORE accurate than the lost original: caught that Zhang 2026 J Control Release is vol 390 (lost original had 379) and refined the Bird 2018 87%-figure attribution to the ΔMdm2Hep model specifically.
  • Lesson: never git checkout/git restore a file with uncommitted working-tree changes; test gates against /tmp or throwaway paths, or remove the injected line by hand.

[2026-06-05] verify | liver-cluster batch — 10 new pages verified (6 cell types + 4 phenotypes)

Two parallel wiki-verifier batches (6 cell types, then 4 phenotypes), full-PDF cross-check, all flipped verified:true (2026-06-05). Cross-consistency enforced against the re-verified tissues/liver.md.

cell types (all verified:true)

  • hepatocytes — corrected [^adekunbi2024] DOI (tentative → 10.1007/s11357-024-01155-7, GeroScience 46(5)) + its overstated claim; Bird 2018 87%-figure re-attributed to ΔMdm2Hep model; Enkhbold 2015 strain (Balb/c) added; Sinha year; Zhang 2026 vol 379→390.
  • hepatic-stellate-cells — NK ligands MICA/MICB → MICA + ULBP2 (Krizhanovsky Fig 5D); Li 2020 “FBP1 KO” → hepatocyte-specific Fbp1 deletion (AAV8-TBG-Cre); p53-KO p=0.008, DKO p<0.01, 7–9-wk female strain added.
  • kupffer-cells — van der Tuin 2018 DOI corrected (10.1161/JAHA.117.007793 → .008105, the seeder-flagged suspect); Bird attribution; M1-rejection framing confirmed consistent with liver.md (no drift).
  • liver-sinusoidal-endothelial-cells — removed seeder-FABRICATED porosity numbers (the “6–8%→3–4%” cited to Le Couteur 2001 was unsupported; replaced with real Jamieson 2007 / Hunt 2019 data, ~30–50% decline); Hunt 2019 “blunted response” was BACKWARDS (paper says responsiveness mostly maintained) → corrected; eNOS→NOS (L-NAME is pan-NOS); caveolin “preserved”→“partially”. Best catches of the batch.
  • cholangiocytes — cell-mass fraction <5% → <10% human / 2–3% rodent; SCTR/CFTR/AE2 “large-duct-only” → medium AND large (Alpini 1996 3-population data); Boulter 2012 model CCl₄ → CDE+DDC; Jagged1 source → portal αSMA+ myofibroblasts.
  • hepatic-progenitor-cells — Tarlow DDC clarification; Hunt & Cogger reclassified mini-review + provenance note (the HPC-aging claim is their synthesis of ref [49], not primary data); niche-driven framing held; contested cell-of-origin kept gap/contradictory-evidence.

phenotypes (all verified:true)

  • masld — corrected fabricated “37 societies / 236 participants” → 236 panellists from 56 countries (Rinella 2023); Duan 2023 13-HODE confirmed consistent with liver.md (KD 2.418µM); resmetirom March-2024 approval confirmed. ICD-11 still null (#gap; WHO MMS API needs OAuth); prevalence-65plus still gap/unsourced (no 65+-stratified meta-analysis exists).
  • liver-fibrosis — propagation fixes from HSC page applied (Krizhanovsky strain/p-values/MICA+ULBP2, gap removed as it’s full-PDF-accessible; Bird ΔMdm2Hep); Poynard 2001 + Pradat 2007 progression rates kept but gap/no-fulltext-access added (closed). ICD-11 null gap.
  • age-related-cholestasis — scope-honesty preserved (not a discrete entity); “larger ducts” → “medium and larger” (cholangiocyte-page consistency); Lucena 2020 reframed — age is NOT a general DILI risk factor, it shifts DILI phenotype toward cholestatic; Tanaka 2024 page numbers corrected; removed a “780 matching records” citation-count artifact.
  • hepatocellular-carcinoma — Li 2020 conditional-KO fix; Rumgay 2022 (661k HCC / ASR 7.3 / 80%) confirmed; Mu 2015 hepatocyte-origin confirmed; flagged “males >65 3–4× higher” as a derived/unverified claim; GLOBOCAN-2022 supersession noted for next cycle. ICD-11 null gap.

discipline / cleanup

  • Verifiers again wrote private-store refs into 2 verified-scope fields (“not available in archive”, “archive status pending”) + 2 phenotype pages got the non-schema literature-checked-through ADDED then I removed — then re-added: phenotype IS atomic-content (carries verified block) and 14 existing phenotype pages already carry the field, so it belongs; the type:phenotype frontmatter block + recency cadence table in CLAUDE.md are simply missing a phenotype row (schema-doc gap, flagged below). All store-refs stripped; full 3-check leak gate clean; footnote integrity intact on all 10.

gaps / schema-doc items surfaced

  • CLAUDE.md schema-doc gap: type: phenotype frontmatter block omits literature-checked-through: and the recency cadence table has no phenotype row, yet ~14 phenotype pages carry it and phenotype is in the atomic-content verified list. Add a phenotype row (suggest 18mo, optional) + the field to the block on a future schema pass.
  • ICD-11 codes unresolved on masld/liver-fibrosis/HCC (WHO ICD-11 MMS API requires OAuth) — gap/needs-canonical-id; resolve via authenticated WHO browser later.
  • masld prevalence-65plus has no clean stratified figure — gap/unsourced retained.
  • GLOBOCAN-2022 (Bray 2024) likely supersedes Rumgay-2022 HCC incidence — refresh next cycle.
  • studies/yang-2023-primate-liver-aging-snrna-srebp2.md still verified:false (PMC OA) — candidate for a future verify pass; also the 2023-slug/2024-year rename still deferred.

Graph View

Backlinks