🧬 Aging Wiki

schema-history

24 min read

schema-history.md — CLAUDE.md schema escalation history

This file records the R-numbered schema escalations, precedent pages, and formalization dates that drove CLAUDE.md to its current form. CLAUDE.md carries the active rules; this file carries the why and the historical context — useful when a future agent is judging an edge case.

Each entry: R# — date — precedent page(s) then a brief description of what was added/changed/clarified.

R13 — Retracted papers convention

  • Precedent: pgc-1alpha
  • Established the bold “RETRACTED” footnote prefix + replacement-paper pattern for primary sources that have been retracted. Wenz 2009 PNAS muscle-PGC-1α paper was retracted Dec 2016 (retraction DOI:10.1073/pnas.1619713114) due to figure-integrity issues; Leick 2010 became the active citation for muscle-aging-PGC-1α.

R14 — Causal-graph fields

  • Added across hallmark/pathway/protein/process/compound/intervention schema:
    • mechanistic-tier: (hallmark) — finer than López-Otín; reflects this wiki’s intervention-evidence ranking
    • intervention-tractability: (hallmark) — whether a clinically usable modulator exists
    • caused-by: / causes: (hallmark, pathway, protein, process) — explicit upstream/downstream nodes in the causal graph
    • druggability-tier: (pathway, protein, process) — 1=clinical drug exists; 2=high-quality probe; 3=predicted druggable; 4=undruggable. From Open Targets Platform.
    • translation-gap: (compound, intervention) — categorical
    • next-experiment: (compound, intervention) — free-text one-liner

R15 — Causal graph driver

  • The R14 fields drive causal-graph-data. Surfacing relationships in frontmatter (not only in body prose) enables Dataview queries across the causal structure.

R20 — niche-signaling field

  • Added niche-signaling: to type: cell-type — upstream signaling pathways from the niche regulating this cell type (e.g., notch-pathway, wnt-beta-catenin for satellite cells).

R21 — Blood-product mode + biologic compound fields

  • Precedents: plasma-exchange, semaglutide (peptide compound, C187H291N45O59), aav-tert
  • Added mode: blood-product to type: intervention — for TPE, young-plasma transfer, NBE, transfusion-medicine aging interventions involving manipulation of blood/plasma constituents (not cells)
  • Added targets: to type: intervention — protein/pathway-level targets (AAV-TERT → TERT/TERC); empty for complex interventions like CR/exercise
  • Added biologic:, who-inn:, molecular-formula:, molecular-weight-da: to type: compound — for monoclonal antibodies, peptides, recombinant proteins where SMILES is impractical

R22 — External-dataset fields

  • Added across protein/cell-type/compound/intervention schema:
    • gtex-aging-correlation: (protein) — Spearman ρ + sign across GTEx tissues
    • mr-causal-evidence: (protein) — MR-validated as causal vs predictive
    • clinical-trials-active: (compound, intervention) — integer count from ClinicalTrials.gov v2 API
    • single-cell-aging-signature: (cell-type) — free-text summary from Tabula Muris Senis / CellxGene Census
  • Established SOPs: sops/finding-tissue-expression.md, sops/finding-population-evidence.md, sops/finding-druggability.md, sops/integrating-clinical-trials.md, sops/finding-singlecell-aging.md

R23d — Senolytics class-page aging-context clinical-stage

  • Precedent: senolytics
  • Established convention: intervention class pages use aging-context clinical-stage: (the stage appropriate to the aging indication, not max regulatory stage). Senolytics class = phase-2 even though dasatinib + quercetin individually have other FDA-approved indications.

R24 — limited-negative + class-page conventions + iPSC empty fields + cell-type verified

  • Precedents: sirtuin-activators (limited-negative), hsct (indication-split safety), iPSCs (ex-vivo cell-type), hematopoietic-stem-cells (HSC prototype)
  • limited-negative enum: Added to human-evidence-level: — for classes where multiple Phase 2/3 trials reported null/negative on primary endpoints (STACs / sirtuin-activators class)
  • Pipe-delimited free-text safety-profile:: When indication-split — e.g., well-established (hematologic disease) | investigational (aging-rejuvenation) for HSCT
  • Class-page aging-context clinical-stage (generalizing R23d senolytics): mtor-inhibitors = phase-2 (Mannick immune-aging trials), not fda-approved (rapamycin transplant indication)
  • iPSC empty fields: tissue-of-origin: [] acceptable (reprogrammed from variable somatic sources, not natively resident); key-aging-phenotypes: [] acceptable when cell type lacks canonical native aging-disease relationship. Document empty-field rationale in body.
  • type:cell-type verified discipline: Formalized — cell-type pages are atomic content originating quantitative claims (markers, niche, aging changes) and carry the four-field verified block. HSC was the prototype.

R24e — Antagonistic pleiotropy Mode B

  • Precedent: antagonistic-pleiotropy
  • Introduced two-mode treatment for type: hypothesis pages:
    • Mode A — evidence-aggregating — for hypotheses with specific testable predictions (free-radical, hyperfunction, information theory, negligible-senescence)
    • Mode B — conceptual-frame — for hypotheses that organize aging biology without sharp predictions (disposable-soma, antagonistic pleiotropy)
  • key-evidence-for: and key-evidence-against: REQUIRED for Mode A; OPTIONAL for Mode B (lint should suppress missing-field warnings for Mode B pages — [] with body-level consistency-with-evidence discussion is acceptable)

R25 — literature-checked-through field

  • Added literature-checked-through: to type: compound, intervention, biomarker, hypothesis
  • Populated by wiki-seeder on initial draft and wiki-verifier on each pass; date-filtered PubMed + Europe PMC search (last 5-7 yr)
  • Lint pass cadences: compound/intervention >12mo, biomarker >18mo, hypothesis >24mo

R26 — Multiple schema clarifications

  • type:pathway verified discipline: Formalized — pathway pages originate quantitative claims (key-node activities, EC50s, KO phenotypes) and carry verification like type:protein/process. Precedents: mtor, heat-shock-response
  • Pathway aging-context druggability-tier: Use aging-context tier (clinical drug exists for an aging indication that engages this pathway), not max-druggability of any node. Examples: mtor = tier 1 (rapamycin); heat-shock-response = tier 2 (HSP90 inhibitors are oncology-only).
  • not-applicable-somatic enum: Added to mr-causal-evidence: for proteins whose aging relevance is mediated by somatic mutation (CHIP drivers: DNMT3A, ASXL1, TET2, JAK2 V617F, SF3B1). Germline MR instruments don’t capture aging-relevant variation. Cite Genovese 2014 / Jaiswal 2014.
  • partial vs not-tested threshold: partial = GWAS hits exist for the gene but no MR study has been published (instrument available, study pending); not-tested = no instruments described OR MR attempted-but-underpowered.
  • Mode A/B key-evidence-* field requirements: REQUIRED for Mode A; OPTIONAL for Mode B (formalizes R24e treatment in schema).
  • FAPs cell-ontology-id null pattern: Precedent: fibro-adipogenic-progenitors. When no CL term resolves at appropriate granularity (FAPs: CL:0002320 “connective tissue cell” is too broad), set cell-ontology-id: null and tag #gap/needs-canonical-id with a one-line rationale. Don’t force-fit an over-broad term. Lint should aggregate #gap/needs-canonical-id into a standing watchlist for periodic CL update sweeps.

R27 — HSP90 + Waziry-2023 + study identifier escalations

  • HSP90 aging-context druggability-tier: Precedent: hsp90 = tier 2 (geldanamycin/tanespimycin advanced-clinical in oncology only, no FDA-approved HSP90 inhibitor and no aging-validated probe). Extends R26 pathway aging-context convention to type:protein.
  • type:study pmid: + pmc: fields: Precedent: waziry-2023-tame-trial. Added optional pmid + pmc fields. PMC particularly important — when local PDF is unavailable but paper is OA via PubMed Central, PMC full text is the canonical verification fallback. PMID useful for efetch-based author-list confirmation (Crossref author lists sometimes incomplete; PubMed efetch more reliable for recent papers).
  • UPR inbound count lesson: ROADMAP [ ] counts decay; inbound-count discovery is the canonical seeding-priority source. UPR was listed as 5 in ROADMAP but actual inbound count was 23. Use lint-pass § Step 3 inbound-count discovery for “top seeding targets” queries, not grep ROADMAP [ ].

R28 — Multiple schema escalations

  • brainstorming/ directory: Added as repository for conversation-derived working drafts; exempt from citation discipline; banner-flagged as non-wiki-quality; promote threads to atomic pages via standard seeder/verifier workflow.
  • introduced: proposed enum: Precedent: disabled-adaptive-immunity. Added proposed to type:hallmark introduced: enum — covers candidate hallmarks repeatedly proposed in literature but not yet in López-Otín’s official 12. Body must clearly state formal-adoption status in opening paragraph.
  • type:hallmark literature-checked-through:: Optional field; populate when hallmark page is intervention-landscape-heavy (e.g., disabled-adaptive-immunity covers active biotech intervention pipeline that turns over fast). Lint flags >24mo.
  • type:model-organism literature-checked-through:: Optional; populate when species is subject of active mechanistic-research literature. Precedents: loxodonta-africana (TP53-retrogene work — Sulak/Abegglen vs Nunney/Vollrath discordance), balaena-mysticetus (DNA-repair biology — Firsanov 2025 superseding earlier framing). Lint flags >24mo.
  • Seeder-brief DOI memory unreliable lesson: Don’t assert specific DOIs/years/journals for older primary sources in seeder briefs from training memory; pull from PubMed/Crossref first or delegate. (Lesson: 5 wrong DOIs across 4 pages, 3 on TAK1 alone.)

R33 — Process literature-checked-through + dietary subdirectory

  • type:process literature-checked-through:: Precedent: gut-microbiome-aging-shifts (2026-05-07). Optional; populate for processes covering fast-moving evidence streams (descriptive-process pages aggregating cohort-level findings); leave null for slow-turnover canonical-mechanism (autophagy, apoptosis). Lint flags >18mo.
  • interventions/dietary/ subdirectory: Added 2026-05-07. Gut-microbiome-modulation cluster justified its own subdirectory; previously dietary-mode interventions were collapsed into lifestyle/. Examples: prebiotics, postbiotics, fmt.
  • ROADMAP filename false positives lesson: Check canonical filename + aliases before dispatching ROADMAP items; pathway pages especially prone to filename-mismatch. (R33: scfa/tgf/wnt all false positives.)

R34 — Mouse-ortholog field formalization (2026-05-13)

  • Precedents: ~164 pages already used the field (Akt1, Sirt1, Tbk1, etc.)
  • Format conventions:
    1. Default — single symbol: mouse gene symbol in MGI sentence-case — Akt1, Sirt1, Cdkn1a
    2. Multi-subunit complexes: semicolon-delimited with parenthetical human-equivalent annotation — Rpa1 (RPA70); Rpa2 (RPA32); Rpa3 (RPA14) or Mfn2 (MFN2); Mfn1 (MFN1)
    3. Paralogs (mouse-duplicated where human has one, or vice versa): slash-delimited — Pot1a / Pot1b (mouse has two POT1 paralogs), Smad2 / Smad3, Ins1 / Ins2
    4. Inline annotation when useful: parenthetical alias — Hcar2 (also known as Puma-g), Has2 (UniProt P70312; ENSMUSG00000056752)
  • Supports lint-pass cross-checks between human (UniProt P04637 = TP53) and mouse (MGI Trp53) when reconciling cross-species claims. Absence does not invalidate a page; flag #gap/needs-mouse-ortholog when cross-species extrapolation discussion exists without an ortholog listed.

R35 — Pathway kegg-secondary + literature-checked-through

  • Precedent: type-i-interferon-signaling (2026-05-13)
  • kegg-secondary:: For pathways spanning multiple KEGG entries (common for broad signaling axes). Type I IFN signaling spans both hsa04623 cytosolic-DNA-sensing (production phase) AND hsa04622 RIG-I-like (RNA-sensor arm). Primary kegg: is the best single canonical match; secondaries supplementary. Lint pass should NOT treat secondary KEGG IDs as conflicting when the same ID appears as primary on a different pathway page (per the cgas-sting / type-i-IFN shared-ID case).
  • type:pathway literature-checked-through:: Optional; 18mo cadence. Populate for pages with active intervention landscapes (type-i-interferon-signaling: JAK inhibitors + NRTIs + STING antagonists; mtor: rapalogs; jak-stat-pathway: ruxolitinib/baricitinib/tofacitinib). Mirrors R28 hallmark + R33 process patterns.

R36 — Protein literature-checked-through

  • Precedents: sting, ikkε, jak1, tyk2, stat1, mavs, ifnar2 — 7+ pages already populated this field in practice before formalization (2026-05-13)
  • Optional; 18mo cadence. Populate for proteins targeted by active clinical-drug-development pipelines (kinases with FDA-approved inhibitors — JAK1/TYK2/IKKε/TBK1; senolytic targets — BCL-xL) or with fast-moving aging-specific research literature (STAT1 SASP-master-regulator papers; MAVS mt-dsRNA aging). Null acceptable for slow-turnover canonical-biology protein pages.

R38 — type:tissue formalization (2026-05-19)

  • Precedents: skeletal-muscle (~6 months de facto use), then skin, dermis, epidermis (R38 seeders flagged the documentation gap during the R38–R44 skin-aging coverage campaign)
  • Tissue pages are atomic content (originate quantitative claims about anatomy + cell composition + aging changes) and carry verification discipline like type:protein/compound/study. NOT navigational MOCs — those are type:framework.
  • Existing tissue pages as of R38 close (8): bone-marrow, brain, heart, myocardium, skeletal-muscle, skin, dermis, epidermis. Future per ROADMAP: hair-follicle, sebaceous-gland, subcutaneous-fat.

R39 — Study publication-type

  • Precedent: purohit-2016 (Letter publication type with condensed methods + limited n reporting + no structured abstract)
  • Optional field on type:study. Values: research-article | letter | brief-communication | review | meta-analysis | case-report | editorial | preprint. Material to evidence-weight interpretation: Letters have condensed methods + limited n reporting; meta-analyses + systematic reviews are higher-weight synthesis. Default research-article when unmarked.

R40 — Study volume/issue/pages

  • volume:, issue:, pages: were in use de facto across many study pages before formalization; recorded as optional but recommended frontmatter.

R41 — Compound chembl-id-active-metabolite

  • Precedent: tazarotene
  • For prodrugs where the active metabolite has a distinct ChEMBL ID. Tazarotene parent ester (CHEMBL1657) → tazarotenic acid active form (CHEMBL1201375). Common pattern in retinoid esters, ester-NSAIDs, statin lactones, peptidomimetic prodrugs. Body must explain the activation cascade + which form binds the target.
  • Canonical-DB memory unreliable lesson: Don’t assert ChEMBL/CAS/UniProt accessions from training memory in seeder briefs; they frequently resolve to unrelated compounds/proteins. Examples: bakuchiol ChEMBL → buparlisib; tazarotene ChEMBL → 5,6-dihydroxy-8-aminoquinoline; retinaldehyde ChEMBL → probenecid.

R42 — Compound pubchem-cid-alt + administration-route

  • Precedents: alpha-tocopherol, lactic-acid (pubchem-cid-alt); niacinamide (administration-route)
  • pubchem-cid-alt:: Optional supplementary CID for stereoisomers / alternative forms with distinct PubChem entries. α-tocopherol natural RRR (14985) + synthetic all-rac (2116); lactic-acid DL (612) + L-form (107689). Body must document which form the page primarily covers + the clinical/pharmacological distinction.
  • administration-route:: Optional; topical | oral | injectable | inhaled | intravenous | multi-route. Critical when route changes mechanism qualitatively — niacinamide topical (barrier repair, PAR-2 melanin transfer inhibition, anti-inflammatory) ≠ niacinamide oral (NAD+ precursor, sirtuin substrate). When mechanisms are conserved across routes, use multi-route or leave null.

R43 — Biomarker model-architecture + multi-model n-cpgs + MR not-applicable

  • Precedents: tapelift-clock-2026 (multi-model n-cpgs string format: elastic-net 157 + PC 5,021), bormann-epidermis-clock-2016 (SVM architecture, no compact CpG list — uses full 450k probe space), skin-autofluorescence-age-reader + age-reader (MR not-applicable)
  • n-cpgs-or-features: string format: For multi-model biomarker pages, accept string-encoded multi-value (e.g., “elastic-net 157 / PC 5021”) rather than forcing a single integer. Lint accepts both.
  • model-architecture: field: Optional; elastic-net | lasso | ridge | svm | pc-clock (Higgins-Chen 2022) | random-forest | composite-other. When svm + n-cpgs-or-features: null, the null is structurally accurate (SVM clocks use full probe space with no compact feature list); lint should NOT flag.
  • mendelian-randomization: not-applicable enum: For structural/device biomarkers that aren’t germline-instrumentable. Distinct from not-tested (no MR done yet but theoretically possible) and not-applicable-somatic (R26 enum, somatic-mutation-only proteins).

R44 — Microbe schema escalations + mode:procedural

  • Precedents: malassezia (kingdom: fungi, gram-stain: not-applicable-eukaryote, genome-size-mb genus-range), cutibacterium-acnes (oxygen-tolerance: aerotolerant-anaerobic, literature-checked-through), staphylococcus-epidermidis (literature-checked-through), dermatologic-resurfacing (mode:procedural)
  • type:microbe kingdom:: Optional; bacteria | archaea | fungi | protist. Critical for Dataview filtering of bacterial vs fungal pages when microbiome cluster grows. Malassezia (Basidiomycota) demonstrated the need vs bacterial precedents (Akkermansia, Cutibacterium, Staphylococcus).
  • gram-stain: not-applicable-eukaryote: Added enum value — fungi lack peptidoglycan, gram-stain is undefined.
  • oxygen-tolerance: aerotolerant-anaerobic: Added enum value — anaerobic-but-tolerates-O2 is a standard microbiology class distinct from facultative. Cutibacterium acnes precedent.
  • genome-size-mb string-range: For genus-level pages where a single number doesn’t fit (Malassezia ~7-9 Mb).
  • type:microbe literature-checked-through:: Optional; 18mo cadence. Populate for microbes with active aging-specific literature (skin microbiome cluster, gut microbiome cluster).
  • type:intervention mode: procedural: Device-mediated (lasers: CO2/Er:YAG/1550nm/1927nm; IPL; microneedling; RF; HIFU) or surgery-based interventions whose mechanism is controlled wounding / energy delivery rather than active pharmacological ingredients. Chemical peels can be classified either as pharmacological (active acid is the bioactive compound) or procedural (controlled wounding is the mechanism) — default to pharmacological per chemical-peels.md precedent.

R45 — type:experiment (2026-05-20)

  • Pages live in experiments/ and capture wet-lab or clinical experiments authored by the user (distinct from studies/, which holds primary sources extracted FROM).
  • Lifecycle: proposed (draft) → accepted (committed) → in-progress (data collection) → complete (data analyzed but not published) → published (results written up as a type:study page in studies/, linked via published-as:) → abandoned (with reason in body).
  • Experiment pages do NOT carry verified: — they are proposals/protocols, not facts about the world. Verification discipline kicks in on the downstream studies/ page when results are published.
  • resolves-edges: and resolves-nodes: are the load-bearing fields connecting an experiment to the causal-graph blocker tracker (see causal-graph-data).
  • Pre-registration encouraged but not required for proposed; increasingly important as status moves toward in-progress. When filed (OSF / ClinicalTrials.gov), set preregistered: true + populate preregistration-doi:.

R48 — Biomarker modality: transcriptomic (2026-05-29)

  • Precedent: transcriptomic-clock-tage (seeded 2026-05-29 from tyshkovskiy-2026-universal-transcriptomic-hallmarks, Nature 2026 — first transcriptomic clock page in the wiki).
  • modality: enum on type: biomarker gains the value transcriptomic (gene-expression / RNA-based aging clocks). Prior enum covered dna-methylation, proteomic, metabolomic, composite-clinical, imaging, telomere-length, physical-performance — but no RNA modality, despite transcriptomic clocks (Peters 2015, RNAAgeCalc, Tyshkovskiy 2026 tAge) being an established and now mortality-validated clock family.
  • A single biomarker page covers the tAge family (chronological / normalized-age / expected-mortality / maximum-lifespan / module-specific clocks) rather than one page per sub-clock, because they share training data, software (TACO / tAge R package) and a single canonical reference. n-cpgs-or-features: accepts the string-encoded multi-value form already permitted under R43.
  • CLAUDE.md status: type: biomarker schema block modality: line should add transcriptomic on next schema-revision pass. Lint should accept modality: transcriptomic.

R47 — type:tissue literature-checked-through (2026-05-23)

  • Precedent: bone (seeded 2026-05-23; active intervention landscape including senolytics, bisphosphonates, denosumab, romosozumab, teriparatide — all with rapidly evolving clinical evidence)
  • literature-checked-through: is formally extended as optional to type: tissue pages, following the same pattern established for type: process (R33), type: pathway (R35), type: protein (R36), type: hallmark (R28), and type: model-organism (R28).
  • Cadence: 18 months (matching protein/pathway/process; bone clinical landscape turns over at a comparable rate to active-drug-pipeline proteins).
  • When to populate: tissue pages with active pharmacological intervention landscapes (e.g., bone — senolytics RCTs, anti-resorptives, anabolics; skin — retinoids, procedural interventions). Leave null for canonical-structure tissue pages with slow-turnover evidence (e.g., a new adipose tissue page summarizing basic anatomy).
  • Lint should NOT flag null literature-checked-through: on type: tissue pages — the field is optional. Lint SHOULD flag the field when it is populated and exceeds the 18-month cadence.
  • CLAUDE.md status: field is already documented in the shared-conventions section for compound/intervention/biomarker/hypothesis; the addition here is precedent for applying the same field to tissue pages. CLAUDE.md type: tissue schema block should be updated to note this optional field on next schema-revision pass.

R46 — type:method (2026-05-20)

  • Precedent: mass-spec-age-hydrolysates (R46 pilot)
  • Lives in methods/. One page per laboratory or analytical technique whose methodological details and limitations recur across multiple study or atomic pages.
  • Explicitly distinct from sops/ (workflow guidance for wiki maintainers) and experiments/ (specific user-authored experimental protocols).
  • Atomic content (originates technical claims about a technique’s principle, validation, limitations, and evidence-weight implications); carries verification discipline.
  • Pilot-grade triage: only seed a methods page when its limitations recur across ≥3 study or atomic pages. Pre-emptive seeding of textbook techniques acceptable (per user direction R46) but should not crowd out evidence-driven coverage.
  • Current Protocols anchoring: when a Current Protocols chapter exists, cite it in current-protocols-citation: and reference it in the Workflow section. The wiki’s methods page summarizes for aging-context interpretation rather than reproducing protocol step-by-step.
  • Verification: cross-check claims against the cited Current Protocols chapter (if any) + canonical methods papers + 1–2 representative aging-context applications.

R49 — type:organ-system + phenotype verified-discipline fix (2026-06-02)

  • type:organ-system (new page type). Body-system navigational overlays live in organ-systems/ (a directory the original directory map reserved but never populated). They are the anatomical counterpart to type: framework: they aggregate and link the tissue/organ, cell-type, phenotype, and pathway atomic pages of one body system, and they do NOT carry the verified block (like frameworks, they originate no primary claims; verification = cross-link integrity).
    • Rationale: the parent-system: field on type: tissue pages (in use since the earliest tissue pages: cardiovascular-system, musculoskeletal-system, integumentary-system, nervous-system, hematopoietic-system) referenced system pages that did not exist, so [[cardiovascular-system]]-style links were all broken. The filename of an organ-system page IS the parent-system: slug, so those links now resolve.
    • Frontmatter: key-organs, key-tissues, key-cell-types, key-aging-phenotypes, related-hallmarks, parent-system (null for top-level; set for sub-systems like central-nervous-system → nervous-system). No verified block.
    • Index: by-organ-system (type: framework, in frameworks/) is the single top-level index + cross-system Dataview rollup (honors the single-Dataview-framework-page convention rather than per-folder _overview files).
    • Initial set (11): cardiovascular, musculoskeletal, integumentary, nervous, hematopoietic, immune (subsumes lymphatic via alias), endocrine, digestive, urinary, respiratory, reproductive. Sparse systems link existing pages + marked #stub planned pages; high-value organ anchors (kidney, liver, lung, pancreas, thymus, spleen) seeded as #stub type: tissue pages.
    • Glands/organs (thyroid, parathyroid, kidney, liver, lung, pancreas) live in tissues/ as type: tissue — there is no separate type: organ. CLAUDE.md type: tissue block updated to note this.
  • phenotype verified-discipline (correcting an earlier omission). type: phenotype was missing from CLAUDE.md’s “AI-extracted vs human-verified” required-types list and from the type: phenotype frontmatter block, even though 27/28 phenotype pages already carried a verified block. Both were corrected (added to the list + the block) during the 2026-06-02 hyperphosphatemia seed. Phenotype pages originate quantitative claims (prevalence, ICD codes, effect sizes, mortality associations) and carry full verification discipline.

R50 — ROADMAP.md retired (2026-06-02)

  • ROADMAP.md deleted; residual forward-intent folded into planned-coverage. The file had decayed: ~80% was closed-campaign archive (R31b–R45) duplicating log/, and its forward sections hadn’t been substantively updated since ~2026-05-20 despite heavy seeding since (bone cluster, organ-systems, etc.). It also disclaimed its own authority throughout and carried stale [ ] entries (e.g. neurodegeneration, already a page).
  • Why it was superseded (this generalizes the R27 lesson): seeding is now driven by (1) inbound-count discovery — lint-pass § Step 3, the canonical priority source; (2) ad-hoc seeding on new studies/user questions; and (3) distributed #stub markers at point-of-use (e.g. organ-system MOCs list their own planned pages), aggregated by the missing-page lint. Together these form a distributed roadmap that made the central static checklist redundant and decay-prone.
  • What survived into gaps/planned-coverage.md: only proactive intent that the reactive mechanisms miss — topics with zero inbound links (telomere DNA-structure page, melanocortin/neurotrophin/cAMP/SPM/melanogenesis protein clusters, several pathways, the HSP90/UPR/TAK1 proteostasis cluster, Tyshkovskiy long-tail) — plus the still-open cross-page propagation backlog.
  • References repointed: .claude/agents/wiki-seeder.md description (now names inbound-count + ad-hoc + planned-coverage), sops/lint-pass.md (§ Step 3 stub-queue destination + EXCLUDE_FILES), index.md header. Historical ROADMAP mentions in log/ and the R27/R33 lessons above are left as audit record. The earlier R27/R33 “use inbound-count, ROADMAP decays / filename-false-positives” lessons are the precedent this formalizes.

R51 — experiments/ relocated to private protocols/experiments/ (2026-06-02)

  • experiments/ (6 type: experiment pages + README) moved from the public top-level into the PRIVATE nested repo at protocols/experiments/. Rationale (user, 2026-06-02): the experiment pages had become user-authored working drafts more akin to protocols/brainstorming/ than to public research — speculative proposals, not extracted facts — so they belong in the private application tree, not the public wiki.
  • type: experiment itself is unchanged (schema, lifecycle, resolves-edges:/resolves-nodes:/adds-nodes: fields all retained); only the home directory + privacy status changed. The page type is still documented in CLAUDE.md § “Page types” alongside the other private-tree types (person/protocol/assessment/lab-panel). When an experiment reaches published, its results still graduate to a public type: study page via published-as:.
  • Link-discipline consequence (the load-bearing part): the experiment↔causal-graph integration is now one-way. Experiment pages may link OUT to public research (incl. [[frameworks/causal-graph-data]]), but public pages must NOT link back. The four public inbound references were de-linked, keeping the prose: frameworks/causal-graph-data.md (worked-example column, scale table, edge matrix, “Resolving experiments” bullets — experiments now named in plain text as proposed/not-yet-run), index.md (the ”## Experiments” Dataview section removed entirely), and phenotypes/chronic-venous-disease.md (two refs reworded to plain literature-gap statements, no dangling “proposed” antecedent). Per user follow-up, public pages must NOT reference a “private experiments tracker” or an unanchored “proposed” experiment — just state the experimental gap.
  • 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 in CLAUDE.local.md and to the public-repo invariant text in CLAUDE.md (both protocols/experiments and the bare experiments/-prefixed form). Documentation describes these forbidden links in prose with bare backticked paths, never the literal bracket form, to avoid self-matching the gate. .gitignore already excludes all of protocols/, so the move needed no .gitignore change.

R52 — clinical-stage: approved-ex-us (2026-06-09)

  • Precedent: pdrn (polydeoxyribonucleotide/polynucleotide injectables — CE-marked + Korea MFDS-approved for aesthetic/wound-healing indications, not FDA-approved). Surfaced during the injectable-skin-boosters build-out (2026-06-09); both the seeder and the verifier independently escalated that the clinical-stage: enum had no value for “approved by a stringent non-US regulator but not FDA,” forcing a phase-2 surrogate.
  • New enum value: approved-ex-us added to the type: compound clinical-stage: enum (preclinical | phase-1 | phase-2 | phase-3 | approved-ex-us | fda-approved | supplement). By the existing convention (aging-context clinical-stage on intervention class pages, CLAUDE.md § druggability conventions) it also applies to clinical-stage: on type: intervention class pages.
  • Semantics: denotes real-world marketing approval for the relevant aging/aesthetic indication by EMA/EU (incl. CE marking), UK MHRA, Japan PMDA, South Korea MFDS, or comparable stringent regulator, without FDA approval. Body must name the jurisdiction(s). When approved in BOTH the US and elsewhere → use fda-approved (body notes the rest); approved-ex-us is specifically the US-gap case.
  • Rationale (user, 2026-06-09): ex-US approval is not a weaker signal than FDA approval, and can be a stronger or earlier one. Motivating cases: PDRN injectables (mainstream in EU/Korea aesthetics, US-investigational) and next-generation sunscreen UV filters (bemotrizinol, bisoctrizole, etc.) that are EMA/global-approved but stuck in the US OTC-monograph backlog — there, “not FDA-approved” reflects US regulatory lag and FDA’s drug-vs-cosmetic classification, not evidence weakness. The wiki needs to record that distinction rather than collapse it into a misleadingly-low phase-N.
  • Applied: pdrn clinical-stage: phase-2 → approved-ex-us; body regulatory note rewritten accordingly. No other existing page currently needs the value (biostimulatory-fillers PLLA/CaHA are FDA-approved; PCL/PDO are CE-only but live on an implemented intervention class page).
  • Lint: accept approved-ex-us as a valid clinical-stage: on compound + intervention pages. No cadence/required-field change.
  • CLAUDE.md status: enum line + a dedicated “Ex-US regulatory-approval convention (R52)” paragraph added to the shared druggability-conventions section in this pass (not deferred).

R53 — type:exposure (2026-06-11)

  • New page type for modifiable environmental/behavioral risk factors that are neither a compound, an intervention, nor a phenotype: tobacco smoking, alcohol, ambient air pollution, and similar exposures. Lives in a new top-level exposures/ directory.
  • Rationale (user, 2026-06-11): surfaced while wiring node-description tooltips + wiki links into the aging-simulator (viz/aging-simulator.html). The simulator’s exogenous-input nodes (smoking, alcohol, PM2.5 air pollution) had zero wiki coverage and no schema home — they are exposures the individual is subjected to / chooses, not drugs taken (type: compound), not deliberate regimens (type: intervention, which frames the modifiable lever — smoking-cessation — awkwardly for an ambient exposure like PM2.5), and not conditions of the organism (type: phenotype). The user chose to formalize a dedicated type rather than force a half-fit or leave them link-less. (The disease nodes from the same pass — COPD, CKD, hypertension, obesity — fit type: phenotype cleanly and were seeded there, not here.)
  • Frontmatter: exposure-class (behavioral | environmental | occupational | dietary), agent (the harmful constituent(s)), measure (how it’s quantified — pack-years, g/day ethanol, µg/m³ PM2.5), modifiable (yes | no | partial), dose-response (monotonic | j-curve | threshold | sublinear | supralinear), target-hallmarks, target-pathways (opt), downstream-phenotypes, mechanisms, human-evidence-level (mirrors compound), reversibility (free-text; cessation effect + time-course), mendelian-randomization (reuses the protein/biomarker enum — smoking/alcohol are MR-instrumented via CHRNA3/ADH1B/ALDH2), literature-checked-through (required, 18mo cadence — risk-factor epidemiology moves but slower than drug pipelines).
  • Verification discipline: atomic content (originates relative risks, dose-response curves, MR estimates) → carries the full four-field verified block like compound/biomarker/phenotype.
  • Initial set (3): smoking, alcohol, air-pollution (seeded verified: false). Distinct from interventions/dietary/sodium-restriction etc., which remain interventions (a chosen reduction regimen); a dietary component viewed as harmful exposure (e.g. dietary saturated fat) stays on its compound page (molecules/compounds/palmitic-acid) unless it earns its own synthesis page.
  • Lint: exposures/ added to the atomic-content directories; type: exposure added to the verified-required + literature-recency-required type lists (18mo).

R54 — type:phenotype literature-checked-through (2026-06-11)

  • Added the optional literature-checked-through: field to type: phenotype (18mo cadence, optional — populate for fast-moving phenotypes, leave null for slow-turnover descriptive ones). Mirrors the existing optional treatment on protein/pathway/process/hallmark/microbe/model-organism.
  • Rationale: during the R53 seed batch (COPD, CKD, hypertension, obesity — all phenotype pages with active intervention + epidemiology literature), three of the four wiki-seeder agents independently flagged that they had run the R25 recency search but had nowhere in the phenotype frontmatter to record literature-checked-through:, so they omitted it (correctly, per the then-current schema). The convergent signal from independent agents made the gap clear: phenotypes like CKD (SGLT2i/finerenone), hypertension (SPRINT-era targets), and obesity (GLP-1/SURMOUNT) have fast-moving therapeutic landscapes where a recency stamp is as load-bearing as on a compound page.
  • Scope: optional (not required) — most phenotypes are slow-turnover descriptive entities (e.g. hair-greying, ear-nose-enlargement) where the field would be noise. Lint flags it only when populated and stale, same as the other optional types.
  • Applied: the four R53 phenotype pages (copd, chronic-kidney-disease, hypertension, obesity) backfilled with literature-checked-through: 2026-06-11 (each had run the recency search at seed time).

R55 — biomarker modality: vital-sign (2026-06-11)

  • Added vital-sign to the type: biomarker modality: enum. For single-parameter resting physiological measures — resting heart rate, resting blood pressure, heart-rate variability, respiratory rate, body temperature — that index aging/mortality risk but are neither molecular clocks nor performance tests.
  • Rationale (user, 2026-06-11): seeding biomarkers/resting-heart-rate-biomarker.md (to ground the aging-simulator’s resting-HR node) exposed that no existing modality fit: it is not a composite-clinical score (single parameter, not a multi-feature index like PhenoAge), not physical-performance (no performance/exertion test — it’s a resting vital sign), and not any of the omics/imaging/telomere modalities. The seeder used physical-performance under protest and flagged the misfit; the verifier confirmed it and (correctly) did not invent an enum value. vital-sign was chosen over the narrower autonomic because it also cleanly homes resting BP, respiratory rate, and temperature (not all of which are purely autonomic).
  • Lint: for modality: vital-sign, the clock-feature fields (n-cpgs-or-features, training-cohort, model-architecture, test-retest-icc) are structurally null/N-A — lint must NOT flag their absence (same precedent as the svm + null n-cpgs-or-features carve-out).
  • Applied: resting-heart-rate-biomarker modality: physical-performance → vital-sign; body note explaining the prior misfit replaced with a note recording the R55 classification.

R56 — biomarker training-target: exposure (2026-06-14)

  • Added exposure to the type: biomarker training-target: enum (chronological-age | mortality | morbidity | pace | exposure). For DNA-methylation / omic predictors trained to estimate an external exposure or behavioral quantity — alcohol intake, smoking intensity, BMI, dietary pattern — rather than a biological age, mortality, morbidity, or pace-of-aging outcome.
  • Rationale (user, 2026-06-14): seeding dnam-alcohol-consumption (the 144-CpG DNAm-Alc predictor behind TruDiagnostic’s “Alcohol epigenetic biomarker”) exposed that no enum value fit — it predicts grams/AUDIT of alcohol consumed, not an age or vital outcome. The seeder used consumption under protest and flagged the gap; the verifier correctly left it for adjudication rather than inventing a value. exposure was chosen over the narrower consumption because it generalizes to the whole class of DNAm exposure-predictors (the same elastic-net/LASSO-on-CpGs architecture is used for smoking pack-years, BMI, and diet scores), several of which are likely future seeds.
  • Note: these exposure-predictor biomarkers are distinct from type: exposure pages (R53) — the exposure page is the risk factor itself (e.g. alcohol); an exposure-target biomarker is the methylation readout that estimates that exposure. The two cross-link (the exposure page links to its DNAm readout, and vice versa).
  • Applied: dnam-alcohol-consumption training-target: consumption → exposure.

Graph View

Backlinks