Editorial Standards

Inclusion requirements

Varia's V1 catalog is intentionally narrow. The retained set covers 49 variants across 12 domains, and every addition is expected to satisfy the same rule set that governs the current catalog. See the Variant Catalog for the full accepted list.

1. Multiple-independent-source requirement. At least two papers from separately-authored teams published in Tier 1 or Tier 2 journals. Single-paper findings, even when that paper claims internal replication across cohorts, wait for an independent second publication.
2. 180-day Watch period. A new SNP candidate first appearing in literature does not enter the database the same review cycle. It sits on the Watch list for 180 days while additional evidence accumulates. After 180 days, if the multi-source requirement is met and no concerning challenges have surfaced, the variant is promoted to merge-candidates in the next monthly review.
3. Domain fit. The variant fits within one of Varia's existing 12 domains. New domain additions are a deliberate product decision, not a content decision.
4. Insight language conformance. Per the existing MONTHLY_DB_REVIEW.md prompt: "associated with" not "causes"; "research suggests" not "proven"; "discuss with your physician" for high or moderate findings; never "you should," "you must," "diagnose," "treat," "prescribe," "clinical-grade," or "medical advice."

Scope discipline

The catalog is curated rather than exhaustive. The goal is not to publish every reported association. The goal is to publish a smaller set of entries that survive explicit source standards, explicit language discipline, and explicit review cadence. Public pages can link outward to The Genome page for scan mechanics and to Sources page for the named authorities behind the curation process.

Inclusion requirements

Varia's V1 catalog stays intentionally small. The retained set covers 49 variants across 12 domains, and new entries are expected to clear the same gate before they appear in the public product. See the Variant Catalog for the full accepted list.

1. Multiple-independent-source requirement. At least two papers from separately-authored teams published in Tier 1 or Tier 2 journals. Single-paper findings, even when that paper claims internal replication across cohorts, wait for an independent second publication.
2. 180-day Watch period. A new SNP candidate first appearing in literature does not enter the database the same review cycle. It sits on the Watch list for 180 days while additional evidence accumulates. After 180 days, if the multi-source requirement is met and no concerning challenges have surfaced, the variant is promoted to merge-candidates in the next monthly review.
3. Domain fit. The variant fits within one of Varia's existing 12 domains. New domain additions are a deliberate product decision, not a content decision.
4. Insight language conformance. Per the existing MONTHLY_DB_REVIEW.md prompt: "associated with" not "causes"; "research suggests" not "proven"; "discuss with your physician" for high or moderate findings; never "you should," "you must," "diagnose," "treat," "prescribe," "clinical-grade," or "medical advice."

Scope discipline

Curation is the whole point. Out of every association that gets published, Varia keeps only the smaller set that clears explicit rules for sources, wording, and review timing. If you want the scan mechanics, The Genome page covers them; for the named reference bodies behind the curation, see the Sources page.

How the pipeline works

Per VEGS § 11, Varia runs a transparent curation pipeline: human-authored rules and thresholds; mechanical application at scale via scripts/editorial_grader/ (four class-specific graders for Pathogenic, Pharmacogenomic, Risk-modifier, and Trait/lifestyle entries). Named detectors monitor upstream authority changes on a documented cadence (ClinVar weekly/monthly, ClinPGx weekly, Retraction Watch weekly, PharmVar RSS plus monthly fallback, ACMG SF via Genetics in Medicine RSS).

Residual human contribution (quantified)

As of 2026-05-28 close artifacts: 32 PMID corrections adjudicated; 15 conflict-of-evidence summaries hand-authored (24 catalog rows with genotype mirrors); 12 CPIC retiering rows human-adjudicated; 9 authority-contradiction reconciliations (1.9% of 470 graded entries, matching PharmGKB's published override-disclosure posture); 628 VEAS citations under human-authored templates. Grader events use algorithm_default_applied vs review_required dispositions. See the Methodology page for the full table.

Override surface

Human-in-the-loop does not mean hidden solo judgment. Ambiguous-threshold Risk-modifier and Trait/lifestyle Well-replicated cases log as algorithm_default_applied (conservative default stands). Authority contradictions, retractions, schema violations, and data-quality gaps emit review_required and fold into the monthly review surface.

How the pipeline works

Varia grades variants with published rules that humans write and software applies consistently. Four graders handle different finding types, and automated checks watch ClinVar, medication guidelines, retractions, and related authorities on a published schedule.

Residual human contribution (quantified)

Varia publishes how much human work remains: 32 PMID fixes reviewed by hand; 15 conflict summaries for split literature; 12 medication-response retiers; 9 authority disagreements reconciled (1.9% of graded entries); 628 citation annotations under human-authored voice rules. Bootstrap threshold cases are logged as algorithm-default, not a standing queue. Details are on the Methodology page.

Override surface

When the rules cannot settle a case on their own, Varia records what happens next: either the conservative default stands, or the case goes to the monthly review surface where the decision is visible. Every call is on the record.

Varia does not fuse ACMG pathogenicity, GWAS population risk, CPIC actionability, and GRADE certainty into one significance ladder because those scales are non-commensurable (VEGS § 6.1). The v1.2 faceted schema keeps four native-scale facets over one provenance core, following the GA4GH/ClinGen native-profile pattern.

Non-equivalence: facet labels are not comparable across facets. Varia sets non_equivalence_declared: true on every entry and surfaces live cross-facet disagreement where facets diverge. Per-facet concordance validation is planned (R7); no concordance numbers are published until that pre-registered study completes.

Different kinds of genetic evidence stay on their own scales. Varia does not force rare-disease grades, population risk, medication guidance, and evidence strength into a single ranking, because they answer different questions and are not interchangeable.

Labels on different parts of a finding are not directly comparable. Varia flags cases where scales disagree. Formal validation against external authorities is planned but not yet published.

Population-risk and polygenic outputs are ancestry-sensitive. European-calibrated effect sizes do not transfer as constants to under-represented ancestries. Varia presents transferability as ranges with uncertainty widening (R3 behavior on Facet B metadata) and tags heterogeneity_by_ancestry conflicts where literature documents attenuation.

Many risk estimates come from studies where most participants had European ancestry. Varia treats those findings as ancestry-sensitive: wider uncertainty when the science is less representative of your background, not one constant for everyone.

Any ranked view Varia shows is an explicit, labeled actionability or attention ordering, never a significance or truth ranking. Varia performs no arithmetic across facets and publishes no cross-facet aggregate score. See the Methodology page for the full policy.

When Varia sorts findings for you, that sort is about what may warrant attention in a clinician conversation, not which finding is "more true" in a single scientific sense. Varia does not add different evidence scales together into one score.

Tier 1

Tier 1 sources are the journals Varia treats as citable without question. The broader conventions file defines the allowlist as these family and specialty groupings:

• Nature family: Nature, Nature Medicine, Nature Genetics, Nature Communications, Nature Neuroscience, Nature Aging, Nature Immunology
• Cell family: Cell, Cell Metabolism, Cell Reports
• Science family: Science, Science Translational Medicine
• NEJM (New England Journal of Medicine)
• Lancet family: Lancet, Lancet Neurology, Lancet Diabetes & Endocrinology, Lancet Healthy Longevity
• JAMA family: JAMA, JAMA Neurology, JAMA Cardiology, JAMA Internal Medicine, JAMA Network Open
• BMJ
• Circulation, Circulation Research, Circulation: Genomics and Precision Medicine
• European Heart Journal (Oxford / European Society of Cardiology)
• JACC (Journal of the American College of Cardiology)
• Brain
• Neuron
• Diabetes (American Diabetes Association), Diabetologia, Diabetes Care
• American Journal of Human Genetics
• Genetics in Medicine (ACMG)
• Alzheimer's & Dementia
• Journal of Clinical Endocrinology and Metabolism (JCEM)
• Endocrine Reviews
• Immunity (Cell Press)
• Journal of Experimental Medicine
• Annals of Internal Medicine, Annals of Neurology

The source registry also keeps explicit anchor entries for the journals the grading code uses directly:

• Nature (Nature): Code-level Tier 1 anchor journal used as part of the direct allowlist that supports high-confidence editorial grading. Last reviewed by Varia: 2026-05-26.
• Cell (Cell): Code-level Tier 1 anchor journal used when Varia checks whether a citation clears the highest editorial source tier. Last reviewed by Varia: 2026-05-26.
• Science (Science): Code-level Tier 1 anchor journal used by the grading pipeline when source-caliber decisions are made. Last reviewed by Varia: 2026-05-26.
• New England Journal of Medicine (NEJM): Code-level Tier 1 anchor journal used for top-tier clinical literature in the editorial grader and source discipline. Last reviewed by Varia: 2026-05-26.
• JAMA (JAMA): Code-level Tier 1 anchor journal that helps define Varia's highest-confidence journal set. Last reviewed by Varia: 2026-05-26.
• Nature Communications (Nat Commun): Code-level Tier 1 anchor journal used to recognize high-caliber translational and population-scale genetics work. Last reviewed by Varia: 2026-05-26.
• Science Translational Medicine (Sci Transl Med): Code-level Tier 1 anchor journal used when translational evidence supports a user-facing editorial claim. Last reviewed by Varia: 2026-05-26.

Tier 2 and excluded sources

Tier 2 sources remain citable, but they carry closer review when a claim is being written into user-facing copy:

• PLOS Genetics, PLOS Medicine, PLOS Biology (PLOS One specifically excluded)
• Genome Medicine, Genome Biology, Genome Research
• Human Molecular Genetics
• Neurology (American Academy of Neurology)
• Molecular Neurodegeneration
• Neurobiology of Aging
• Translational Psychiatry
• Arteriosclerosis, Thrombosis, and Vascular Biology (ATVB)
• eLife
• Atherosclerosis
• Journal of Lipid Research
• Aging Cell
• Pharmacogenetics and Genomics
• Clinical Pharmacology and Therapeutics
• Acta Neuropathologica

Tier 3 and Tier 3 override

Tier 3 sources can ground Strength labels of Mixed evidence, Single-study, or Disputed by recent work per VEGS § 8. Tier 3 may ground Strength Consistent only when all of the following hold: cohort size N > 50,000; genome-wide significance (p < 5×10⁻⁸ for GWAS findings); effect-size consistency with prior smaller cohorts in the same direction; and documented override rationale in strength_rationale.

Excluded sources do not enter the primary citation backbone:

• All MDPI journals (Nutrients, Cells, IJMS, Biomolecules, etc.)
• Most Frontiers family journals
• PLOS One specifically (despite PLOS Genetics/Medicine/Biology being citable)
• Hindawi journals (acquired by Wiley after paper-mill scandal; brand contaminated)
• Any journal not MEDLINE-indexed
• Conference abstracts and proceedings
• Preprints (bioRxiv/medRxiv), surface in What We're Watching only, never as primary citation until peer-reviewed publication

Tier 1

Tier 1 sources are the journals Varia treats as trusted enough to cite without hesitation. The broader conventions file defines the allowlist through these family and specialty groupings:

• Nature family: Nature, Nature Medicine, Nature Genetics, Nature Communications, Nature Neuroscience, Nature Aging, Nature Immunology
• Cell family: Cell, Cell Metabolism, Cell Reports
• Science family: Science, Science Translational Medicine
• NEJM (New England Journal of Medicine)
• Lancet family: Lancet, Lancet Neurology, Lancet Diabetes & Endocrinology, Lancet Healthy Longevity
• JAMA family: JAMA, JAMA Neurology, JAMA Cardiology, JAMA Internal Medicine, JAMA Network Open
• BMJ
• Circulation, Circulation Research, Circulation: Genomics and Precision Medicine
• European Heart Journal (Oxford / European Society of Cardiology)
• JACC (Journal of the American College of Cardiology)
• Brain
• Neuron
• Diabetes (American Diabetes Association), Diabetologia, Diabetes Care
• American Journal of Human Genetics
• Genetics in Medicine (ACMG)
• Alzheimer's & Dementia
• Journal of Clinical Endocrinology and Metabolism (JCEM)
• Endocrine Reviews
• Immunity (Cell Press)
• Journal of Experimental Medicine
• Annals of Internal Medicine, Annals of Neurology

The source registry also keeps direct anchor entries for the journals the grading code uses by name:

• Nature (Nature): Code-level Tier 1 anchor journal used as part of the direct allowlist that supports high-confidence editorial grading. Last reviewed by Varia: 2026-05-26.
• Cell (Cell): Code-level Tier 1 anchor journal used when Varia checks whether a citation clears the highest editorial source tier. Last reviewed by Varia: 2026-05-26.
• Science (Science): Code-level Tier 1 anchor journal used by the grading pipeline when source-caliber decisions are made. Last reviewed by Varia: 2026-05-26.
• New England Journal of Medicine (NEJM): Code-level Tier 1 anchor journal used for top-tier clinical literature in the editorial grader and source discipline. Last reviewed by Varia: 2026-05-26.
• JAMA (JAMA): Code-level Tier 1 anchor journal that helps define Varia's highest-confidence journal set. Last reviewed by Varia: 2026-05-26.
• Nature Communications (Nat Commun): Code-level Tier 1 anchor journal used to recognize high-caliber translational and population-scale genetics work. Last reviewed by Varia: 2026-05-26.
• Science Translational Medicine (Sci Transl Med): Code-level Tier 1 anchor journal used when translational evidence supports a user-facing editorial claim. Last reviewed by Varia: 2026-05-26.

Tier 2 and excluded sources

Tier 2 sources are still usable, but they get closer scrutiny before they are turned into user-facing claims:

• PLOS Genetics, PLOS Medicine, PLOS Biology (PLOS One specifically excluded)
• Genome Medicine, Genome Biology, Genome Research
• Human Molecular Genetics
• Neurology (American Academy of Neurology)
• Molecular Neurodegeneration
• Neurobiology of Aging
• Translational Psychiatry
• Arteriosclerosis, Thrombosis, and Vascular Biology (ATVB)
• eLife
• Atherosclerosis
• Journal of Lipid Research
• Aging Cell
• Pharmacogenetics and Genomics
• Clinical Pharmacology and Therapeutics
• Acta Neuropathologica

Tier 3 and Tier 3 override

Tier 3 sources can support Mixed evidence, Single-study, or Disputed by recent work grades per VEGS § 8. They can support Consistent only when the study has more than 50,000 participants, reaches genome-wide significance for GWAS findings, agrees in effect direction with earlier smaller studies, and carries a written rationale in the catalog.

Excluded sources do not become the main citation for a Varia claim:

• All MDPI journals (Nutrients, Cells, IJMS, Biomolecules, etc.)
• Most Frontiers family journals
• PLOS One specifically (despite PLOS Genetics/Medicine/Biology being citable)
• Hindawi journals (acquired by Wiley after paper-mill scandal; brand contaminated)
• Any journal not MEDLINE-indexed
• Conference abstracts and proceedings
• Preprints (bioRxiv/medRxiv), surface in What We're Watching only, never as primary citation until peer-reviewed publication

Varia surfaces institutional authority data (ACMG/AMP via ClinVar, CPIC and PharmGKB via ClinPGx, ClinGen, PharmVar) verbatim where a clinical framework already grades the variant. VEGS applies Varia's own Strength and Conversation Priority grading only to Risk-modifier and Trait/lifestyle classes, where no authoritative clinical grader exists.

The attribution "Backed by Varia editorial assessment" applies only to VEGS-graded classes. Pathogenic and Pharmacogenomic findings display the authority's label, not a parallel Varia-invented grade.

Conversation Context (descriptive: which authorities cover the finding) and Conversation Priority (Varia editorial framing of discussion warranting) are defined on the Methodology page per VEGS § 4 Axis 2 and Axis 2b.

When a medical authority has already graded a variant, Varia shows that grade directly instead of inventing a second one. That includes ACMG classifications through ClinVar, medication guidelines through CPIC and PharmGKB (via ClinPGx), ClinGen, and PharmVar star-allele naming.

Varia's own grading applies only to common-trait risk modifiers and lifestyle traits, where no clinical authority publishes a grade. The label "Backed by Varia editorial assessment" appears only on those Varia-graded findings.

For how Conversation Context and Conversation Priority work together, see the Methodology page.

Primary citation rules

The default posture is primary literature first, authority surface second, prose third. A claim should trace cleanly to a named paper, a named institutional authority, or a named registry entry. When a finding depends on a source page rather than a paper, Varia names that authority directly and links it on the Sources page or the Institutional Authority References page.

Recommendation boundary

Varia does not use the Editorial Standards page to prescribe treatment, supplements, products, or protocols. The page describes source discipline, citation discipline, and evidence discipline. When Varia names a biomarker or a drug class, it is naming the discussion object, not issuing a treatment instruction.

Primary citation rules

The default rule is simple: a claim should point back to a paper, a named authority, or a hand-authored registry entry. If the support comes from an authority page rather than a paper, Varia names that authority directly and links it on the Sources page or the Institutional Authority References page.

Recommendation boundary

Varia does not use this page to tell readers what product to buy, what supplement to take, or what treatment to start. When Varia names a biomarker or a drug class, it is naming the topic for discussion, not giving a treatment instruction.

Trigger categories

The override surface uses explicit dispositions. algorithm_default_applied covers ambiguous-threshold Risk-modifier cases, low effect-direction extraction, and Trait/lifestyle Well-replicated flags. review_required covers authority contradictions, retractions, hard-rule conflicts, and incomplete authority data.

Public log

The event log is append-only and reviewable. review_required items follow the 14-day SLA with auto-downgrade on miss. That is the operational contract behind the public claim that editorial discipline is visible rather than implied.

Trigger categories

When the published rules hit a gray zone, Varia labels it in the log: either the safe default already applied, or it belongs on the monthly review list. Nothing disappears into untracked solo judgment.

Public log

The event log is append-only and public. Genuine review items carry a 14-day SLA; a missed month makes routing more conservative, not silently wrong.

Conflict types (VEGS § 6)

When evidence conflicts, each conflict carries a conflict_type label. Eleven values are in scope for V1:

• failed_replication
• effect_direction_reversal
• heterogeneity_by_ancestry
• heterogeneity_by_phenotype
• heterogeneity_by_sex
• publication_bias_suspected
• retracted_supporting_paper
• withdrawn_by_authors
• expression_of_concern (v1.1)
• corrected_post_publication (v1.1)
• major_methodology_revision (v1.1)

When Varia calls evidence mixed

Varia distinguishes between disagreement in the literature and low evidence in the literature. Mixed evidence means conflicting direction or conflicting replication across multiple studies. Single-study means the literature is still thin. Disputed by recent work means newer evidence contradicts the earlier consensus. Those categories are public because they materially change how a reader should interpret a citation list.

What the page promises

When evidence conflicts, Varia surfaces the conflict rather than smoothing it away. The public-facing result is the "Evidence is mixed" callout, paired with conflicting citations and plain-language summary text. The Editorial Standards page exists in part so a reader can see that conflict handling is part of the design, not a special case added after the fact.

Conflict types (VEGS § 6)

When studies disagree, Varia tags the disagreement with one of eleven named conflict types, including failed replication, effect direction reversal, heterogeneity by ancestry, phenotype, or sex, publication bias suspected, retracted or withdrawn supporting papers, expression of concern, post-publication correction, and major methodology revision.

When Varia calls evidence mixed

Varia separates disagreement from thin evidence. Mixed evidence means multiple studies disagree. Single-study means the literature is still sparse. Disputed by recent work means newer evidence pushes back on an earlier consensus. Those labels matter because they change how a reader should treat the citation list.

What the page promises

When evidence conflicts, Varia surfaces the conflict rather than flattening it. The public-facing result is the "Evidence is mixed" callout with contradictory citations and plain-language summary text. The point is to show that conflict handling is built into the editorial system from the start.

Per VEGS § 11, Varia's public commitment is human-in-the-loop curation with published algorithmic constraints: rule-based automation applies human-authored thresholds and source policy at scale; residual human steps are documented, quantified, and auditable; cases the rules cannot resolve surface to a public override surface.

Varia does not claim that grading removes all human judgment, that there is no per-variant human adjudication on the override surface, that grading is purely objective, that a clinical geneticist reviews the catalog, or that grading has zero subjective inputs. Meta-layer decisions (thresholds, allowlist, override rules, voice-anchor criteria) remain human-authored and logged. See the Methodology page for quantified residual contribution and seven meta-editorial governance touchpoints.

Varia uses published rules to grade variants consistently, logs measured human editorial work, and surfaces override cases publicly. Varia does not claim the software removes all judgment, that no human ever adjudicates an override, that every grade is purely objective, or that a clinical geneticist personally reviewed your variant.

The Methodology page states plainly what Varia does and does not claim about editorial discipline.

Temporary flags

Some citations can appear in a provisional state while a verification pass is still closing. That status is visible rather than hidden. The point of the flag is to make unfinished verification legible to the reader and auditable to the editor.

Closeout path

Provisional citations are expected to close through the same verification stack that feeds the editorial grader and the event log. A provisional citation is therefore a temporary publication state, not a second-class evidence tier.

Temporary flags

Some citations can appear with a provisional flag while a verification pass is still being closed. That flag stays visible so unfinished verification is obvious to the reader and easy to audit later.

Closeout path

Provisional citations are expected to close through the same verification stack that feeds the editorial grader and the event log. In other words, provisional is a temporary workflow state, not a shadow evidence tier.

What readers can inspect

Readers can inspect the named source tiers, the named institutional authorities, the named detector cadence, the public event log, the human-in-the-loop grading code path, quantified residual contribution counts, and the markdown drafts that eventually port into the public HTML pages. Those are the mechanisms behind Varia's trust claim.

What this page is not

This page is not a curation roadmap, not a list of future variants, not a substitute for the Methodology page, and not a substitute for the Sources page. It is the compact statement of how Varia decides what counts as citeable, what triggers review, and what obligations the project makes in public.

What readers can inspect

Readers can inspect the named source tiers, the named institutional authorities, the detector cadence, the public event log, the human-in-the-loop grading code path, measured human editorial counts, and the markdown drafts that later turn into HTML pages. Those are the visible parts of Varia's trust promise.

What this page is not

This page is not a future-content roadmap, not a list of planned variants, not a replacement for the Methodology page, and not a replacement for the Sources page. It is the short statement of how Varia decides what can be cited, what triggers review, and what the project commits to in public.