gwas_effects.csv

Identity

Row model	GwasEffectRow (just_dna_format.gwas)
Becomes	gwas_effects.parquet
Authored or derived	derived — an enricher pass writes it; an author corrects it via overrides.csv
Draftable	no
Written and checked by	just-dna-enricher gwas (gwas) from gwas_catalog
Natural key	association_id
Fact signature	yes — its sidecar carries one
In the attestation binding	no

Columns

Column	Type	Required	Values	Meaning
association_id	str	required		The Catalog's own association id. Identity, the same source-accession pattern as PgsRow.pgs_id and PharmVariantRow.annotation_id — one variant carries dozens of associations and only the archive's id tells them apart.
variant_key	str	required		Coordinate-derived identity of the locus (matches the post-expansion weights key)
rsid	str | None	optional		dbSNP identifier, as the Catalog itself names it. Position-level, so it names the locus rather than the allele — which is effect_allele's job.
effect_allele	str | None	optional		The allele effect_size is stated relative to, parsed from the Catalog's riskAlleleName (rs4149056-C → C). Null when the source wrote -?, which it does often — 2 of the first 3 rs4149056 associations — meaning the study did not establish which allele carries the effect. An effect size relative to an unknown allele cannot be used as a weight, so the row is kept and counted rather than dropped: that is a fact a consumer needs, not a defect to hide.
effect_size	float | None	optional		The reported magnitude. Its meaning is effect_measure; its scale is effect_unit.
effect_measure	str | None	optional	suggested: HR, NR, OR, RR, beta, log(HR), log(OR)	What kind of magnitude it is — OR or beta in practice. The Catalog keeps these in two mutually exclusive fields (orPerCopyNum, betaNum) rather than the single ambiguous 'OR or BETA' column its download TSV uses, so this maps cleanly. Open, matching StudyRow.effect_measure.
effect_unit	str | None	optional		The unit a beta is in, verbatim from the source — 'umol/l', 'unit', 'cm'. Kept verbatim including the uninformative ones: 'unit' really is all many studies report, and recording that honestly is the difference between a scale a consumer can reason about and one it cannot. Null for an OR, which is dimensionless.
effect_direction	str | None	optional	one of: decrease, increase	Whether the effect allele increases or decreases the measured trait. Not clinical direction — VariantRow.direction is protective|risk|neutral|unknown and answers a different question, since increasing a trait may be good, bad or neither. Folding the two together would overload one field with two axes (Principle 5); the source keeps them apart (betaDirection) and so does this.
standard_error	float | None	optional		Standard error of the effect size, when the study reported one.
confidence_interval	str | None	optional		The reported 95% CI, verbatim — '[0.14-0.17]', and also '[NR]' where the study reported none. A string because that is what the source publishes and the bracket forms vary; parsing it into two floats here would discard the cases that do not fit.
risk_allele_frequency	float | None	optional		Frequency of the effect allele in the study's sample. The source writes 'NR' for not reported, which is an ABSENCE and arrives here as null — distinct from a reported 0.
p_value	str | None	optional		Raw p-value string, verbatim, mirroring StudyRow.p_value.
p_value_num	float | None	optional		The queryable p-value, mirroring StudyRow.p_value_num. The Catalog's own mantissa/exponent pair is deliberately not carried, and the recorded 0.5 rejection has two halves of which only one still applies: 'a cost every author pays' does not transfer to a machine-written table, but 'a catalogue-of-millions problem rather than a curated-module one' does, and a module cites tens of associations, not millions.
trait	str | None	optional		The reported trait in the Catalog's words. Descriptive — it is re-worded between releases for an unchanged ontology term, so trait_efo_id is the fact and this is what makes the row readable.
trait_efo_id	str | None	optional		EFO trait id(s) — the join back to VariantRow.trait_efo_id.
pmid	str | None	optional		PubMed id of the publication, free-form under the same grammar as StudyRow.pmid.
study_accession	str | None	optional		The Catalog's study accession, e.g. 'GCST001234'.
ancestry	str | None	optional		The study population, free text as the Catalog records it ('European', 'East Asian', 'Hispanic or Latin American'). Free rather than a vocabulary because it is prose there and closing it would make a future release unloadable; and NOT merged with PgsRow.training_ancestry, which is a 1000G superpopulation code list — vocab.py forbids collapsing two ancestry vocabularies into one.
dataset	str	required		Which Catalog release this row came from, e.g. 'gwas_catalog_2026-08-01'. A FACT: the Catalog re-curates, and an association whose effect size moved is a different fact.
source	str | None	optional		The licensed data source: gwas_catalog|manual|reversed (open). Joins sources.csv.source. Names the SOURCE, never the route.
status	str | None	optional	one of: ambiguous, not_found, resolved	Outcome: resolved|not_found|ambiguous. not_found is a FACT — the Catalog was consulted and reports no association for this variant — and differs from a variant never queried, which has no row at all.
fetched_at	str | None	optional		ISO-8601 UTC timestamp, second resolution. Canonicalized on load; records when this row was written by a pass, not when the Catalog published anything — that is dataset.

Generated from the row model at build time — reference.authoring_reference(), the same answer describe_table gives an authoring tool. Nothing on this page is hand-kept.