pgs.csv

One row is one curated PGS Catalog entry — the accession and the trait, not the weights. The scoring file itself is deliberately not authored here: a polygenic score's variant weights are a data file, and carrying them as an authored table is tracked separately rather than assumed.

This table reaches no gate, and research_tier is not a licence axis. Its two members are a calibration judgement — research_only pins as data that a score yields a within-reference Z/percentile, never an ancestry-calibrated absolute risk, and calibrated says the opposite — so research_only is about what the number means, not about who may use it. Nothing in pgs.csv changes what compiles. The paragraph here said the reverse until 2026-09-13, and an author who had read it believed the accession carried its restriction to the gate by itself, so an empty licence ledger read as nothing restrictive here rather than as nobody declared anything (S101).

What the table does owe is a licence row. The Catalog publishes license per score record and it varies, so a score whose record restricts use needs its own row in licensing.csv — and that row is what the gate reads, keyed on that file and nothing else.

Identity

Row model	PgsRow (just_dna_format.pgs)
Becomes	pgs.parquet — lead parquet: pgs.parquet
Authored or derived	authored — a person writes it (a drafter may append rows)
Draftable	yes — draft can append rows
Drafted by	no drafting provider targets this table — draft writes no row of it
Natural key	pgs_id, trait_efo_id
Fact signature	no
In the attestation binding	yes — manifest.inputs[]

Columns

Column	Type	Required	Values	Meaning
pgs_id	str	required		PGS Catalog id, e.g. PGS000135
trait_efo_id	str | None	optional		EFO/MONDO/OBA/HP trait ontology id(s) — joins with variant modules
note	str | None	optional		Free-text note
group	str | None	optional		Grouping label within the module
training_ancestry	list[str] | None	optional	one of: AFR, AMR, EAS, EUR, SAS, multi	Superpopulation(s) the score was developed and evaluated in: the PGS Catalog's development and evaluation samples, not the discovery GWAS behind its weights (1000G superpop codes; multi-valued)
training_cohort	str | None	optional		Optional free-form sub-superpop cohort, e.g. 'FIN', 'Ashkenazi', 'UK Biobank NW-EUR'
match_rate_floor	float | None	optional		Author-set variant-match floor in [0,1]; a score computed below it is invalid. Only the floor lives in-module — the observed per-sample match rate is a measurement (consumer-side).
research_tier	str | None	optional	one of: calibrated, research_only	Calibration frame, not a licence term (VALID_RESEARCH_TIERS): research_only is a within-reference Z/percentile, calibrated is ancestry-calibrated absolute risk. Reaches no compile gate — a score whose Catalog record restricts use needs a licensing.csv row, which is the only thing the gate reads.

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.