Data dictionary

sample_metadata_wide.parquet

column	type	meaning
sample_key	string	ENA/BioSample accession (SAMEA/SAMN/SAMD) — primary key
study_accession	str	BioProject (PRJ…)
secondary_sample	str	ENA secondary sample accession (ERS/SRS/DRS)
sample_title	str	submitter sample title
body_site_class	str	vocabulary (v1.2.1): primary — infant gut/stool/meconium/rectal sample; unknown — no body-site evidence, kept in body-site scope; excluded — non-gut site (skin, oral, milk, vaginal, nasal, blood…) or, since v1.2.1, a non-human host / isolate-genome sample (exclusion_reason_code); linked — mother/other sample linked to an infant subject, outside body-site scope. Body-site scope = {primary, unknown}; headline scope = catalog_scope (age scope ∧ body-site scope)
collection_date	str	archive collection_date as given
is_gold_heldout	bool	True if in the curatedMetagenomicData evaluation set
probiotic_exposure	str	yes
preterm_status	str	preterm
gestational_age_weeks	float64	float weeks at birth
delivery_mode	str	vaginal
feeding_mode	str	exclusive_breast
antibiotic_exposure	str	yes
age_at_collection_days	float64	integer days postnatal
birth_weight_grams	float64	integer grams
country	str	ISO-3166 alpha-2
maternal_antibiotics	str	yes
hmo_supplementation	str	yes
nec_status	str	yes
health_condition	str	healthy_control
multiple_birth	str	singleton
sibling_in_study	str	yes
geo_subregion	str	'City, Region' free text as given by submitter (never reverse-geocoded)
sex	str	male
timepoint_label	str	study's own label (free text)
subject_id	str	study's own subject identifier (free text)
probiotic_exposure__confidence	float	confidence (0–1) of probiotic_exposure
preterm_status__confidence	float	confidence (0–1) of preterm_status
gestational_age_weeks__confidence	float	confidence (0–1) of gestational_age_weeks
delivery_mode__confidence	float	confidence (0–1) of delivery_mode
feeding_mode__confidence	float	confidence (0–1) of feeding_mode
antibiotic_exposure__confidence	float	confidence (0–1) of antibiotic_exposure
age_at_collection_days__confidence	float	confidence (0–1) of age_at_collection_days
birth_weight_grams__confidence	float	confidence (0–1) of birth_weight_grams
country__confidence	float	confidence (0–1) of country
maternal_antibiotics__confidence	float	confidence (0–1) of maternal_antibiotics
hmo_supplementation__confidence	float	confidence (0–1) of hmo_supplementation
nec_status__confidence	float	confidence (0–1) of nec_status
health_condition__confidence	float	confidence (0–1) of health_condition
multiple_birth__confidence	float	confidence (0–1) of multiple_birth
sibling_in_study__confidence	float	confidence (0–1) of sibling_in_study
geo_subregion__confidence	float	confidence (0–1) of geo_subregion
sex__confidence	float	confidence (0–1) of sex
timepoint_label__confidence	float	confidence (0–1) of timepoint_label
subject_id__confidence	float	confidence (0–1) of subject_id
probiotic_exposure__route	str	evidence route of probiotic_exposure (R1–R4)
preterm_status__route	str	evidence route of preterm_status (R1–R4)
gestational_age_weeks__route	str	evidence route of gestational_age_weeks (R1–R4)
delivery_mode__route	str	evidence route of delivery_mode (R1–R4)
feeding_mode__route	str	evidence route of feeding_mode (R1–R4)
antibiotic_exposure__route	str	evidence route of antibiotic_exposure (R1–R4)
age_at_collection_days__route	str	evidence route of age_at_collection_days (R1–R4)
birth_weight_grams__route	str	evidence route of birth_weight_grams (R1–R4)
country__route	str	evidence route of country (R1–R4)
maternal_antibiotics__route	str	evidence route of maternal_antibiotics (R1–R4)
hmo_supplementation__route	str	evidence route of hmo_supplementation (R1–R4)
nec_status__route	str	evidence route of nec_status (R1–R4)
health_condition__route	str	evidence route of health_condition (R1–R4)
multiple_birth__route	str	evidence route of multiple_birth (R1–R4)
sibling_in_study__route	str	evidence route of sibling_in_study (R1–R4)
geo_subregion__route	str	evidence route of geo_subregion (R1–R4)
sex__route	str	evidence route of sex (R1–R4)
timepoint_label__route	str	evidence route of timepoint_label (R1–R4)
subject_id__route	str	evidence route of subject_id (R1–R4)
subject_key	str	resolved subject id (study-scoped)
role	str	infant · mother · other · unknown · adult (age ≥ 18 y) · child (1,101 d – 18 y). infant only when evidenced: an attribute/sample-name role, or age_scope ∈ {infant_evidenced, study_all_infant}; unevidenced body-site defaults are unknown (v1.2)
role_source	str	what evidences role: attr_role, sample_name_token, body_site_raw, body_site_class=linked, age_scope:<value>, none
t_index	float64	ordinal timepoint within subject
n_timepoints_subject	float64	timepoints for this subject
linked_infant_subject_key	str	for mother samples: the infant's subject_key
run_accessions	str	';'-joined run accessions
n_runs	int64	runs for this sample
instrument_model	str	instrument model(s)
library_layout	str	PAIRED/SINGLE
read_count_total	int64	sum of read_count over runs
study_title	str	BioProject title
cohort_id	str	cohort cluster id (COH…)
cohort_name	str	cohort name
first_public_min	str	earliest run release date in study
adult_age_flag	bool	Rule (v1.2): True iff age_at_collection_days > 1,100 — i.e. the sample carries a committed age_at_collection_days determination whose parse_note starts with out_of_scope_adult (an archive age attribute, R1, that the validator's in-scope check validate_age rejected as age N days outside 0-1100). The flag is therefore always evidence-linked; the 12 v1.1 samples that carried both an in-scope and an out-of-range age were resolved in favour of the archive attribute (reasons in value_history.parquet, change_stage = auditor_review:B9)
n_fields_with_value	int64	count of populated metadata fields
ena_sample_url	str	ENA browser link
ena_study_url	str	ENA browser link

sample_determinations.parquet

column	meaning
sample_key, study_accession, field_name	key
field_value	raw value as found in the source
value_normalized	value in the controlled vocabulary/unit
confidence	0–1
evidence_source	label: sample.attr.<key>, run.library_name, sample_id_pattern, paper.supp.table, paper.fulltext.methods, paper.abstract, study.description …
evidence_locator	attribute key / pmcid:file:sheet:column / section
evidence_quote	≤12-word verbatim quote
evidence_limited_to_abstract	1 when the paper was paywalled
determined_by	parser or model id
route	R1–R4
scope	sample
parse_note	derivation notes (unit resolution, propagation, supersession, review flags)
group_audit	keep/downgrade for group statements (Opus audit)
src_track	pipeline pass that produced the row
release_added	release in which this row first became visible (R<YYYY>.<n> or a pre-numbered package semver 1.0.0–1.2.2); pre-1.3.0 values are reconstructed — see DATA_DICTIONARY 'Release columns'
release_retired	release that replaced/removed the row; null (empty in CSV) = current row
package_added	semver of the data package in which the row first appeared

Routes

R1 = archive sample attribute or sample-name convention (per sample); R2 = supplementary table row (per sample; 'subject_level_join' in parse_note when copied from a per-subject row); R3 = statement in the paper text applied to a defined group; R4 = abstract/ENA description statement (group; confidence ≤0.5).

Field vocabularies

field	values
delivery_mode	vaginal
feeding_mode	exclusive_breast
preterm_status	preterm
antibiotic_exposure	yes
maternal_antibiotics	yes
probiotic_exposure	yes
hmo_supplementation	yes
nec_status	yes
age_at_collection_days	integer days postnatal
gestational_age_weeks	float weeks at birth
birth_weight_grams	integer grams
country	ISO-3166 alpha-2
sex	male
timepoint_label	study's own label (free text)
subject_id	study's own subject identifier (free text)
health_condition	healthy_control
multiple_birth	singleton
sibling_in_study	yes
geo_subregion	'City, Region' free text as given by submitter (never reverse-geocoded)

study_metadata_wide.parquet

column	meaning
study_accession, study_title, n_samples, n_runs, first_public_min	archive identity and size
catalog_status, triage_verdict, decision_stage, confidence, evidence	INFANT-catalog inclusion verdict and its evidence (historical; not all-age catalog membership — use gut_studies.parquet)
universe_slice	enumeration channel that found the study
recov_age, recov_delivery, …	recoverability tier predicted before extraction (R1 archive / R2 table / R3 paper / R4 abstract / R0 none)
cov_<field>	fraction of the study's infant-scope samples with a value
cohort_id, cohort_name, n_linked_papers, linked_pmids	cohort and literature
n_unique_infants_est	distinct resolved subject_key among role = infant samples — populated only when n_unique_infants_source ∈ {subject_ids, subject_ids_partial}; NaN otherwise (v1.2; v1.1 wrote 0 or n = n_samples sentinels)
n_unique_infants_source	subject_ids (every infant sample has a resolved subject and subjects repeat), subject_ids_partial (subjects resolved on part of the infant samples), subject_ids_all_distinct (one subject id per sample — cannot distinguish cross-sectional from unresolved; render "≤ n"), no_subject_evidence (render "unknown (≤ n)"), no_infant_role_samples
n_unique_infants_upper, n_unique_infants_display	upper bound (number of infant-role samples) and the string the site should render
n_infant_role_samples, n_infant_samples_with_subject	inputs to the above
n_infant_evidenced, n_study_all_infant, n_age_unknown_mixed_study, n_age_unknown_no_study_estimate, n_adult_flagged, n_non_infant_role, n_age_scope_infant	per-study age_scope counts; n_age_scope_infant = infant_evidenced + study_all_infant
mixed_age_deposit	True when more than half of the study's sample rows are age_unknown_mixed_study or adult_flagged — render a banner
n_samples, n_runs, n_biosamples	from runs.parquet: runs of the study; distinct BioSamples; n_samples = runs for run-unit studies, else BioSamples. n_sample_rows = rows of the sample table keyed to this study (differs when a BioSample carries runs from two BioProjects; n_biosamples_shared_with_other_study)
longitudinal, max_timepoints, n_mothers	subject resolution (infant-role samples)
worklist_rank, effort_class	curation priority
ena_url, ncbi_url	links

Sample unit (added after the auditor's run-vs-BioSample finding)

column	meaning
biosample_accession	str
run_accession	str
age_scope	str
age_scope_basis	str
sample_unit	biosample (default: one BioSample = one stool), run (child row: the submitter registered one BioSample per infant and one run per stool — 6 studies, 521 rows), biosample_pooled rows were moved to parent_biosamples.parquet in v1.2 — the sample table contains only biosample and run units
parent_biosample	for run rows, the pooled BioSample they belong to
studies.n_samples	catalog sample rows (biosample + run units); n_biosamples keeps the archive's BioSample count (e.g. PRJNA294605: 158 samples, 11 BioSamples)

Class-A decision rule (deterministic, sample_unit_classification.csv): BioSample text says 'Assembly of N … samples'/'pooled', or a run-keyed supplementary table gives ≥2 distinct ages/timepoints across the BioSample's runs, or NCBI BioSample carries a collection-date range. Technical multi-run BioSamples (lanes, replicates, paired/unpaired, long+short reads; 33 studies) keep the BioSample as unit.

Confidence (v1.2 clarification)

confidence (and every <field>__confidence) is an ordinal reliability tier assigned by the route and engine that produced the value, not a calibrated probability. Deterministic R1 parsers emit 0.90 (title parser 0.70, unit resolution 0.75–0.85, date arithmetic 0.85, subject propagation 0.70–0.85); R2 table extraction emits 0.65–0.85 (0.75 typical, 0.85 when a paper's own table maps by exact run/sample accession); R3 paper-text group statements 0.25–0.80 (Opus-downgraded statements are capped at 0.40); R4 abstract statements 0.40–0.50. The full tier list is confidence_tiers.csv (route × determiner × scope, with the discrete values emitted). The empirical precision of each tier against the curatedMetagenomicData gold join is in tier_field_precision.csv (gold_set = hires excludes gold rows whose age is an integer number of years; all is the full 3,670-sample join; value mapping: cMD c-section subtypes → c_section, No Breastfeeding/Exclusively Formula Feeding → formula; age tolerance max(7 d, 10 %); GA ± 1 wk). On that join the tiers are monotone for age (R1 0.90 → 1.00, R1 0.85 date/propagation → 0.70, R2 0.85 → 1.00, R2 0.75 → 0.98) but not for feeding (R2 0.75 → 0.48, R2 0.85 → 0.94), and no gold exists for R4 or for exposure fields — treat the number as a tier label and read the precision table for the field you use.

parent_biosamples.parquet

The 16 BioSamples that the submitters of 6 deposits registered one-per-infant with one run per stool. Their runs are the catalog's run sample units (n_child_run_units, child_run_units). Same columns as the sample table plus n_determinations_moved (their 97 v1.1 determinations are in value_history.parquet, status moved_to_parent_biosamples). Excluded from every count.

study_verdict_history.parquet

One row per (study, stage table, replicate). stage_order (v1.2.1) follows the stage ontology stage_rank/stage_level: screen < tier1_rubric < confirm r1–r3 < adjudication < growth_wave (wave order ff, gap, grow, grow_w2, grow_w3, gsa, loose, nm, r3, v3_delta; r1 < r2 < wave adjudication) < growth_wave_merged < biosample_rejudge < human_review (round 1, round 2) < consolidated_triage_v2 < consolidated_final; ties broken by date, replicate, stage. is_consolidated marks rows copied from a consolidated table (consolidated_kind = final / triage_v2 / growth_wave) — render only the final row as the single 'final' line. verdict_norm (v1.2.1): Haiku screen yes → uncertain (screen-pass, not a verdict), no → exclude, ERROR and empty verdicts → null; verdict_norm_note explains each mapping. stage_family groups tables (haiku_screen, sonnet_rubric_tier1, sonnet_confirm:, opus_adjudication:, growth_triage_merged:*, biosample_rejudge, human_review_round1/2, triage_v2_consolidated, catalog_final). verdict is the raw label of that stage (include/exclude/unsure/uncertain; the Haiku screen's yes/no; ERROR for failed requests); verdict_norm maps it to include/exclude/uncertain/None. is_final marks the row that equals universe_studies_all. evidence is the stage's JSON evidence list; src_artifact/src_version_id name the release-bundle table the row was copied from.

value_history.parquet

Determinations that are not in sample_determinations.parquet. status: superseded (a higher-precedence value replaced it; replaced_by names it), rejected (validator or rule rejection; reason), dropped (group statement failed the Opus checklist audit; reason = drop_reason), recommitted_out_of_scope_adult (an R1 age the v1.1 validator rejected as out of range that v1.2 commits with parse_note out_of_scope_adult to evidence adult_age_flag), not_committed_duplicate (second out-of-range age attribute on the same sample), moved_to_parent_biosamples, auditor_finding_applied (study-level auditor findings and what was done). change_stage names the pipeline stage that made the change (auditor_review:B2/B9 for v1.2.0; auditor_review:R3-3 for the v1.2.1 host/isolate rule, statuses recommitted_out_of_scope_host and recommitted_out_of_scope_isolate, field_name body_site_class, evidence source run.scientific_name / run.library_source).

Release columns (R2026.1, package 1.3.0)

Fact tables — sample_determinations.parquet, universe_studies_all.parquet, study_metadata_wide.parquet, cohorts.csv, study_paper_links.csv, sandpiper_sample_summary.parquet, sandpiper_run_qc.parquet, sandpiper_top_genera.parquet, sandpiper_study_panels.parquet, study_metadata_wide.csv — end with three columns:

column	dtype	meaning
release_added	string	release in which this row first became visible (R<YYYY>.<n> or a pre-numbered package semver 1.0.0–1.2.2); pre-1.3.0 values are reconstructed — see DATA_DICTIONARY 'Release columns'
release_retired	string | null	release that replaced/removed the row; null (empty in CSV) = current row
package_added	string	semver of the data package in which the row first appeared

Release ids are R<YYYY>.<n> (first: R2026.1 = package 1.3.0) or, for states published before numbering, the package semver 1.0.0 (first public package, 2026-09-25), 1.1.0 (release v11), 1.2.0, 1.2.1, 1.2.2. "Current" = release_retired IS NULL; in the shipped fact tables every row is current. sample_metadata_wide is derived from sample_determinations and carries no release columns.

Reconstruction caveat. Packages before 1.3.0 carried no release columns, so release_added of pre-existing rows was reconstructed: newest value_history.change_stage touching the (sample_key, field_name) or sample_determinations.src_track, mapped to the package the CHANGELOG records for that stage (sample_unit_fix → 1.1.0, auditor_review:B2/B9 → 1.2.0, auditor_review:R3-3 → 1.2.1, owner_decision → 1.2.2). Exact for rows those stages touched (13191 current determinations: src_track = sample_unit_fix → 1.1.0, src_track = adult_scope_fix → 1.2.0); every other pre-1.3.0 row is labelled 1.0.0 by assumption (the first package is not archived as a table, so this cannot be verified row by row). package_added equals release_added for all pre-1.3.0 rows. Study verdict rows are all 1.0.0 (no verdict change is dated after the first package); Sandpiper rows are 1.2.0 (when Sandpiper landed; the v1.2.1 recomputation of existing rows is recorded in the CHANGELOG, not as retire+add).

sample_determinations_all.parquet (R2026.1)

sample_determinations columns + the three release columns + retired_reason (string|null; value_history.reason) + retired_change_stage (string|null; value_history.change_stage). Rows: every current determination (618,898, release_retired null) ∪ every retired published value (1,316) reconstructed from value_history rows whose status ∈ {superseded, moved_to_parent_biosamples, recommitted_out_of_scope_host, recommitted_out_of_scope_isolate} (release_retired = the package of their change_stage). value_history rows with status rejected, dropped, not_committed_duplicate, recommitted_out_of_scope_adult (the pre-history of a row that IS current) or auditor_finding_applied were never published and are not rows here. Exactly one current row per (sample_key, field_name); retired rows may repeat a key. sample_determinations_superseded.parquet (223 rows) is the same set as the superseded rows of stages sample_unit_fix (75) and auditor_review:B9 (148) and is kept for the v1.1 layout.

releases.csv (R2026.1)

One row per release id, oldest first: release_id, package_version, release_date, data_tag, site_tag, doi (empty until Zenodo mints one), n_studies_included, n_samples, n_catalog_scope, n_determinations_current, sandpiper_version, notes_file. Historical rows carry only the counts the CHANGELOGs state (empty otherwise — never invented).

Contribute worklist (R2026.2, package 1.4.0)

contribute_worklist.csv — one row per open study (universe_studies_all.triage_verdict ∈ {include, uncertain} and at least one of the six worklist fields below 0.5 coverage on its catalog_scope samples; studies with 0 catalog_scope samples are judged on body-site scope body_site_class ∈ {primary, unknown}). Complete studies are not listed. Rules, templates and vocabularies: the pipeline's config/contribute.yaml and docs/CONTRIBUTE.md. Read-only: nothing here is a curated value — every column is derived from the tables of this package and from the recoverability / R2-rescue / supplement-inventory artifacts registered in config/inputs.json (group contribute).

contribute_worklist.csv (439 rows)

column	dtype	meaning
rank	int64	1 = highest priority_score (ties: n_samples desc, accession)
study_accession	string	BioProject accession (universe_studies_all key)
study_title	string	universe_studies_all.study_title
cohort_id	string	cohorts.csv id
cohort_name	string	cohorts.csv name
triage_verdict	string	include
n_samples	int64	sample units of the study (universe_studies_all.n_samples)
n_catalog_scope	int64	sample_metadata_wide rows with catalog_scope = True
n_infant_samples_est	float64	null
missing_fields	string	';'-joined short field names (age, delivery, feeding, preterm, antibiotics, probiotic) whose coverage < 0.5
n_missing_fields	int64	count of missing_fields (≥ 1 for every listed study)
coverage_<field>	float64	fraction of catalog_scope samples with a value for the field (body-site scope {primary, unknown} when n_catalog_scope = 0); six columns: age, delivery, feeding, preterm, antibiotics, probiotic
best_tier_<field>	string	best recoverability tier for study × field (R1 archive attribute · R2 supplementary table · R3 paper text · R4 abstract · R0 none); six columns
blocker_code	string	dominant reason the study is open — vocabulary in config/contribute.yaml blocker_codes (controlled_access, no_linked_paper, paywalled_abstract_only, tables_unjoinable_need_key, pdf_only_supplement, no_supplement_found, archive_only_uncertain, unitless_age_needs_curator, partial_coverage); decision order blocker_order
blocker_detail	string ≤ 200	the deciding evidence (counts, access tier, sample-ID forms from RESCUE_REPORT_v2, human-review note)
unlock_text	string ≤ 200	imperative 'what would unlock this' sentence from the unlock_templates of the blocker (id form / missing fields substituted)
contribution_type	string	primary ask: per_sample_table
n_linked_papers	int64	rows in study_paper_links for the study
own_data_pmids	string	';'-joined PMIDs of the linked papers (empty when none)
n_supp_tables_inventoried	int64	supp_inventory members with member_type = table across the linked papers
controlled_access	bool	study or its cohort is in controlled_access_registry.csv (non-open tier) or flagged controlled in extraction_worklist / universe_studies_all
priority_score	float64	Σ over missing fields of weight × (1 − coverage) × log10(n_catalog_scope + 1); weights age 3, delivery 2, feeding 2, preterm 1.5, antibiotics 1, probiotic 0.5
ena_url	string	ENA browser URL of the study
ncbi_url	string	NCBI BioProject URL
issue_url	string	prefilled GitHub Issue (form catalog-contribution.yml, label contribution; query keys study_accession, contribution_type, release_tag, title)
release_added	string	release in which this row first became visible (R<YYYY>.<n> or a pre-numbered package semver 1.0.0–1.2.2); pre-1.3.0 values are reconstructed — see DATA_DICTIONARY 'Release columns'
release_retired	string	release that replaced/removed the row; null (empty in CSV) = current row
package_added	string	semver of the data package in which the row first appeared

contribute_worklist_fields.csv (2,634 rows = 6 per worklist study)

column	dtype	meaning
study_accession	string	BioProject accession
field	string	age
coverage	float64	fraction of in-scope samples with a value (same scope rule as the worklist)
n_with_value	int64	in-scope samples with a value
n_catalog_scope	int64	catalog_scope samples of the study
best_tier	string	best recoverability tier R0–R4 for the study × field
blocker_code	string	complete when coverage ≥ 0.5, else the study blocker (only the age field carries unitless_age_needs_curator; other fields then partial_coverage)
evidence	string ≤ 120	tier + first recoverability evidence quote/source + note
release_added	string	release in which this row first became visible (R<YYYY>.<n> or a pre-numbered package semver 1.0.0–1.2.2); pre-1.3.0 values are reconstructed — see DATA_DICTIONARY 'Release columns'
release_retired	string	release that replaced/removed the row; null (empty in CSV) = current row
package_added	string	semver of the data package in which the row first appeared

Blocker codes (one per study, decision order = listing order for included studies; uncertain studies always archive_only_uncertain): controlled_access (study/cohort in the controlled-access registry) · no_linked_paper (0 rows in study_paper_links) · paywalled_abstract_only (extraction_worklist access_tier C/D/E) · tables_unjoinable_need_key (R2 rescue: supplementary tables keyed by paper-internal names) · pdf_only_supplement (only PDF/DOCX supplements) · no_supplement_found (paper but no inventoried supplement) · unitless_age_needs_curator (age column without unit) · partial_coverage (paper + tables processed, fields still missing). Field-level code complete = field at/above threshold. Contribution types: per_sample_table, id_key, paper_pointer, age_schedule, verdict_evidence (dropdown of the GitHub Issue form catalog-contribution.yml that issue_url opens prefilled). Rows get release_retired when the study becomes complete in a later release.

Registry tier (R2026.4, package 1.6.0)

The registry is the outer tier of the catalog (docs/EXPANSION.md): every ENA study with a human shotgun-metagenome signal, any body site, any age, classified at STUDY level from ENA study/sample/run metadata (no paper reading). Enumeration slices: S1 library_source=METAGENOMIC × WGS|WXS; S2 misfiled GENOMIC on verified human-metagenome taxa; S3 OTHER|Targeted-Capture|WGA adjudication (per-slice completeness in registry_universe_audit.csv). Classification stages: deterministic_prior (infant-universe verdict carried over), deterministic_rule, sonnet_x2 (two rubric replicates agree), opus_adjudicated (replicates disagreed → adjudication), pending. Vocabularies: the pipeline's config/vocab/{body_sites,life_stages,assay,population_flags}.yaml; scope rules: config/scope.yaml. The curated infant catalog is the scope infant_gut inside this registry (in_infant_catalog mirrors universe_studies_all.triage_verdict). The run-level table registry_runs.parquet (all runs of the universe, 46 ENA fields + found_by) is attached to the GitHub Release of the data repository as registry_runs_v<version>.parquet (too large for this package). Bitemporal columns follow the package convention (release_added, release_retired, package_added).

registry_studies.parquet (54,410 rows)

column	dtype	meaning
study_accession	string	ENA/INSDC BioProject accession (PRJ…); primary key
secondary_study_accession	string	ENA secondary study accession (ERP/SRP/DRP)
study_title	string	ENA study title
description_short	string	ENA study description truncated to 300 characters
center_name	string	Submitting centre
first_public_min	string	Earliest run first_public date (ISO)
first_public_max	string	Latest run first_public date (ISO)
n_runs	int64	Runs in the registry universe for this study
n_samples	int64	Distinct sample_accession values
n_biosamples	int64	Distinct BioSample accessions
library_strategies	string	JSON dict strategy → run count
library_sources	string	';'-joined distinct library_source values
instrument_platforms	string	';'-joined distinct instrument_platform values
scientific_names_top	string	';'-joined top-5 scientific_name values with run counts, e.g. 'human gut metagenome (120)'
host_tax_ids	string	';'-joined distinct host_tax_id values
n_runs_host_9606	int64	Runs with host_tax_id 9606
n_runs_nonhuman_host	int64	Runs with a non-9606 host_tax_id
human_signal	bool	Frame-free sweep human-signal rule outcome (pre-filter)
human_signal_rule	string	A
ambiguous	bool	Human-signal rule tie flag
host_human	string	Classified host
host_evidence	string	JSON list of {source, quote} (quote ≤ 12 words; source from the curation-skill label list)
assay	string	Assay class (config/vocab/assay.yaml)
access	string	Access tier
body_sites	string	';'-joined body-site codes (config/vocab/body_sites.yaml)
body_site_primary	string	Primary body-site code (multi_site when ≥ 3 strong sites)
body_site_evidence	string	JSON evidence list
life_stages	string	';'-joined life-stage codes (config/vocab/life_stages.yaml)
life_stage_primary	string	Primary life-stage code (mixed_ages when ≥ 3 strong stages)
life_stage_evidence	string	JSON evidence list
population_flags	string	';'-joined flags (config/vocab/population_flags.yaml)
health_context	string	≤ 120 chars free text or null
classification_stage	string	Cascade stage that produced the classification
classification_confidence	float64	0–1; minimum over components
classification_model	string	'deterministic' or the resolved model id of the deciding stage
in_infant_catalog	string	Infant-catalog verdict
infant_reason_code	string	Infant triage reason_code (curation-skill vocabulary) or null
scope_memberships	string	';'-joined scope ids from config/scope.yaml
universe_slice	string	Enumeration slice that found the study
release_added	string	Bitemporal: release id in which the row first appeared
release_retired	string	Bitemporal: release id in which the row was retired; null = current
package_added	string	Bitemporal: package semver of first appearance

registry_universe_audit.csv

column	dtype	meaning
slice	string	Slice tag / ENA query label
ena_query	string	ENA portal query string
ena_count	int64	Count returned by the ENA count endpoint
rows_pulled	int64	read_run rows actually streamed
n_studies	int64	Distinct studies in the slice
completeness	float64	rows_pulled / ena_count
pulled_at	string	ISO timestamp of the pull

registry_biosamples.parquet (611,601 rows) — one row per harvested BioSample of a registry study (S2: human_all studies outside the curated infant catalog) — added R2026.5

column	dtype	meaning
sample_accession	string	BioSample accession (SAMN/SAME/SAMD) as listed in registry_biosample_index; primary key
study_accession	string	ENA study (BioProject) the sample belongs to
n_attributes	int64	Non-placeholder attribute rows harvested for the sample
source	string	ena_xml
body_site_code	string	config/vocab/body_sites.yaml code from the normalised body-site attribute; unknown_site when an attribute existed but mapped to no site; null when no body-site attribute
body_site_raw_key	string	Normalised attribute key the code came from (e.g. host_body_site, isolation_source)
body_site_raw_value	string	Raw attribute value the code came from
life_stage	string	config/vocab/life_stages.yaml code from the age attribute (or from age_days); unknown_age when an age attribute existed but was not interpretable; null when none
age_days	float64	Age at collection in days when a numeric age with unit was given
age_raw_key	string	Attribute key the age came from
age_raw_value	string	Raw age value
sex	string	female
sex_raw_key	string	Attribute key the sex came from
sex_raw_value	string	Raw sex/gender value
country_iso2	string	ISO-3166-1 alpha-2 from geo_loc_name / country attributes
country_raw_key	string	Attribute key the country came from
country_raw_value	string	Raw geographic value
collection_date_raw	string	Raw collection date attribute
collection_year	Int64	Year extracted from collection_date_raw
disease_raw	string	First non-placeholder disease/health/phenotype attribute value (≤200 chars; NOT normalised — no controlled vocabulary yet)
release_added	string	Release id in which the row first appeared
release_retired	string	Release id in which the row was retired (null = current)
package_added	string	Package version that first carried the row

registry_study_papers.parquet (6,398 rows) — one row per study × linked paper (accession mention in Europe PMC and/or publication declared in the NCBI BioProject record); 4,217 human_all studies outside the curated infant catalog — added R2026.5

column	dtype	meaning
study_accession	str	ENA study (BioProject) accession
paper_id	str	Europe PMC id (MED:
pmid	str	PubMed id of the linked paper
pmcid	str	PubMed Central id
doi	str	DOI
title	str	Paper title
journal	str	Journal
year	Int64	Publication year
pub_type	str	Europe PMC publication types
is_open_access	boolean	Europe PMC open-access flag
in_epmc_fulltext	object	Full text available in Europe PMC
source	str	europepmc_mention
match_field	str	abstract_or_title
match_accession	str	Accession matched (primary PRJ… or secondary ERP/SRP/DRP)
bioproject_declared	bool	Paper is listed as a publication in the NCBI BioProject record
n_registry_studies	int64	How many processed registry studies this paper mentions
relation	str	data_paper
method	str	Rule that produced the relation (det_bioproject_declared 0.9, det_single_mention_abstract 0.75, det_single_mention_fulltext 0.6)
confidence	float64	Rule confidence (null for unsure)
contested	bool	True when the paper mentions ≥ 2 processed studies
contest_reason	str	multi_study
epmc_rank	float64	Rank of the paper in the Europe PMC hit list for the study
epmc_hit_count	float64	Europe PMC hitCount for the study query
epmc_hits_capped	object	Hit list truncated at 300 (never true in R2026.5)
body_site_primary	str	registry_studies.body_site_primary at harvest time (convenience)
release_added	string	Release id in which the row first appeared
release_retired	string	Release id in which the row was retired (null = current)
package_added	string	Package version that first carried the row

registry_bioproject_records.parquet (4,217 rows) — one row per processed registry study (NCBI BioProject XML) — added R2026.5

column	dtype	meaning
study_accession	str	ENA study (BioProject) accession
secondary_study_accession	str	ENA secondary study accession
found_in_ncbi	bool	BioProject record found via esearch
bioproject_uid	str	NCBI BioProject UID
archive	str	NCBI
center_id	str	Submitting center id
center_id_center	str	Center attribute of center_id
title	str	Paper title
name	str	BioProject name
organisation	str	Submitting organisation (first)
submitter_owner	str	Submitter/owner organisation
organizations_json	str	JSON list of organisations (name, abbr, role)
n_publications	float64	Publications declared in the record
publication_pmids	str	; -joined PMIDs declared
publication_dois	str	; -joined DOIs declared
publications_json	str	JSON list of declared publications
grants_json	str	JSON list of grants
external_links_json	str	JSON list of external links
registration_date	str	ProjectReleaseDate when present (239 studies)
submitted	str	Submission date
last_update	str	Last update date
submission_access	str	public
data_types	str	Declared data types
target_material	str	BioProject target material (eGenome, …)
target_capture	str	Target capture (eWhole, …)
target_sample_scope	str	Sample scope (eMultiisolate, eEnvironment, …)
method_type	str	Method type (eSequencing, …)
relevance_medical	str	Medical relevance flag
body_site_primary	str	registry_studies.body_site_primary at harvest time (convenience)
release_added	string	Release id in which the row first appeared
release_retired	string	Release id in which the row was retired (null = current)
package_added	string	Package version that first carried the row

registry_authors.parquet (52,110 rows) — one row per study × paper × author (Europe PMC core records of ≤ 5 linked papers per study) — added R2026.5

column	dtype	meaning
study_accession	str	ENA study (BioProject) accession
author_full_name	str	Author full name (Europe PMC)
author_display	str	Surname + initials
author_surname	str	Surname
author_initials	str	Initials
is_group	bool	Collective/group author
source	str	europepmc_mention
pmid	str	PubMed id of the linked paper
pmcid	str	PubMed Central id
doi	str	DOI
paper_title	str	Paper title
pub_year	Int64	Publication year
position	int64	Author position (1 = first); part of the key
n_authors	int64	Authors on the paper
is_first	bool	First author
is_last	bool	Last author
affiliation	str	Affiliation string (e-mail addresses removed, rule F7)
orcid	str	ORCID iD when present
paper_relation	str	relation of the study×paper link (registry_study_papers)
link_method	str	method of the link
author_source	str	europepmc_core
author_key	str	lower-case surname + initials key for de-duplication
release_added	string	Release id in which the row first appeared
release_retired	string	Release id in which the row was retired (null = current)
package_added	string	Package version that first carried the row

gut_sample_determinations.parquet (1,659,128 rows) — one row per sample × field of the gut_all curated scope (current value; precedence infant catalog > R1 > R2 > R3 > R4) — added R2026.7

column	dtype	meaning
sample_key	string	BioSample accession (or run accession for run-unit samples); with field_name the key
field_name	string	Field of config/packs/gut.yaml
study_accession	string	ENA study
field_value	string	Raw value as found in the source
value_normalized	string	Normalised value (days for age, vocabulary codes, ISO-2 country, yes/no)
confidence	float64	Route/parser confidence tier (R1 0.85–0.9, R2 0.8–0.85, R3 ≤ 0.7, R4 ≤ 0.5)
evidence_source	string	Labelled source: sample.attr.
evidence_locator	string	biosample_attr
evidence_quote	string	Verbatim quote ≤ 12 words
evidence_limited_to_abstract	float64	1 when the evidence is abstract/description only (R4)
determined_by	string	Parser / stage id
route	string	R1
scope	string	sample (rows expanded from study_all statements carry 'expanded from study_all' in parse_note)
parse_note	string	Parser notes
group_audit	string	Audit note for group statements (infant rows)
src_track	string	infant_catalog (copied from the curated infant tables)
release_added	string	Release id in which the row first appeared
release_retired	string	Release id in which the row was retired (null = current)
package_added	string	Package version that first carried the row

gut_sample_metadata_wide.parquet (579,252 rows) — one row per sample of the gut_all curated scope (one column per field with its route and confidence) — added R2026.7

column	dtype	meaning
sample_key	string	BioSample accession; primary key
study_accession	string	ENA study
biosample_accession	string	BioSample accession
secondary_sample	string	ERS/SRS/DRS accession (infant rows)
sample_unit	string	biosample
body_site_code	string	Registry body-site code of the sample (from its attributes) or the infant catalog's class
sample_life_stage	string	Registry life stage from the sample's age attribute
curated_source	string	infant_catalog
in_infant_catalog	bool	Study is an included infant-catalog study
age_at_collection_days	float64	Age at collection in days
sex	string	female
bmi	float64	Body-mass index kg/m²
country	string	ISO-3166-1 alpha-2
health_condition	string	config/vocab/health_conditions.yaml code
health_condition_detail	string	Raw health / disease text
antibiotic_exposure	string	yes
subject_id	string	Subject / participant identifier as given by the source
timepoint_label	string	Timepoint / visit label as given
age_at_collection_days__confidence	float64	Confidence of the committed age_at_collection_days value
age_at_collection_days__route	string	Route of the committed age_at_collection_days value
sex__confidence	float64	Confidence of the committed sex value
sex__route	string	Route of the committed sex value
bmi__confidence	float64	Confidence of the committed bmi value
bmi__route	string	Route of the committed bmi value
country__confidence	float64	Confidence of the committed country value
country__route	string	Route of the committed country value
health_condition__confidence	float64	Confidence of the committed health_condition value
health_condition__route	string	Route of the committed health_condition value
health_condition_detail__confidence	float64	Confidence of the committed health_condition_detail value
health_condition_detail__route	string	Route of the committed health_condition_detail value
antibiotic_exposure__confidence	float64	Confidence of the committed antibiotic_exposure value
antibiotic_exposure__route	string	Route of the committed antibiotic_exposure value
subject_id__confidence	float64	Confidence of the committed subject_id value
subject_id__route	string	Route of the committed subject_id value
timepoint_label__confidence	float64	Confidence of the committed timepoint_label value
timepoint_label__route	string	Route of the committed timepoint_label value
delivery_mode	string	Infant-catalog field (only on infant rows)
feeding_mode	string	Infant-catalog field (only on infant rows)
preterm_status	string	Infant-catalog field (only on infant rows)
gestational_age_weeks	string	Infant-catalog field (only on infant rows)
birth_weight_grams	string	Infant-catalog field (only on infant rows)
maternal_antibiotics	string	Infant-catalog field (only on infant rows)
probiotic_exposure	string	Infant-catalog field (only on infant rows)
hmo_supplementation	string	Infant-catalog field (only on infant rows)
nec_status	string	Infant-catalog field (only on infant rows)
age_category	string	neonate
age_category_basis	string	age_at_collection_days
body_site_class	string	primary (gut/stool)
body_site_basis	string	sample_attribute
infant_scope	bool	== the infant catalog's catalog_scope rule (reproduces the infant catalog exactly)
n_fields_with_value	int64	Pack fields with a committed value
release_added	string	Release id
release_retired	string	Release id when retired (null = current)
package_added	string	Package version

gut_studies.parquet (2,837 rows) — one row per study of the gut_all curated scope — registry columns plus curation coverage — added R2026.7

column	dtype	meaning
study_accession	string	ENA study; primary key
secondary_study_accession	string	registry_studies column (see registry_studies.parquet)
study_title	string	registry_studies column (see registry_studies.parquet)
description_short	string	registry_studies column (see registry_studies.parquet)
center_name	string	registry_studies column (see registry_studies.parquet)
first_public_min	string	registry_studies column (see registry_studies.parquet)
first_public_max	string	registry_studies column (see registry_studies.parquet)
n_runs	int64	Runs
n_samples	int64	Samples
n_biosamples	int64	BioSamples
n_runs_sandpiper	int64	Runs with Sandpiper profiles
…registry_studies columns…		all registry_studies columns are carried (classification, evidence, scope memberships, sample roll-ups)
cov_<field>	float64	Share of the study's samples with a value for each pack field
n_samples_curated	int64	Samples in gut_sample_metadata_wide
age_categories	string	JSON dict age_category → samples
health_conditions	string	JSON dict health_condition → samples (top 6)
curated_depth	string	Routes present among the study's committed values (e.g. R1;R2;R4)
curated_source	string	infant_catalog
release_added	string	Release id
release_retired	string	Release id when retired
package_added	string	Package version

gut_runs.parquet (721,678 rows) — one row per sequencing run of a gut_all catalog study (registry_runs ∩ gut_studies) with the catalog sample_key — added R2026.12 / 1.12.0 — added R2026.12

column	dtype	meaning
run_accession	string	ENA/SRA run accession (SRR/ERR/DRR)
study_accession	string	BioProject of the run (a gut_studies row)
sample_accession	string	BioSample accession(s) of the run (';'-joined when the run pools several)
secondary_sample_accession	string	SRS/ERS secondary sample accession(s)
experiment_accession	string	SRX/ERX experiment accession
library_name	string	Submitter library name
library_strategy	string	ENA library_strategy (WGS / WXS / OTHER …)
library_source	string	ENA library_source (METAGENOMIC …)
library_layout	string	SINGLE
instrument_platform	string	ENA instrument_platform
instrument_model	string	ENA instrument_model
read_count	string	ENA read_count (as delivered by the portal)
base_count	string	ENA base_count (as delivered by the portal)
first_public	string	ENA first_public date (YYYY-MM-DD); its year bounds collection_date (R1 parser)
sandpiper_profiled	bool	Run present in the Sandpiper snapshot (registry_runs_sandpiper)
sample_key	string	gut_sample_metadata_wide.sample_key of the run's sample (run accession for run-unit samples; null when the BioSample is not a catalog sample, e.g. pooled runs)

gut_sandpiper_sample_summary.parquet (339,626 rows) — one row per catalog sample with >= 1 run in the Sandpiper 2.0.0 snapshot (339,626 samples / 2,084 studies) — added R2026.12 — added R2026.12

column	dtype	meaning
sample_key	str	catalog sample key (gut_sample_metadata_wide)
study_accession	str	BioProject
n_runs_profiled	int64	runs of the sample with a Sandpiper profile
n_runs_total	int64	runs of the sample in gut_runs
qc_partial	bool	1 = some runs of the sample are unprofiled
root_coverage_sum	float64	summed SingleM root coverage over the profiled runs (depth proxy)
richness_genus	int64	genera with relative abundance >= 0.001
shannon_genus	float64	Shannon index over genus relative abundances (natural log)
shannon_genus_assigned	float64	Shannon over genus-assigned coverage only (unassigned remainder excluded)
top_genus	str	most abundant GTDB genus
top_genus_relabund	float64	its relative abundance (0-1)
unassigned_genus_relabund	float64	share of root coverage not resolved to a genus
unassigned_species_relabund	float64	share of root coverage not resolved to a species
spf	float64	Sandpiper single-copy-marker profile fraction (per_acc_summary)
known_species_fraction	float64	fraction of coverage in known species (per_acc_summary)
qc_flags	str	';'-joined QC flags (qc_low_depth, qc_low_complexity, qc_non_metagenome, qc_partial, ...) — flagged, never dropped
qc_any_flag	bool	1 = at least one QC flag set
qc_low_depth	bool	root coverage below the depth threshold
qc_low_complexity	bool	Sandpiper low_complexity flag
qc_readfraction_warning	bool	Sandpiper read-fraction warning
qc_non_metagenome	bool	organism prediction not a metagenome (loose rule)
qc_non_metagenome_strict	bool	organism prediction not a metagenome (strict rule)
qc_synthetic	bool	Sandpiper synthetic-sample flag
qc_rna	bool	RNA / non-DNA library flag
qc_predicted_ecological	bool	Sandpiper host_or_not prediction = ecological
qc_no_genus_assigned	bool	no coverage resolved to any genus
organism_labels	str	';'-joined Sandpiper organism labels of the runs
runs_profiled	str	';'-joined profiled run accessions
sandpiper_url	str	Sandpiper page of the first profiled run
age_category	str	catalog age category (copied from the wide table)
body_site_class	str	catalog body-site class
infant_scope	bool	infant-extension filter flag
taxonomy_db	str	taxonomy database (GTDB)
taxonomy_version	str	GTDB release (R232)
sandpiper_version	str	Sandpiper version (2.0.0)
zenodo_record	str	Zenodo record of the bulk snapshot (20419175)

gut_sandpiper_pca_scores.parquet (335,956 rows) — one row per scored sample of the genus-level CLR-PCA (335,956 samples x 383 genera; root coverage >= 2) — added R2026.12 — added R2026.12

column	dtype	meaning
sample_key	str	catalog sample key
pc1	float32	PCA score, component 1 (16.3 % variance)
pc2	float32	component 2 (6.5 %)
pc3	float32	component 3 (3.8 %)
pc4	float32	component 4 (3.2 %)
pc5	float32	component 5 (2.3 %)
study_accession	str	BioProject
root_coverage_sum	float64	summed root coverage (depth proxy)
taxonomy_db	str	taxonomy database (GTDB)
taxonomy_version	str	GTDB release (R232)
sandpiper_version	str	Sandpiper version (2.0.0)
zenodo_record	str	Zenodo record of the bulk snapshot (20419175)
pca_method	str	CLR on 383 genera (prevalence >= 1 %), multiplicative pseudocount, randomized SVD

gut_sandpiper_pca_loadings.parquet (383 rows) — one row per genus x component loading of the CLR-PCA — added R2026.12 — added R2026.12

column	dtype	meaning
genus	str	GTDB R232 genus
pc1	float64	loading
pc2	float64	loading
pc3	float64	loading
pc4	float64	loading
pc5	float64	loading
prevalence_ge_0p1pct	float64	share of samples where the genus is >= 0.1 % relative abundance
mean_relabund	float64	mean relative abundance over scored samples
clr_mean	float64	mean CLR value
taxonomy_db	str	taxonomy database (GTDB)
taxonomy_version	str	GTDB release (R232)
sandpiper_version	str	Sandpiper version (2.0.0)

gut_sandpiper_pca_variance.csv (5 rows) — one row per PCA component: explained variance ratio — added R2026.12 — added R2026.12

column	dtype	meaning
pc	str	component index
explained_variance_ratio	float64	share of CLR variance
singular_value	float64	singular value
cumulative	float64	cumulative explained variance ratio
n_samples	int64	samples in the PCA
n_genera	int64	genera in the PCA
root_coverage_min	int64	minimum root coverage for inclusion
prevalence_min	float64	minimum genus prevalence for inclusion
prevalence_detection	float64	relative-abundance threshold used for prevalence
pseudocount	float64	multiplicative pseudocount before CLR
total_clr_variance	float64	total variance of the CLR matrix

gut_sandpiper_study_coverage.csv (2,837 rows) — one row per catalog study: Sandpiper join coverage — added R2026.12 — added R2026.12

column	dtype	meaning
study_accession	str	BioProject
study_title	str	study title
n_runs	int64	runs in gut_runs
n_runs_profiled	int64	runs with a profile
n_samples	int64	samples
n_infant_scope	int64	infant-scope samples
n_samples_profiled	int64	samples with >= 1 profiled run
n_samples_partial	int64	samples with unprofiled runs
n_low_depth	int64	samples flagged low depth
n_low_complexity	int64	samples flagged low complexity
n_any_flag	int64	samples with any QC flag
root_coverage_median	float64	median root coverage
richness_genus_median	float64	median genus richness
shannon_genus_median	float64	median Shannon
frac_runs_profiled	float64	profiled / total runs
frac_samples_profiled	float64	profiled / total samples
coverage_class	str	full
top_genus_mode	str	most frequent top genus
prefix	str	accession prefix (SRR/ERR/DRR)
sandpiper_version	str	Sandpiper version (2.0.0)
taxonomy_db	str	taxonomy database (GTDB)
taxonomy_version	str	GTDB release (R232)
zenodo_record	int64	Zenodo record of the bulk snapshot (20419175)

gut_relatedness.parquet (5,858 rows) — one row per family or shared setting with >= 2 catalogued people — added R2026.23 / 1.19.0 — added R2026.23

column	dtype	meaning
study_accession	str	catalog project
group_kind	str	family
group_type	str	family, or the shared setting type
group_id	str	family_id / shared_setting_id as given in the source
n_people	int64	distinct people (person_key)
n_samples	int64	catalog samples
roles	str	JSON {family_role: n_people}
motif	str	family motif, e.g. mother + child×2
people	str	';'-joined person keys
sample_keys	str	';'-joined sample keys

gut_relations.parquet (1,015 rows) — one row per explicit pairwise relation stated by the authors — added R2026.23 / 1.19.0 — added R2026.23

column	dtype	meaning
study_accession	str	catalog project
subject_a	str	person (subject id or label)
subject_b	str	related person
relation	str	donor_of
evidence_source	str	where the relation is stated
evidence_quote	str	verbatim quote
confidence	float64	0-1
sample_key_a	str	a sample of subject_a when known
notes	str	free text

gut_pass_coverage.csv (2,798 rows) — one row per catalog project: status of every annotation pass — added R2026.25 / 1.20.0 (release check scripts/check_pass_coverage.py) — added R2026.25

column	dtype	meaning
study_accession	string	catalog project
has_pmcid	bool	project has an uncontested linked paper with a PMCID (full-text / supplementary passes apply)
fulltext	string	full-text cohort statements (R3 + R3b): done
abstract	string	abstract / ENA-description statements (R4 + R4b)
interventions	string	study-level intervention classification
deep	string	per-sample deep annotation (archive attributes, sample names, paper tables)
supplementary	string	supplementary / body-table search (R2)
subject	string	subject_id pass (applies when subject_id coverage < 95 %)
design	string	relatedness + longitudinal study-design pass
n_missing	int	passes that apply but were never attempted

gut_metadata_contradictions.parquet (8,853 rows) — one row per withheld sample x field: self-contradictory metadata — added R2026.25 / 1.20.0 (catalog.scopes.contradictions) — added R2026.25

column	dtype	meaning
sample_key	string	catalog sample
study_accession	string	project
field_name	string	field whose value is withheld
kind	string	sources_disagree
value_a	string	value of the published-precedence source (or the sample's own value)
route_a	string	route of value_a
source_a	string	evidence source of value_a
quote_a	string	evidence quote of value_a
value_b	string	incompatible value from the other source
route_b	string	route of value_b
source_b	string	evidence source of value_b
quote_b	string	evidence quote of value_b
handling	string	what was published (normally: neither value)
note	string	e.g. the sex split of the subject

Sandpiper columns (added v1.2.0)

Source: SingleM community profiles from Sandpiper 2.0.0 (Woodcroft et al. 2025, Nat Biotechnol; Zenodo record 20419175, CC-BY), taxonomy GTDB R232. Profiles are keyed by run; for a catalog sample with several profiled runs the filled coverage per taxon is summed across runs and then normalised (never averaged). Every *_ra column is a fraction of prokaryotic (Bacteria + Archaea) coverage — approximately a cell proportion, not a read fraction; not comparable with MetaPhlAn or 16S numbers.

column	meaning
sp_profiled	True when at least one run of the sample has a Sandpiper profile
sandpiper_url	https://sandpiper.qut.edu.au/run/<run> — the RANDOM-selection profiled run with the highest root coverage (v1.2.1)
sp_n_runs_total, sp_n_runs_profiled, sp_partial	run counts; sp_partial = some runs unprofiled
sp_spf	prokaryotic read fraction (%), metagenome-size-weighted over runs
sp_known_species_fraction	% of prokaryotic coverage assigned to a named GTDB species
sp_root_coverage	summed root coverage (× genome equivalents); sp_low_depth = root < 2× (bars hidden)
sp_top_genus, sp_top_genus_ra	most abundant genus-level bin (may be unassigned_at_genus)
sp_ra_g_Bifidobacterium	GTDB g__Bifidobacterium (stable genus)
sp_ra_f_Bacteroidaceae, sp_ra_g_Bacteroides, sp_ra_g_Phocaeicola	GTDB moved B. vulgatus/dorei/plebeius/coprocola to Phocaeicola; the family ≈ NCBI-sense Bacteroides
sp_ra_enterobacterales_core	Σ of six GTDB genera incl. their alphabetic-suffix genera (^g__<Name>(_[A-Z]+)?$): Escherichia, Klebsiella, Enterobacter(+_B,_D), Citrobacter(+_A = amalonaticus/farmeri, _B = koseri, _C, _D), Salmonella, Serratia(+_B,_G) (GTDB f__Enterobacteriaceae is broader than NCBI's, so the family is not used)
sp_ra_g_Escherichia, sp_ra_g_Klebsiella, sp_ra_f_Lachnospiraceae, sp_ra_f_Lactobacillaceae (GTDB splits Lactobacillus), sp_ra_g_Streptococcus, sp_ra_g_Staphylococcus, sp_ra_g_Enterococcus, sp_ra_g_Veillonella, sp_ra_g_Clostridioides (C. difficile)	indicator taxa
sp_ra_unassigned_genus	coverage not resolved to a genus (novel fraction; never renormalised away)
sp_shannon_genus, sp_n_genera_ge1pct	Shannon (ln) over genus bins incl. the unassigned bin; named genera ≥ 1 %
sp_flag_low_complexity	Sandpiper low-complexity rule — expected for Bifidobacterium-/Enterobacterales-dominated neonatal stool; display only, never triage
sp_nonmeta_class, sp_flag_non_metagenome, sp_flag_synthetic, sp_flag_rna, sp_flag_readfraction_warning	see sandpiper_flag_table.csv (single source for field, site label, definition, level; sandpiper_flag_definitions.json is generated from it). sp_flag_non_metagenome = sp_nonmeta_class ∈ {named_nonhuman_host, named_microbe_or_other} over the sample's runs (v1.2.1)
sp_run_concordance_bc, sp_runs_discordant	max pairwise genus-level Bray–Curtis between the sample's runs; discordant = BC > 0.5 (human-review hint, not a verdict)
taxonomy_db, taxonomy_version, sandpiper_version	GTDB / R232 / 2.0.0 on every profiled row

GTDB alphabetic-suffix genera (v1.2.1, R3-1)

Every genus indicator matches ^g__<Name>(_[A-Z]+)?$, i.e. the GTDB genus and its alphabetic-suffix split genera, so that the NCBI-sense genus is covered. gtdb_to_ncbi_note per indicator:

indicator	GTDB taxa summed	gtdb_to_ncbi_note
sp_ra_g_Bifidobacterium	g__Bifidobacterium	single GTDB genus; equals NCBI Bifidobacterium
sp_ra_g_Bacteroides	g__Bacteroides, g__Bacteroides_D, g__Bacteroides_E, g__Bacteroides_F	Bacteroides_D/_E/_F are NCBI Bacteroides species split by GTDB (e.g. B. pectinophilus)
sp_ra_g_Phocaeicola	g__Phocaeicola, g__Phocaeicola_A	Phocaeicola_A added; the family f__Bacteroidaceae ≈ NCBI-sense Bacteroides
sp_ra_enterobacterales_core	g__Citrobacter, g__Citrobacter_A, g__Citrobacter_B, g__Citrobacter_C, g__Citrobacter_D, g__Enterobacter, g__Enterobacter_B, g__Enterobacter_D, g__Escherichia, g__Klebsiella, g__Salmonella, g__Serratia, g__Serratia_B, g__Serratia_G	includes Citrobacter_A (amalonaticus/farmeri), Citrobacter_B (koseri), Enterobacter_B/_D, Serratia_B/_G; Escherichia includes NCBI Shigella
sp_ra_g_Escherichia	g__Escherichia	GTDB Escherichia includes NCBI Shigella
sp_ra_g_Klebsiella	g__Klebsiella	single GTDB genus
sp_ra_g_Streptococcus	g__Streptococcus	single GTDB genus in this dataset
sp_ra_g_Staphylococcus	g__Staphylococcus	single GTDB genus in this dataset
sp_ra_g_Enterococcus	g__Enterococcus, g__Enterococcus_A, g__Enterococcus_B, g__Enterococcus_C, g__Enterococcus_D, g__Enterococcus_E, g__Enterococcus_F, g__Enterococcus_G, g__Enterococcus_H, g__Enterococcus_I, g__Enterococcus_J, g__Enterococcus_K, g__Enterococcus_L	includes GTDB Enterococcus_A (avium/raffinosus/gilvus), _B (faecium/lactis/hirae/durans), _C (asini/dispar), _D (gallinarum/casseliflavus), _E (cecorum), _F … _L; plain g__Enterococcus = E. faecalis group
sp_ra_g_Veillonella	g__Veillonella, g__Veillonella_A	includes Veillonella_A (seminalis/ratti and unnamed species)
sp_ra_g_Clostridioides	g__Clostridioides	single GTDB genus (C. difficile)

Placeholder genera: GTDB R232 contains alphanumeric placeholder genera without a Latin name, e.g. g__ECMA0423 (Enterobacteriaceae; one species, detected in ~34 % of profiled samples at ~0.25 % mean RA) and g__G047199095. They are real GTDB genera, appear in sandpiper_top_genera.parquet and in study panels under their placeholder name, and are not counted in any named indicator column. Bifidobacterium species shares (share_within_bifidobacterium, off-package table) are normalised over named species only; in ~20 % of samples > 10 % of genus coverage is unresolved at species rank, so shares overstate named species — use sp_ra_unassigned_species as the denominator caveat.

<field>__scope (16 coverage fields): scope of the winning determination — sample, subject, biosample (propagated from a parent BioSample) or group (R3/R4 statement applied to a defined group). Apply the README reading rule to group-scope values.

Study columns sp_*: sp_frac_samples_profiled = profiled rows / n_sample_rows (one denominator, v1.2.1); sp_median_* are computed over panel-scope rows (catalog_scope ∧ profiled ∧ NOT sp_low_depth) and are null with sp_panel_scope giving the fallback label when a study has none; sp_frac_runs_flagged counts runs whose organism label is a non-human host or a named microbe, or synthetic / RNA (the routine "Homo sapiens" host label is not counted). sp_auditor_attention/sp_auditor_reason: the Sandpiper track's hint that a study deserves a human look (never a verdict).

Author columns (added v1.2.0)

authors.parquet: one row per study × author × paper (author_display, author_surname, author_initials, is_group, source ∈ linked_paper / bioproject_publication / ena_study_xref, pmid, doi, position, n_authors, is_first, is_last, paper_relation, catalog_status). study_authors_summary.csv: per screened study first_author, last_author, n_authors, n_papers, organisations (ENA centre / broker and NCBI BioProject owner names, ;-joined). Names are string-matched from Europe PMC authorString; the same person may appear under several initial variants and no disambiguation was attempted.

sandpiper_run_qc.parquet (documented v1.2.1)

One row per run of the 389 included studies (174,022 rows). Keys: run_accession, study_accession, sample_accession (BioSample), catalog_sample_key (sample-table key), sample_unit (biosample / run — v1.2.0 wrote the retired value biosample_pooled for the 521 run-unit runs; renamed in v1.2.1 so a join on sample_unit = 'run' works).

column	meaning
in_sandpiper	True when the run has a profile in the Zenodo snapshot
sp_miss_reason	why not (precedence v1.2.1: strategy → platform → access → horizon → size → snapshot gap): profiled · non_wgs_strategy (library_strategy in {AMPLICON, Targeted-Capture, WXS} — the strategies with zero profiled runs in the snapshot; OTHER and WGA are profiled by Sandpiper and are not miss reasons) · non_illumina_platform · controlled_access_study · published_after_snapshot_horizon · unknown_size (ENA base_count ≤ 0, so size cannot be judged) · small_run_lt_100Mbp · ena_illumina_not_in_snapshot / sra_illumina_not_in_snapshot / ddbj_illumina_not_in_snapshot (eligible by every rule but absent from the snapshot)
library_strategy, library_source, instrument_platform, base_count, first_public, organism	ENA run/sample fields used by the rules
sp_root_coverage, sp_species_coverage	root / species-level coverage (genome equivalents)
top1_order_fraction, top3_order_fraction	share of coverage in the top 1 / top 3 orders (low-complexity inputs)
sp_spf, sp_known_species_fraction	prokaryotic read fraction (%), known-species fraction (%)
bacterial_archaeal_bases, metagenome_size, average_bacterial_archaeal_genome_size	Sandpiper size estimates (bases)
sp_warning_present	Sandpiper emitted a warning for the run
sp_prediction, sp_host_or_not	Sandpiper ecological prediction and host-associated call
sp_nonmeta_class	organism-label class: not_flagged · named_human_host · named_nonhuman_host · named_microbe_or_other (see sandpiper_flag_table.csv)
sp_flag_*, sp_low_depth, sp_predicted_ecological, sp_low_known_species_high_depth	run-level flags — every one is defined in sandpiper_flag_table.csv
sandpiper_url, sandpiper_version, taxonomy_db, taxonomy_version, zenodo_record	provenance (Sandpiper 2.0.0 outputs, Zenodo 20419175, GTDB R232; the live Sandpiper site may show a newer version)

sandpiper_study_panels.parquet (v2, v1.2.1)

Mean relative abundance per study over panel-scope rows = catalog_scope ∧ sp_profiled ∧ NOT sp_low_depth (top-12 phyla, top-15 genera, plus the unassigned_at_* bins with rank_order = 0). Columns: study_accession, rank, taxon, mean_rel_abundance, n_samples_panel (rows averaged; asserted equal to the recomputation from the shipped sample table), rank_order, n_samples_study (rows of the study), n_catalog_scope, n_profiled, n_profiled_catalog_scope, n_profiled_age_infant / n_profiled_age_adult / n_profiled_age_unknown_or_other (age_scope breakdown of all profiled rows, for the caption), n_low_depth_excluded_from_panel, frac_samples_profiled, panel_scope, panel_definition, taxonomy_db, taxonomy_version. Studies without panel-scope rows have no rows here; their fallback label is in sandpiper_study_panel_status.csv and in study_metadata_wide.sp_panel_scope. The v1.2.0 merge artefacts n_samples_x/n_samples_y are gone.

sandpiper_flag_table.csv (v1.2.1)

Single source for every Sandpiper flag: field, level (run / sample / study), site_label, definition, used. sandpiper_flag_definitions.json is generated from it; the GlobDB domain-only flags are kept with used = false (the catalog uses GTDB profiles only).

organisations.parquet (v2, v1.2.1) and study_authors_summary.csv

One row per study × organisation token × source (bioproject_organization = NCBI BioProject Organization, submitter_center = ENA center_name, ena_broker). v1.2.1 adds not_org_reason (submission_id for ^SUB\d+( \| SUB\d+)*$, email_or_handle for tokens containing @, placeholder for n/a / None / '.', lowercase_username for single all-lowercase tokens such as hpsun), is_organisation (not_org_reason is null; such rows get org_type = not_an_organisation) and display_eligible (is_organisation AND, for PRJNA studies with an NCBI BioProject organisation, not an ENA center_name). organisations_index.json, study_authors_summary.organisations and study_metadata_wide.organisations are built from display_eligible rows only; a check asserts that no published organisation string matches SUB\d{6,} or contains @. Single lowercase tokens include some genuine company/institute acronyms (e.g. novogene, rivm) — they stay in the parquet for owner whitelisting but are not displayed.

authors.parquet (v2): author_key folds names with a transliteration map (ø→o, æ→ae, ß→ss, ł→l, đ→d, İ→i, Ð→d, Þ→th) before NFKD stripping, and folds initials too; the group regex includes office / production / editorial; name_needs_check flags no-initials multi-token person names for manual review.

Complete column reference (generated from the parquet schema; the build fails on an undocumented column)

sample_metadata_wide.parquet (147 columns)

column	dtype	description
sample_key	str	primary key: BioSample accession for biosample units, run accession for run units
study_accession	str	BioProject accession
secondary_sample	str	ENA secondary sample accession
sample_title	str	submitter sample title
body_site_class	str	body-site scope class — vocabulary: primary (infant gut/stool/meconium/rectal), unknown (no body-site evidence; kept in body-site scope), excluded (non-gut site: skin, oral, milk, vaginal, nasal…; or, since v1.2.1, non-human host / isolate genome — see exclusion_reason_code), linked (mother/other sample linked to an infant subject; out of body-site scope)
collection_date	str	archive collection_date as given
is_gold_heldout	bool	True if in the curatedMetagenomicData evaluation set
probiotic_exposure	str	metadata field probiotic_exposure (see Field vocabularies)
preterm_status	str	metadata field preterm_status (see Field vocabularies)
gestational_age_weeks	float64	metadata field gestational_age_weeks (see Field vocabularies)
delivery_mode	str	metadata field delivery_mode (see Field vocabularies)
feeding_mode	str	metadata field feeding_mode (see Field vocabularies)
antibiotic_exposure	str	metadata field antibiotic_exposure (see Field vocabularies)
age_at_collection_days	float64	metadata field age_at_collection_days (see Field vocabularies)
birth_weight_grams	float64	metadata field birth_weight_grams (see Field vocabularies)
country	str	metadata field country (see Field vocabularies)
maternal_antibiotics	str	metadata field maternal_antibiotics (see Field vocabularies)
hmo_supplementation	str	metadata field hmo_supplementation (see Field vocabularies)
nec_status	str	metadata field nec_status (see Field vocabularies)
health_condition	str	metadata field health_condition (see Field vocabularies)
multiple_birth	str	metadata field multiple_birth (see Field vocabularies)
sibling_in_study	str	metadata field sibling_in_study (see Field vocabularies)
geo_subregion	str	metadata field geo_subregion (see Field vocabularies)
sex	str	metadata field sex (see Field vocabularies)
timepoint_label	str	metadata field timepoint_label (see Field vocabularies)
subject_id	str	metadata field subject_id (see Field vocabularies)
probiotic_exposure__confidence	float64	confidence of probiotic_exposure (0–1)
preterm_status__confidence	float64	confidence of preterm_status (0–1)
gestational_age_weeks__confidence	float64	confidence of gestational_age_weeks (0–1)
delivery_mode__confidence	float64	confidence of delivery_mode (0–1)
feeding_mode__confidence	float64	confidence of feeding_mode (0–1)
antibiotic_exposure__confidence	float64	confidence of antibiotic_exposure (0–1)
age_at_collection_days__confidence	float64	confidence of age_at_collection_days (0–1)
birth_weight_grams__confidence	float64	confidence of birth_weight_grams (0–1)
country__confidence	float64	confidence of country (0–1)
maternal_antibiotics__confidence	float64	confidence of maternal_antibiotics (0–1)
hmo_supplementation__confidence	float64	confidence of hmo_supplementation (0–1)
nec_status__confidence	float64	confidence of nec_status (0–1)
health_condition__confidence	float64	confidence of health_condition (0–1)
multiple_birth__confidence	float64	confidence of multiple_birth (0–1)
sibling_in_study__confidence	float64	confidence of sibling_in_study (0–1)
geo_subregion__confidence	float64	confidence of geo_subregion (0–1)
sex__confidence	float64	confidence of sex (0–1)
timepoint_label__confidence	float64	confidence of timepoint_label (0–1)
subject_id__confidence	float64	confidence of subject_id (0–1)
probiotic_exposure__route	str	route of probiotic_exposure (R1–R4, see Routes)
preterm_status__route	str	route of preterm_status (R1–R4, see Routes)
gestational_age_weeks__route	str	route of gestational_age_weeks (R1–R4, see Routes)
delivery_mode__route	str	route of delivery_mode (R1–R4, see Routes)
feeding_mode__route	str	route of feeding_mode (R1–R4, see Routes)
antibiotic_exposure__route	str	route of antibiotic_exposure (R1–R4, see Routes)
age_at_collection_days__route	str	route of age_at_collection_days (R1–R4, see Routes)
birth_weight_grams__route	str	route of birth_weight_grams (R1–R4, see Routes)
country__route	str	route of country (R1–R4, see Routes)
maternal_antibiotics__route	str	route of maternal_antibiotics (R1–R4, see Routes)
hmo_supplementation__route	str	route of hmo_supplementation (R1–R4, see Routes)
nec_status__route	str	route of nec_status (R1–R4, see Routes)
health_condition__route	str	route of health_condition (R1–R4, see Routes)
multiple_birth__route	str	route of multiple_birth (R1–R4, see Routes)
sibling_in_study__route	str	route of sibling_in_study (R1–R4, see Routes)
geo_subregion__route	str	route of geo_subregion (R1–R4, see Routes)
sex__route	str	route of sex (R1–R4, see Routes)
timepoint_label__route	str	route of timepoint_label (R1–R4, see Routes)
subject_id__route	str	route of subject_id (R1–R4, see Routes)
subject_key	str	resolved subject id (study-scoped)
role	str	infant · mother · other · unknown · adult · child (see Sample unit section)
t_index	float64	timepoint index within subject
n_timepoints_subject	float64	timepoints of the subject
linked_infant_subject_key	str	infant subject a mother/other sample is linked to
run_accessions	str	;-joined run accessions of the sample
n_runs	float64	number of runs
instrument_model	str	instrument model (top)
library_layout	str	PAIRED/SINGLE
read_count_total	float64	summed read_count over runs
study_title	str	BioProject title
cohort_id	str	cohort id (cohorts.csv)
cohort_name	str	cohort name
first_public_min	str	earliest first_public of the runs
adult_age_flag	bool	True when a committed age is > 1,100 d (see age_scope = adult_flagged)
n_fields_with_value	int64	count of the 19 fields with a value
ena_sample_url	str	ENA browser link
ena_study_url	str	ENA study link
sample_unit	str	biosample or run (see Sample unit section)
parent_biosample	str	parent BioSample of a run unit
biosample_accession	str	BioSample accession (always populated)
run_accession	str	run accession for run units, else null
age_scope	str	age-scope class (see Sample unit section). Footnote: adult_flagged means "over-age flagged" — any committed age > 1,100 d, which includes children aged 3–18 y (role = child), not only adults
age_scope_basis	str	evidence behind age_scope
role_source	str	evidence behind role
sp_profiled	bool	True when at least one run has a Sandpiper profile
sandpiper_url	str	link to the representative profiled run: RANDOM-selection run with the highest root coverage (v1.2.1; v1.2.0 used the alphabetically first run)
sp_n_runs_total	int64	runs of the sample (populated for every row since v1.2.1)
sp_n_runs_profiled	int64	profiled runs (0 when unprofiled)
sp_partial	bool	True when 0 < profiled runs < total runs; False for unprofiled rows
sp_spf	float64	prokaryotic read fraction (%)
sp_known_species_fraction	float64	% of prokaryotic coverage assigned to a named GTDB species
sp_root_coverage	float64	summed root coverage (genome equivalents)
sp_low_depth	boolean	root coverage < 2 (excluded from panels/medians)
sp_top_genus	str	most abundant genus bin
sp_top_genus_ra	float64	its relative abundance
sp_ra_g_Bifidobacterium	float64	GTDB g__Bifidobacterium (no suffix genera in R232)
sp_ra_f_Bacteroidaceae	float64	GTDB f__Bacteroidaceae
sp_ra_g_Bacteroides	float64	GTDB g__Bacteroides + suffix genera Bacteroides_D/_E/_F
sp_ra_g_Phocaeicola	float64	GTDB g__Phocaeicola + Phocaeicola_A
sp_ra_enterobacterales_core	float64	Σ of GTDB Escherichia, Klebsiella, Enterobacter(+_B,_D), Citrobacter(+_A,_B,_C,_D), Salmonella, Serratia(+_B,_G)
sp_ra_g_Escherichia	float64	GTDB g__Escherichia (includes NCBI Shigella)
sp_ra_g_Klebsiella	float64	GTDB g__Klebsiella
sp_ra_f_Lachnospiraceae	float64	GTDB f__Lachnospiraceae
sp_ra_f_Lactobacillaceae	float64	GTDB f__Lactobacillaceae (GTDB splits NCBI Lactobacillus into many genera)
sp_ra_g_Streptococcus	float64	GTDB g__Streptococcus
sp_ra_g_Staphylococcus	float64	GTDB g__Staphylococcus
sp_ra_g_Enterococcus	float64	GTDB g__Enterococcus (= E. faecalis group) + Enterococcus_A (avium/raffinosus/gilvus) + _B (faecium/lactis/hirae/durans) + _C … _L
sp_ra_g_Veillonella	float64	GTDB g__Veillonella + Veillonella_A
sp_ra_g_Clostridioides	float64	GTDB g__Clostridioides (C. difficile)
sp_ra_unassigned_genus	float64	coverage not resolved to a genus
sp_shannon_genus	float64	Shannon (ln) over genus bins incl. unassigned
sp_n_genera_ge1pct	float64	named genera ≥ 1 %
sp_flag_low_complexity	boolean	see sandpiper_flag_table.csv
sp_flag_non_metagenome	bool	see sandpiper_flag_table.csv (derived from sp_nonmeta_class since v1.2.1)
sp_flag_synthetic	boolean	see sandpiper_flag_table.csv
sp_flag_rna	boolean	see sandpiper_flag_table.csv
sp_flag_readfraction_warning	boolean	see sandpiper_flag_table.csv
sp_run_concordance_bc	float64	max pairwise genus Bray–Curtis between runs
sp_runs_discordant	boolean	BC > 0.5
taxonomy_db	str	GTDB
taxonomy_version	str	R232
sandpiper_version	str	2.0.0 (Zenodo snapshot 20419175; the live Sandpiper site may show a newer version)
age_at_collection_days__scope	string	scope of the winning age_at_collection_days determination: sample / subject / biosample / group
antibiotic_exposure__scope	string	scope of the winning antibiotic_exposure determination: sample / subject / biosample / group
birth_weight_grams__scope	string	scope of the winning birth_weight_grams determination: sample / subject / biosample / group
country__scope	string	scope of the winning country determination: sample / subject / biosample / group
delivery_mode__scope	string	scope of the winning delivery_mode determination: sample / subject / biosample / group
feeding_mode__scope	string	scope of the winning feeding_mode determination: sample / subject / biosample / group
geo_subregion__scope	string	scope of the winning geo_subregion determination: sample / subject / biosample / group
gestational_age_weeks__scope	string	scope of the winning gestational_age_weeks determination: sample / subject / biosample / group
health_condition__scope	string	scope of the winning health_condition determination: sample / subject / biosample / group
hmo_supplementation__scope	string	scope of the winning hmo_supplementation determination: sample / subject / biosample / group
maternal_antibiotics__scope	string	scope of the winning maternal_antibiotics determination: sample / subject / biosample / group
multiple_birth__scope	string	scope of the winning multiple_birth determination: sample / subject / biosample / group
nec_status__scope	string	scope of the winning nec_status determination: sample / subject / biosample / group
preterm_status__scope	string	scope of the winning preterm_status determination: sample / subject / biosample / group
probiotic_exposure__scope	string	scope of the winning probiotic_exposure determination: sample / subject / biosample / group
sibling_in_study__scope	string	scope of the winning sibling_in_study determination: sample / subject / biosample / group
exclusion_reason_code	object	controlled reason code when a deterministic sample rule set body_site_class = excluded in v1.2.1: host_nonhuman (ENA scientific_name/tax_id is a non-human animal) or assay_isolate_genome (all runs GENOMIC with a named microbe organism); null otherwise. Rows are logged in value_history (change_stage auditor_review:R3-3)
sp_nonmeta_class	str	any-run organism-label class over ALL runs of the sample (see sandpiper_flag_table.csv): not_flagged · named_human_host · named_nonhuman_host · named_microbe_or_other
catalog_scope	bool	headline scope (v1.2.1) = age_scope ∈ {infant_evidenced, study_all_infant} AND body_site_class ∈ {primary, unknown}; the intersection used for headline counts (71,795 rows) and for Sandpiper study panels/medians
panel_scope	bool	True when the row enters its study's Sandpiper panel/medians: catalog_scope AND sp_profiled AND NOT sp_low_depth

study_metadata_wide.parquet (120 columns)

column	dtype	description
study_accession	str	BioProject accession (key)
triage_verdict	str	include/exclude/uncertain from the infant triage cascade. INFANT-CATALOG verdict (historical, 2026-09 infant screen) — NOT membership in the all-age catalog: a study can be excluded here (e.g. age_adult_only) and still be in gut_studies. Catalog membership = presence in gut_studies.parquet.
outcome	str	contract outcome of the final verdict row
confidence	float64	confidence of the final verdict
reason_code	str	controlled exclusion reason of the INFANT screen (null for included); age_* codes are infant-age reasons, not catalog exclusions
evidence	str	JSON evidence list of the final verdict
model	str	model/stage that produced the final verdict
decision_stage	str	stage table of the final verdict
note	str	free-text note of the final verdict
n_infant_samples_est	float64	triage estimate of infant samples
body_site_call	str	triage body-site call for the study
slot	str	verdict slot name
validator_ok	object	validator result on the verdict row
validator_msg	str	validator message
study_title	str	BioProject title
n_samples	int64	from runs.parquet: distinct BioSamples with runs registered under this study (runs for run-unit studies). Sums to 154,268 over studies because 62 BioSamples carry runs from two BioProjects; use n_sample_rows for rows in the sample table
n_runs	int64	runs registered under the study
first_public_min	str	earliest first_public
sig_infant_hit	bool	enumeration infant signal hit
sig_infant_title	bool	infant term in title
screen_haiku	str	Haiku screen label (yes/no)
linked_infant_paper	bool	a linked paper is infant-scoped
universe_slice	str	enumeration channel
catalog_status	str	included / excluded / human_review in the INFANT catalog. INFANT-CATALOG verdict (historical, 2026-09 infant screen) — NOT membership in the all-age catalog: a study can be excluded here (e.g. age_adult_only) and still be in gut_studies. Catalog membership = presence in gut_studies.parquet.
assessed_downstream	bool	reached the extraction stage
recov_age	str	recoverability tier for age (R0–R4)
recov_antibiotics	str	recoverability tier for antibiotics (R0–R4)
recov_delivery	str	recoverability tier for delivery (R0–R4)
recov_feeding	str	recoverability tier for feeding (R0–R4)
recov_preterm	str	recoverability tier for preterm (R0–R4)
recov_probiotic	str	recoverability tier for probiotic (R0–R4)
cohort_id	str	cohort id
cohort_name	str	cohort name
cohort_crosswalk_names	str	alternative cohort names
n_biosample_infant_flagged	int64	BioSamples flagged infant by archive attributes at triage
n_linked_papers	int64	linked papers
linked_pmids	str	;-joined PMIDs
median_gbp	float64	median bases per run (Gbp)
median_read_len	float64	median nominal read length
paired_share	float64	fraction of PAIRED runs
instrument_model_top	str	most frequent instrument model
generation_top	str	sequencer generation of instrument_model_top
timepoint_class	str	cross_sectional / longitudinal heuristic from sample titles
country_top	str	most frequent country
worklist_rank	float64	curation priority rank
expected_field_yield	float64	predicted number of recoverable fields
effort_class	str	curation effort class
controlled_access	object	True when the study is controlled-access (dbGaP/EGA)
n_biosamples	int64	BioSample-unit rows of the study in the sample table (v1.2.1: parents of run units no longer counted here; see n_run_units, n_parent_biosamples)
sample_unit	str	dominant sample unit of the study: biosample / run
cov_probiotic_exposure	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a probiotic_exposure value
cov_preterm_status	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a preterm_status value
cov_gestational_age_weeks	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a gestational_age_weeks value
cov_delivery_mode	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a delivery_mode value
cov_feeding_mode	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a feeding_mode value
cov_antibiotic_exposure	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a antibiotic_exposure value
cov_age_at_collection_days	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a age_at_collection_days value
cov_birth_weight_grams	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a birth_weight_grams value
cov_country	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a country value
cov_maternal_antibiotics	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a maternal_antibiotics value
cov_hmo_supplementation	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a hmo_supplementation value
cov_nec_status	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a nec_status value
cov_health_condition	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a health_condition value
cov_multiple_birth	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a multiple_birth value
cov_sibling_in_study	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a sibling_in_study value
cov_geo_subregion	float64	fraction of catalog-scope-eligible rows (n_infant_scope_samples denominator) with a geo_subregion value
n_unique_infants_est	float64	distinct infant subjects (see Sample unit section)
max_timepoints	float64	max timepoints per infant subject
longitudinal	bool	True when max_timepoints > 1
n_mothers	int64	mother-role subjects
n_infant_scope_samples	int64	body-site scope count: rows with body_site_class ∈ {primary, unknown} (name kept for compatibility; not age-evidenced)
ena_url	str	ENA link
ncbi_url	str	NCBI link
n_age_scope_infant	int64	rows with age_scope ∈ {infant_evidenced, study_all_infant}
n_adult_flagged	int64	rows with age_scope = adult_flagged (over-age, includes children)
n_age_unknown_mixed_study	int64	rows age_unknown_mixed_study
n_age_unknown_no_study_estimate	int64	rows age_unknown_no_study_estimate
n_infant_evidenced	int64	rows infant_evidenced
n_non_infant_role	int64	rows non_infant_role
n_study_all_infant	int64	rows study_all_infant
mixed_age_deposit	bool	> 50 % of rows are age_unknown_mixed_study or adult_flagged
n_infant_role_samples	int64	rows with role = infant
n_infant_samples_with_subject	int64	infant rows with a resolved subject
n_unique_infants_source	str	source class of the unique-infant estimate
n_unique_infants_upper	float64	upper bound
n_unique_infants_display	str	string to render
n_sample_rows	int64	rows of the sample table keyed to this study
n_biosamples_shared_with_other_study	int64	BioSamples of this study also registered under another BioProject
sp_n_samples_profiled	int64	sample rows with a profile
sp_frac_samples_profiled	float64	sp_n_samples_profiled / n_sample_rows (single denominator since v1.2.1)
sp_median_bifidobacterium_ra	float64	median sp_ra_g_Bifidobacterium over panel-scope rows (catalog_scope ∧ profiled ∧ NOT low depth); null when the study has none
sp_median_shannon_genus	float64	median Shannon over panel-scope rows
sp_n_runs_profiled	int64	profiled runs registered under this study
sp_frac_runs_profiled	float64	sp_n_runs_profiled / n_runs
sp_coverage_class	str	none (no profiled rows) · partial · full (every row profiled); defined on sample rows since v1.2.1
sp_frac_runs_flagged	float64	see sandpiper_flag_table.csv
sp_frac_low_complexity_profiled	float64	see sandpiper_flag_table.csv
sp_frac_low_depth_profiled	float64	see sandpiper_flag_table.csv
sp_median_spf	float64	median prokaryotic read fraction over panel-scope rows
sp_median_known_species_fraction	float64	median known-species fraction over panel-scope rows
sp_auditor_attention	bool	see sandpiper_flag_table.csv
sp_auditor_reason	str	reason text for sp_auditor_attention
first_author	str	first author of the primary linked paper
last_author	str	last author
n_authors	int64	distinct author keys over linked papers
n_author_papers	int64	papers contributing authors
organisations	str	;-joined organisation names (v1.2.1: submission ids, e-mail addresses, placeholders and single lowercase tokens removed; NCBI BioProject Organization preferred over ENA center_name for PRJNA studies)
sp_median_scope	str	scope label of the medians (same vocabulary as sp_panel_scope)
sp_n_samples_panel	int64	rows in the panel scope
sp_panel_scope	str	catalog_scope_adequate_depth (panel shown) · fallback:no_catalog_scope_samples · fallback:no_age_evidenced_profiled_samples · fallback:all_age_evidenced_profiled_samples_low_depth · no_profiled_samples — render the fallback text instead of a panel
sp_coverage_note	str	set when rows are profiled only through runs registered under a sibling BioProject
n_catalog_scope	int64	rows with catalog_scope = True (age scope ∧ body-site scope) — the headline denominator
n_excluded_host_or_isolate	int64	rows set to body_site_class = excluded by the v1.2.1 host/isolate rule (exclusion_reason_code not null)
n_body_site_excluded	int64	rows with body_site_class = excluded
n_run_units	int64	run-unit rows (one BioSample per infant, one run per stool)
n_parent_biosamples	int64	parent BioSamples of the run units (listed in parent_biosamples.parquet, counted nowhere else)
shared_biosample_note	object	human-readable note for the five studies sharing BioSamples (F6)
release_added	string	release in which this row first became visible (R<YYYY>.<n> or a pre-numbered package semver 1.0.0–1.2.2); pre-1.3.0 values are reconstructed — see DATA_DICTIONARY 'Release columns'
release_retired	string	release that replaced/removed the row; null (empty in CSV) = current row
package_added	string	semver of the data package in which the row first appeared