Appendix B — Glossary
The words this book uses with a specific meaning, each ending with the chapter that says more.
bottle — a Niskin bottle closed at a depth during a cast; in the database a sample of type bottle under its cast, with the water-chemistry measurements taken from it in obs_env. → The database
cast — one lowering of the CTD or the bottle rosette at a station; a root sample of type cast with bottles under it. → CTD QA/QC
catalog — catalog.json, the release’s manifest: every table, its rows and the parquet objects that hold it; the contract every reader resolves through. → Releases
core — the small family of tables every consumer reads — sample, obs_bio, obs_env, obs_attribute, sample_measurement and the references — that every dataset projects into. → The database
cruise — one voyage of one ship; keyed YYYY-MM-NODC by the month SWFSC designates for it and the ship’s NODC code, never by the month an event fell in. → Keys and integrity
dataset key — {provider}_{dataset}, the name of everything about a dataset: its notebook, its metadata folder, its record, its page, the provenance stamp on every observation. → Naming conventions
grain — what one row stands for: a physical sampling event in sample, one measured value in obs_env, one taxon’s headline in obs_bio, one bin in obs_attribute. → The database
grid cell — a cell of the CalCOFI station grid a sample falls in (grid_key); several real stations share one nearshore cell. → Keys and integrity
hexagon — an H3 cell; every observation carries its resolution-10 cell (hex_id) and coarser hexagons are computed from it. → Explore
holding — a dataset CalCOFI has or knows of but has not ingested; it has a page and a record but no rows. → Metadata & the ingest loop
ingest — the reproducible notebook that turns one source dataset into the shared schema. → Ingesting a dataset
life stage — the developmental stage of a counted organism, from a registered vocabulary with Darwin Core and NERC ids. → Naming conventions
line and station — CalCOFI’s survey grid: a line runs offshore from the coast, a station is a position along it; together the site_key (093.3 026.4). → Keys and integrity
measurement type — a registered measured quantity with its units and physical bounds; every obs_env row names one. → Naming conventions
net, tow — one deployment of a net (a tow) and each net on it; sample rows of type tow and net under a site, with the counts from the net in obs_bio. → The database
observation — one measured or counted value at one place and time, in obs_env or obs_bio; obs is the two under one name. → Access the data
provider — the organization that curates, holds and can license a dataset; never the portal that serves it. → Naming conventions
quality flag — a dataset’s own verdict on a value (measurement_qual), kept uninterpreted; qual_ok is the release’s reading of it. → Access the data
realm — bio or env: whether an observation counts an organism or measures the water. → The database
record — datasets.json, one entry per dataset generated at release, from which every portal surface is derived. → Portals
release — an immutable, versioned, cited copy of every table, cut every few weeks; latest is the promoted one. → Releases
root sample — a sample with no parent — a site, a cast, an underway record — and the root_id the browser objects join on. → Keys and integrity
sample — a physical sampling event of any kind, at its native grain, with its position, time and depth span, and its parent. → The database
sidecar — a JSON or markdown file beside the parquet — the catalog, the metadata, the keys, the record, the notes — that describes the release and never needs the bytes to be read. → Releases
supplemental — a table hosted and catalogued but left out of the default set: the full-resolution CTD and meteorology series, the root-sample join twin. → Access the data
taxon key — worms:<AphiaID>, or itis:<TSN> for birds: the accepted identifier of an organism in the authority that carries it. → The database
working store — the CTD team’s PostgreSQL database: originals verbatim, flags beside them, built to feed the next release through a nightly snapshot and never read by a consumer. Running and available since August 2026; not yet the team’s practice, so today CTD flags still arrive with the provider’s files. → CTD QA/QC