Reference

The database schema

One pipeline, four layers: a jurisdiction is identified, its adopted code is retrieved chapter by chapter, each chapter is read into rules, and each rule yields discrete requirements — every layer keeping the citation back to the text. The row counts below are live.

zoning_locality

17,523 rows · one zoning jurisdiction each

The national roster. A row is created when a jurisdiction is identified, long before its ordinance is retrieved — which is what lets the site distinguish no rules exist from not read yet. TIGER FIPS columns join it to Census geometry; the nza_* columns carry National Zoning Atlas figures verbatim for the 118 jurisdictions the Atlas has read.

Column	Type	Meaning
locality_id	bigint	Primary key, used in every /localities URL
locality_name · locality_type	text	Name and kind — municipality, county unincorporated, township…
state_abbr · state_fips	char	State, as postal code and FIPS
placefp · cousubfp · fips_county	char	TIGER joins to Census place / county-subdivision / county geometry
discovery_status	text	pending → found → acquired; or no_code / unavailable / error
has_zoning	boolean	NULL = unknown; FALSE = confirmed to have no zoning code
ordinance_url	text	Where the adopted code was located
population · housing_units	integer	Census figures, where joined
nza_*	numeric…	National Zoning Atlas statistics, in acres — kept verbatim and shown separately

Exposed atLocalities Zoning overview API

zoning_ordinance_content

3,931 rows · one retrieved chapter each

The full text. Each chapter of an adopted code is retrieved from its host, OCR’d, and structured into markdown; both forms are kept. A functional GIN index over the text is what answers full-text search, and structured_markdown is what the chapter pages render — with extracted standards highlighted in place.

Column	Type	Meaning
content_id	bigint	Primary key, used in every /chapters URL
locality_id	bigint	→ zoning_locality
source · source_url · document_url	text	Which host the chapter came from, and where
chapter · document_title	text	The chapter heading as published
ocr_text	text	Raw OCR output
structured_markdown	text	The same text, structured — headings, tables, emphasis
char_count	integer	Generated column over the OCR text; chapters average 67 KB, the largest is 3 MB
status	text	pending → ocr_done → markdown_done → rules_extracted

Exposed atChapters Text search API

zoning_ordinance_rules

42,251 rows · one extracted rule each

The heart of the database. A rule with a zone_code is a district — R-1, B-2 — carrying the ordinance’s dimensional standards for that zone across roughly fifty typed columns. A rule with rule_scope = ‘general’ is a citywide provision: most of a code’s volume. Every rule keeps the section and excerpt it was read from.

Column	Type	Meaning
rule_id	bigint	Primary key, used in every /districts URL
content_id · locality_id	bigint	→ the chapter it was read from, and the jurisdiction
rule_scope	text	district · general · overlay
zone_code · zone_name · zone_category	text	The district identity; category drives the site-wide legend
min_lot_area_sqft … max_far	numeric ×50	Dimensional standards: lot, density, setbacks, height, bulk, parking
permitted/conditional/accessory/prohibited_uses	text[]	The use lists, as the ordinance names them
sf/mf2/mf3/mf4plus_treatment	text	How 1–4+ family housing is treated: by_right · conditional · prohibited
adu_allowed	boolean	Accessory dwelling units; NULL = the ordinance does not say
source_section · source_excerpt	text	The citation: where in the code this was read, and its language
confidence · needs_review	real · boolean	Extractor self-assessment, surfaced on the page when flagged

Exposed atDistrict pages District search API

zoning_ordinance_standard

170,998 rows · one discrete requirement each

Requirements lifted out of the prose, one per row — a maximum sign area, a fence setback, a riparian buffer — each typed by value_kind and each carrying raw_text, the sentence it was read from. That sentence is what the chapter pages locate and highlight.

Column	Type	Meaning
standard_id	bigint	Primary key; the #standard-N anchors on chapter pages
rule_id · content_id · locality_id	bigint	→ the rule, chapter and jurisdiction it belongs to
category	text	One of 28 checked values — signage, parking, height, floodplain…
standard_key · label	text	Machine-generated key and human label
numeric_value · numeric_value_max · unit	numeric · text	The value, or a range, with its unit
bool_value · text_value · enum_value	various	Non-numeric values, per value_kind
raw_text	text	The exact sentence the requirement was read from
applies_when	jsonb	Conditions scoping the requirement, where stated

Exposed atStandards API

zoning_standard_def

195,908 rows · one standard key each

The vocabulary of standard_keys with default units and value kinds. Machine-generated and definition-free — which is why the site’s glossary is authored rather than derived from it.

Column	Type	Meaning
standard_key	text	Primary key
category · label · default_unit	text	Grouping, display label, unit
value_kind	text	numeric · range · boolean · text · enum

state_zoning_statute

50 rows · one state enabling act each

The state law behind every local ordinance: the authority model (who may zone), the primary citation, the delegating text and a plain summary, with a verification flag.

Column	Type	Meaning
state_abbr · state_name	text	One row per state
authority_model	text	county_and_municipal · municipal_delegation · state_level, with local nuance
primary_citation · statute_title · source_url	text	The citation and where to read it
statute_text · summary	text	The delegating language and a plain-English summary
verified · confidence	boolean · text	Whether the record was checked against the source

Exposed atStatutes API

zoning_ordinance_source

18,119 rows · one candidate source each

Discovery bookkeeping: every candidate URL found while hunting for a jurisdiction’s adopted code, with its kind and status. Feeds the pipeline; not exposed on the site directly.

Column	Type	Meaning
locality_id	bigint	→ zoning_locality
source · url · kind · title	text	Where a code might live, and what kind of host it is
status	text	Whether the candidate panned out

Database notes

NULL means silence, not no. Every extracted column distinguishes what the ordinance does not say from what it forbids — adu_allowed is NULL where the code never addresses it, and the site renders that as “not stated” throughout.

On the roster ≠ read. A locality row exists as soon as a jurisdiction is identified; only districts extracted from its chapters make it searchable. And a locality with rules is not necessarily one with districts — a retrieved subdivision chapter yields general provisions and nothing else.

Numerics travel as strings. Postgres numeric preserves precision, so the JSON API returns dimensional values as strings — parse them, don’t assume numbers.

The read-only grants the site runs under are documented in .env.example; the optional text-search migration ships as sql/search_index.sql and belongs to the pipeline, not the site.