Data dictionary

Every column, defined.

The same schema for every county, state and the nationwide file, in CSV, Excel or Parquet. How complete each column is varies by department — every coverage page shows it.

Column	Type	Description
`permit_number`Permit number	text	The permit or case number as issued by the building department.
`permit_type`Permit type	text	The department’s own classification — e.g. Residential Electrical, Roofing, Solar, New Construction, Demolition.
`work_description`Work description	text	Free-text description of the permitted work, as written on the application.
`permit_status`Status	text	Lifecycle status as reported by the source: applied, issued, finaled, expired, withdrawn …
`application_date`Application date	date	Date the permit application was filed.
`issue_date`Issue date	date	Date the permit was issued.
`complete_date`Completion date	date	Date of final inspection, certificate of occupancy, or close-out.
`expiration_date`Expiration date	date	Date the permit expires or expired.
`valuation`Job value	numeric (USD)	Declared cost or valuation of the work.
`contractor`Contractor	text	Contractor or applicant of record.
`address`Site address	text	Street address of the permitted work.
`parcel_id`Parcel identifier	text	The parcel / tax-map identifier recorded with the permit, for joining to parcel and assessment data.
`parcel_id_kind`Parcel identifier kind	enum	What parcel_id holds: tax_account, parcel_number, account_alt, geo (a GIS feature id), address (no id, address only) or none.
`locality_geoid`Locality FIPS	text (5)	Census FIPS code of the county or independent city.
`locality_name`Locality	text	County or independent-city name.
`historical`Historical	boolean	True when the record comes from the department’s archive or historical feed rather than its current one.
`source`Source	text	The building-department dataset the record was collected from.
`raw`Original record	JSON	The untouched source record, every original field included — nothing the department published is dropped.
`fetched_at`Collected at	timestamp	When the record was last collected from the source.

parcel_id_kindWhat the parcel key is.

tax_accountTax account: The assessor’s tax account or real-estate account number.
parcel_numberParcel number: A parcel / tax-map number (PIN, APN, GPIN, map-lot).
account_altAlternate account: A secondary account number the department uses for the property.
geoGIS feature id: The id of a GIS feature (often the parcel polygon’s object id) the permit is attached to.
addressAddress only: No parcel id was recorded; the site address is the only key.
noneNone: Neither a parcel id nor a usable address was recorded.

How the file is built.

Department vocabulary, kept verbatim

permit_type and permit_status are exactly what the department recorded — “NHC Residential Electrical”, “B-CONVERT”, “Final”. Nothing is collapsed into a house taxonomy that could hide what the source meant; the original record is alongside if you need any field we do not normalize.

Rows and permits

Where a department publishes the same permit in more than one dataset (a current feed and an archive, say), the file carries one row per source record and the coverage pages count each permit number once.

Dates are dates

Every date column is an ISO date (YYYY-MM-DD), parsed from whatever the source used — epoch milliseconds, US-style strings, Excel serials.

Subscription files roll forward

Each monthly subscription file holds every permit dated in the trailing 12 months — the whole window, not a diff — so loading it is a replace. Status changes on permits still in the window are applied.