Methodology

Data as of 2026-09-05. Republishes new-business registrations from FOUR states' own official open-data registries (CO, CT, NY, OR) for the window each state publishes — NOT a national register, not an official formation statistic, and never a list of people.

One snapshot, two formats

The JSON files and CSV sample use the same collected state-registry records. Use JSON for software and the sample CSV for a spreadsheet. Registration dates describe filings; collection dates describe when the source was read. Neither establishes today's legal status. Paid ongoing delivery is not available yet.

Sources (official, free, keyless)

• CO — Colorado Secretary of State (via Colorado Information Marketplace): dataset. Statewide business-entity register on the state's Socrata portal, keyless. Weekly pull keeps entities whose entityformdate falls in the trailing 28 days. Entity types are SoS codes (DLLC = domestic LLC, DPC = domestic profit corp, FLLC = foreign LLC, DNC = domestic nonprofit corp, ...); the code ships verbatim in entity_type. Agent-name columns exist in the source and are never selected.
• CT — Connecticut Secretary of the State (Business Registry on data.ct.gov): dataset. The Business Master table of CT's published business registry, keyless Socrata. Principals and agents live in SEPARATE datasets (ka36-64k6, qh2m-n44y) which this asset never touches. business_type is plain (LLC / Stock / Non-Stock / LLP / Limited Partnership / Religious).
• NY — New York Department of State (Division of Corporations): dataset. DOS publishes every filing processed in the previous 30 days. Only NEW-ENTITY filing types are kept (see NY_FORMATION_TYPES); biennial statements, amendments, dissolutions, and assumed names are excluded. Location is the county of the principal office (cnty_prin_ofc) — NY publishes no city in this file. filer_name / sop_name (people) are never selected.
• OR — Oregon Secretary of State (Corporation Division): dataset. Official monthly new-registrations file, keyless Socrata. The file is one row per associated name; only the entity row set (grouped) plus PRINCIPAL PLACE OF BUSINESS / MAILING ADDRESS rows are fetched for the city — member/president/agent rows are never selected. Includes ASSUMED BUSINESS NAME trade names, which face the entity-only gate.

Honesty rules (enforced by the pipeline, not prose)

• Entity-level only. Owner, officer, member, registered-agent and filer fields are never even fetched from the sources. Trade names shaped like a bare personal name are excluded and counted ({"OR": 1134} in this build), never stored.
• Every record carries its state and the state’s own registry identifier (entity_id) plus the official dataset URL it came from — verifiable by anyone on the state’s portal.
• No derived counts. records_in_this_file is exactly the number of records in that file. It is not “new businesses in Colorado this week” — the feed applies an entity gate and (NY) a formation-type filter, and states publish on their own schedules.
• Refuse-don’t-overwrite. A state whose portal changes format goes stale upstream and keeps its last-known-good records (status shipped per state in /v1/meta.json); if the store itself fails its floors the API is not re-rendered at all.
• Partial weeks are labeled. ISO weeks touching the edge of a state’s window carry week_complete: false.

Weeks

ISO 8601 weeks (Monday–Sunday), keyed YYYY-Www. A record’s week is the week of its registration_date as published by the state.

Name lookup keys and sharding

The exact key and the coarse phonetic key (TM-KEY v1) follow a published normalization rule applied to entity_name: see the full rule table. shard = fnv1a32(key) % 4096, lowercase 3-digit hex; all 4096 files exist in both trees so a missing key is an empty lookup in a present file, never a 404. Each key carries n, the true total in this feed, and at most 100 exact / 50 phonetic records, newest first. A hit means a business by that name registered in a covered state within its window — it says nothing about trademarks, availability or any other state.

Refresh

data_as_of is the stored collection date, not today's date or a promise of a new download. The covered registration dates differ by state; see the dated ranges. Oregon's monthly source covers an earlier period. Ongoing paid delivery is not currently available.