Sources

UTF-8 with BOM. 18 columns. Carries the registered office address and the min/max director counts, but NOT director names — the federal extract is materially less personal than the CRA charities extract.

Same 18 columns as the CBCA file. Governing legislation distinguishes the Not-for-profit Corporations Act, Cooperatives Act, Boards of Trade Act and the older Canada Corporations Act.

Larger than the active file. Status Detail distinguishes "Dissolved by the corporation" from "Inactive - Amalgamated", which is the difference between an entity that ended and one that continues under another name.

GET /api/v4/topic?page=1&page_size=2 returns total 2,808,571 — the API IS enumerable, contrary to what this entry said for two days. The blocker is licensing, not capability. The BC Data Catalogue record for "OrgBook BC API" gives license_id 22, "Access Only", owned by Digital Identity and Trust; the licence page states plainly that reproduction is not permitted without written permission and points to a Copyright Permission Request Form.

Probed via the Données Québec CKAN API (package_show). The publisher states license_id CC-BY-SA-NC-4.0, titled in full: Attribution, pas d’utilisation commerciale, partage dans les mêmes conditions. This registry previously recorded the licence as CC BY 4.0, which was wrong and would have led straight to ingesting it. The dataset itself is real and rich: a ZIP of several tables behind an ASP.NET download page, plus a PDF layout guide. Nothing was downloaded.

Fields confirmed on a live 2-row read: getbusid, tradename, homeoccind, address, comdistcd, comdistnm, licencetypes, first_iss_dt, exp_dt, jobstatusdesc, point. `homeoccind` = Y marks a HOME-BASED business, which is an explicit publisher flag that the address is residential — the only source in the registry that states this outright. robots.txt allows /resource with Crawl-delay 1.

Explore v2 API. `businessname` contains SOLE PROPRIETOR PERSONAL NAMES verbatim — the second record returned was "Leon Vance Michael Hiebert (Leon Hiebert)" trading as Skyzone Cabling Solutions. The individual/ organisation classifier is load-bearing on this source. Opendatasoft caps offset+limit at 10,000, so 205k records are unreachable by pagination and must be read through /exports/json.

package_show confirmed the CSV resource 169e90ba-3ae0-43dd-8b2f-919e87002f50 has datastore_active: true, so datastore_search works. CKAN answers 409 to ANY unknown parameter — never append Socrata-style $limit here.

Official version 1 ZIP verified and inspected. The 112,017,039-byte CSV contains 446,575 rows from 35 providers: 446,575 names, 394,537 addresses, 382,146 derived NAICS values, 128,344 employee fields and 299,157 licence numbers. Inputs were collected May–December 2022, so publisher status values are retained as historical categories rather than presented as current standing. Statistics Canada labels the release occasional; the annual cadence here is a conservative update check, not a claim that a new edition appears every year.

The id "9qtb-9xdd" once guessed here returned dataset.missing; the real id was found via the catalog search API ($q=business licence), not by guessing again. Edmonton uses its own bracketed redaction convention — `<Home Based Business>` and `<REDACTED FOR PRIVACY>` — which the shape must detect explicitly or it would publish the literal placeholder text as if it were a real address. Adult Services and Massage Practitioner rows redact name, address AND the licence id together at the source; the shape drops those rows entirely rather than working around the publisher's own privacy decision. Short-term-rental rows redact only the address, keeping a real name — the one case in the whole registry where a publisher discloses a business but deliberately withholds exactly where it operates.

datastore_search confirmed live. Fields: business_id, name, address, city, state, type, statut, date_statut, latitude, longitude, x, y. business_id is unique per row (1,000 distinct in a 1,000-row read). The `statut` vocabulary is five values and was counted in full: Fermé 20,318; Ouvert 16,840; Fermé changement d’exploitant 16,376; En traitement 601; Sous inspection fédérale 304. The third of those is a trap — it means the premises changed operator, not that the business failed, and a prefix match on "Fermé" would close 16,376 restaurants that are trading tonight. `city` is the borough or a demerged municipality (Côte Saint-Luc, Anjou), not always Montréal.

Fields confirmed live: TradeName, FullAddress, Unit, House, Street, LicenceNo, Category, LicenceFee, Description, IssuedDate, ExpiresYear, IssuedYear, Covers_FROM, Covers_TO, gislink, ObjectId. maxRecordCount 1000, supportsPagination true. ONE ROW PER LICENCE YEAR and LicenceNo CHANGES between years — the same business appears as 40784 and 40447 — so the licence number cannot anchor a business. FullAddress embeds a CRLF and a truncated postal ("VICTORIA BC V8"); mobile vendors carry the literal street "BUSINESS - MOBILE". Licence text read in a browser because the page is client-rendered: it is OGL-BC v2.0 reworded for the city, granting use "including for commercial purposes", with exemption 6(a) for Personal Information.

Fields confirmed live: Licence_Account_Number, Name, Type_Code, Type_Description, Business_Description, Licence_Location, Within_City_of_Kelowna, ObjectId. maxRecordCount 1000, supportsPagination true. Unlike Victoria, the account number IS stable per business and anchors it directly. There is no date column of any kind — the file is a current snapshot, so a Kelowna licence can say that a business trades but never since when. `Within_City_of_Kelowna` distinguishes resident from non-resident licence holders. Licence text extracted from the publisher’s PDF: OGL-Canada v2.0 reworded for the city, "including for commercial purposes".

A MapServer layer rather than a FeatureServer — the /query endpoint is identical, so the arcgis adapter reads it unchanged. maxRecordCount 2000, supportsPagination true. Fields: OBJECTID, UNIT, PROPERTY_NUMBER, LICENCE_NUMBER, TRADE_NAME, LICENCE_TYPE_NAME, COVERS_FROM, COVERS_TO, ACCOUNT_NAME, HOUSE, STREET, LEGACY_LICENCE_NUMBER, LEGAL_TYPE, LICENCE_STATUS. 1,985 distinct LICENCE_NUMBER in a 2,000-row sample, so it is one row per licence and anchors directly. COVERS_FROM/COVERS_TO are EPOCH MILLISECONDS, not dates. ACCOUNT_NAME holds the licence holder and is frequently a private individual — "ALBERT WARKENTIN" trading as BA TRUCKING — and the licence's own exemption 6(a) grants no right to use Personal Information, so the shape drops it. "BUSINESS - NON RESIDENT" appears in STREET as a marker for out-of-city contractors, not as an address.

THE MOST PERSONAL MUNICIPAL FILE IN THIS REGISTRY. It publishes COL_BUSINESSPHONE and EMAILADDRESS in the clear, and the addresses are personal ones — "s.almasi@me.com" was the second row read. Neither column is carried past the shape at all, not even to internalText: the licence's exemption 6(a) grants no right to use Personal Information, and an email has no value for entity resolution to weigh against holding it. LAT and LONG ARE MISLABELLED — they hold Web Mercator metres (6319331, -13679963), not degrees; converted they give -122.889, 49.261, which is Coquitlam. COL_FOLDER is unique per row but year-prefixed ("26 116374 00"), so anchoring on it would make every business vanish and reappear each year in the change layer. Statuses seen: Issued, Renewal. maxRecordCount 1000. On 2026-10-25 the layer reported 84,851 raw rows, but the same licence assertions were being republished under new OBJECTIDs. The 2026-10-15 capture contained 18,073 rows and normalized to 6,873 distinct assertions; Bronze retains every publisher row while Silver suppresses only byte-equivalent normalized repeats.

Fields: OBJECTID, BusinessID, Unit, House, Street, PostalCode, BusinessName, BusinessType, BusinessSubType, BusinessDescription, IssueDate, LicenceCategory, WebsiteAddress, Phone1, Phone1Description, Keywords. BusinessID is stable and anchors directly. IssueDate is EPOCH MILLISECONDS. Two things no other BC source here has: a FULL POSTAL CODE, which is what lets a municipal licence merge onto a federal corporation, and `LicenceCategory` = "HOME BASED", an explicit residential flag. Phone1 is present and never carried. The portal's own licence page requires a sign-in, so the terms were read from the City's Terms of Use page instead: OGL-BC v2.0 reworded for Maple Ridge, "including for commercial purposes", with the City's Terms of Use added as also applying.

The richest municipal licence file in the registry. Fields: PropertyNumber, Business_Name, Civic_Unit, Civic_House, Civic_Street, Mailing_Address, GISLink, CommunityName, Phone, Licence_Desc, Category, Category_Desc, Detail, Detail_Desc, NAICS_Primary, NAICS_Primary_Desc, Approved_Date, Permitted_Use, Permitted_Use_Desc, Seasonal_Employees, PartTime_Employees, FullTime_Employees, Latitude, Longitude, FID. THREE SEPARATE EMPLOYEE COUNTS — full-time, part-time and seasonal — which makes this only the second source here that says how big a business is, after the Montréal supplier list. Also a NAICS code, which no source in the registry carried before this batch. The object-id field is FID, not OBJECTID. Phone is present and never carried.

The city publishes six overlapping layers — All, Residents, Nonresidents, Intermunicipal, Childcare and New This Year. The ALL layer is used, and 3,844 is exactly Residents (2,729) plus Nonresidents (1,115), so the others are subsets and ingesting more than one would count the same business twice. LICENCEE_NAME is a PERSON on a large share of rows and is written surname-first — "RO, KYOUNG S", "AIT TALEB, AHMAD" — and is never carried. CIVIC_ADDRESS uses "BUSINESS - OUT OF TOWN" as a marker for non-resident holders and embeds a CRLF. Carries NAICS_CODE and NAICS_DESCRIPTION, and YEAR_OPENED, which is the year the business started rather than the year of this licence.

Fields: LicenceNumber, DateFrom, DateTo, TradeName, LicenceDesc, LicenceCategory, Unit, Address, StreeName, ESRI_OID. The street field is spelled "StreeName" in the publisher's schema — a typo in their column name, not in this note, and reading "StreetName" returns nothing. The object-id field is ESRI_OID. Licence text extracted from the city's PDF: OGL-BC v2.0 reworded for Prince George, "including for commercial purposes".

The district publishes an annual snapshot for every year from 2018 to 2025 AND a monthly current file, and two of the "annual" datasets point at the monthly layer, so ingesting them all would multiply the same businesses eight ways. Only the monthly layer is used. Fields: OBJECTID, GIS_ID, Roll_Number, Business_Licence, Business_Name, Business_Location, NAICS, Date_Opened, NonResident, Licence_Status, Licence_Type, Licence_Code, Licence_Description, Licence_Conditions, NAICS_Sector, NAICS_Description, Squamish_Sector, Squamish_Subsector, Neighbourhood. Date_Opened is EPOCH MILLISECONDS and reaches back to 1990. Licence_Type "HOME OCCUPATION, OFFICE" is an explicit residential flag.

A plain CSV at a stable URL, so no adapter was needed. 120,639 rows covering 2007-2026 — the deepest municipal history in the registry — holding 17,694 distinct licence numbers, which ARE stable across renewals and anchor directly. Columns: TRADE_NAME, LICENCE_NUMBER, ADDRESS, NEIGHBOURHOOD, COVERAGE_YEAR, LICENSE_TYPE, RENEWAL_STATUS, LICENSE_STATUS, ACCOUNT_STATUS. 39.2% OF ROWS CARRY A CATEGORY WHERE THE NAME SHOULD BE: HOME OCCUPATION (38,583), NON-RES HAWKER/HUCKSTER (3,758), TAXI OPERATOR (3,578), RESIDENT HAWKER/HUCKSTER (1,809), NON-RESIDENT VENDOR (20) — the city withholding the trading name of home-based and itinerant licensees. 8,618 real names remain. THE LICENCE IS WORDED DIFFERENTLY FROM EVERY OTHER ONE HERE: it grants a "worldwide, royalty-free, non-exclusive license... for any lawful purpose" and does NOT contain the phrase "including for commercial purposes" that the OGL variants carry, nor the word "perpetual". There is no non-commercial clause and no share-alike clause anywhere in it, and commercial use is a lawful purpose, so the grant covers it — but the difference is recorded here rather than smoothed over, and the licence name is added to the allowlist in licence.test.ts as a deliberate act with this reasoning attached.

THE ONLY MUNICIPAL LICENCE FILE IN THIS REGISTRY WITH NO PERSONAL DATA IN IT AT ALL. The complete field set is OBJECTID, LICENSE_NUMBER, LICENSE_TYPE, SUB_DESCRIPTION, BUSINESS_NAME and EXPIRY_DATE — no licensee, no holder, no contact, no address. Every other city here publishes at least one column that has to be dropped. EXPIRY_DATE is epoch milliseconds. Layer 1 of the service, not layer 0. Licence read at the publisher: "worldwide, royalty-free, perpetual, non-exclusive licence... including for commercial purposes", exempting personal information under MFIPPA.

THE RECORD COUNT IS MISLEADING AND THE SHAPE KNOWS IT. 13,757 rows, but only 275 distinct trade names, and 10,976 rows (80%) are "License - Rooming Houses": dwellings, with trade_name set to the literal "(none)" and the address column ABSENT ENTIRELY because the publisher withheld it. The usable remainder is Personal Sales (1,665) and Personal Services (1,051), one row per licence YEAR, so a tattoo studio licensed since 2010 appears fifteen times. Real yield is a few hundred businesses. Socrata licenseId is OGL_CANADA, verified through the views API rather than assumed from the portal footer. robots.txt sets Crawl-delay: 1 and disallows only /browse query strings; /resource is allowed.

The download 302s to opencanada.blob.core.windows.net with a SIGNED, EXPIRING SAS URL (se=/sig= query parameters). The signed URL must never be cached in this registry — the redirect is followed at fetch time. Free-text description fields make this pii: free-text.

2.3 GB — the largest file in the registry, and the reason the whole read path is streaming. Recipients include INDIVIDUALS as well as organisations, so the individual/organisation classifier is mandatory here, not optional.

Fields confirmed live: permitnum, statuscurrent, applieddate, issueddate, completeddate, permittype, permitclass, permitclassgroup, permitclassmapped, workclass, description, applicantname, contractorname, housingunits, estprojectcost, totalsqft, originaladdress, communityname, latitude, longitude, point. 12,147 distinct contractors; JAYMAN BUILT leads with 12,440 permits and $2.83B of estimated project cost. 199,029 rows carry NO contractor at all (Socrata omits null fields entirely) and the shape drops them. LICENCE WAS CHECKED RATHER THAN ASSUMED: the Socrata licenseId is the unhelpful "SEE_TERMS_OF_USE", not the OGL string the business-licence dataset carries, and the terms had to be read from the story page — they are the standard OGL text, granting a "worldwide, royalty-free, perpetual, non-exclusive license to use the Information, including for commercial purposes". Exemption 6(a) grants no right to use Personal Information, which is the licence itself pointing at the publishing gate.

Fields: id_poursuite, business_id, date, description, adresse, date_jugement, etablissement, montant, proprietaire, ville, statut, date_statut, categorie. `business_id` is the SAME namespace as the establishments file, so an offence attaches to its premises by anchor equality rather than by name matching. `proprietaire` mixes numbered companies (9068-9811 QUEBEC INC.) with sole proprietors’ personal names and is never carried past Silver. `date` is the offence and `date_jugement` the ruling; only the ruling dates the record.

Resource listed live on the 2024 package. The charity registration number is a BN15 (nine-digit BN + RR0001), so its first nine digits join directly to the federal corporations BN.

This entry previously pointed at financial_section_a_b_and_c, which holds program codes rather than the financial statement — the wrong resource for what the entry claimed. Section D is financial_d_and_schedule_6_2024_updated.csv. Columns are bare T3010 line numbers and were mapped from CRA guide T4033 rather than guessed: 4700 total revenue, 4950 total expenditures, 5000 charitable programs, 5010 management and administration, 5020 fundraising, 5050 gifts to qualified donees, 4200 total assets, 4350 total liabilities. Blank is not zero and is not read as one.

Live JSON:API query filtered on entity.legalAddress.country=CA. The first page reported Golden Copy publishDate 2026-10-15 and 59,158 Canadian records. Deep numbered pagination fails at 10,000 even though the API advertises later page links; complete reads must begin with page[cursor]=* and follow cursor links. LEI reference data is CC0. The payload is nested and carries legal name, legal address, entity status, creation date, legal form, registration authority, local registry identifier, LEI renewal status and corroboration level. RA000072 was verified through GLEIF’s registration-authority endpoint as Corporations Canada, allowing its registeredAs value to attach by exact federal corporation number instead of by name.

license_id confirmed as ca-ogl-lgo via the Open Government Portal API directly, not the dataset's own landing page, which states no licence at all. 19 columns; no dollar value field exists anywhere in the file — this is qualification status, never money. `commodity` and `commodity-description_en` are double-quoted BY THE PUBLISHER inside an already-quoted CSV field, so correct RFC 4180 parsing still leaves a literal leading and trailing quote character in the value; the shape strips it. No business number, corporation number, or address of any kind — attachment to an entity is by supplier name only.

Fields: Numero_Fournisseur, Nom_Fournisseur, Autre_Nom, Num_Entreprise_Quebec, Structure_Juridique, Nombre_Employes, Nombre_Employes_Montreal, Politique_Dans_Dev_Durable, Politique_Approv_Responsable, Entreprise_Economie_Sociale, Lieu_Arrondissement, Lieu_Ville, Lieu_Etat_Province, Lieu_Pays, Lieu_Zone_Achat. `Autre_Nom` is NOT an alternate name on most rows — the publisher reuses the column to repeat the NEQ as "NEQ:1169283323". Structure_Juridique is prefixed with an asterisk.

BN, FPE, Form ID, #, Last Name, First Name, Initials, Position, "At Arm's Length", Start Date, End Date — 11 columns, confirmed against a live sample. Carries NO address, phone or email of any kind; the narrowest possible slate. Position is free-typed by the filer (seen: "Director", "Director/Chair", "Treasurer/Director", "Directoor" — a literal typo in the live data), so it is stored verbatim, never parsed.

Official RR 2.1 Golden Copy discovered through GLEIF’s current-file API. The 16:00 UTC release contains 491,100 global relationship rows. Exactly 12,825 rows begin at one of the 59,158 Canadian LEIs in the current Level 1 capture: 2,848 direct accounting-consolidation parents, 3,074 ultimate accounting-consolidation parents, 6,761 fund-manager links and 142 other typed fund or branch relationships. 10,366 targets also carry a Canadian LEI; 2,459 point to a foreign entity. Joins use the child and target LEIs exactly — never company names.