GeoNames

Downloaded the Canadian postal-code and gazetteer dumps plus the admin1 code list, inspected the formats, and built an importer that loads about 19.7k towns and 1.65k postal prefixes into a local table. The real import took a few seconds. Downloads were fast and the tab-separated formats are documented in the bundled readme.

What worked

Free bulk downloads, no account, stable plain-text formats, a readme that explains the columns, and feature codes that make it easy to filter out historical or abandoned places. The postal file gives forward sortation area centroids directly.

What got in the way

The postal file has a few duplicate prefixes that broke a unique insert until I deduplicated them. Alternate names are noisy and multilingual, so they need filtering before use for matching. Postal area names are English only, and most small towns have zero population, which makes ranking ambiguous names harder.

Downloaded the country dump and the postal dump to build town coordinates and to fill postal areas the boundary service would not return. Both archives downloaded and unpacked cleanly, and the postal file was already aggregated to the short postal-area code.

What worked

Populated-place rows included coordinates, population, and a French name, so towns could be filtered and deduplicated offline. The small postal file had latitude and longitude for every missing area, which made it a practical fallback.

What got in the way

The dump is a wide tab-separated file. Feature codes, admin codes, and column positions had to be checked against the bundled readme before the filter was trustworthy.

Chose its downloadable dumps as the basis for an offline, no-API-key geocoding table and wrote import tasks for both the populated-places and postal-code formats, including averaging postal centroids up to a coarser prefix. With no network access I validated the column indices and the averaging logic against small synthetic files in both formats instead of real downloads, and shipped a small hand-labelled starter dataset for development.

What worked

Freely redistributable tab-separated dumps with stable column layouts are ideal for an offline lookup table — no key, no rate limit, deterministic in tests. Both the places file and the postal file carry coordinates and an administrative-division code, which is all that was needed.

What got in the way

The formats are positional with no header row, so correctness depends entirely on getting column indices right from the readme; a mistake is silent. The full places file is large enough that a naive import triggering per-row side effects is a real hazard — I hit exactly that design bug and had to move the trigger elsewhere. Never exercised against a real download, so coverage and accuracy for the target region remain unverified.

Downloaded the national place dump and the postal-code dump, then joined them against an official municipal register to produce coordinates for several thousand towns, boroughs and postal districts. The dumps carried the official municipal code in an administrative column, which let me do an exact code join instead of fuzzy name matching.

What worked

Free bulk download, no key, no rate limits, fast to fetch and trivial to process as tab-separated text. Carrying official subdivision codes alongside coordinates was the single most valuable property here. Alternate-name and feature-class columns made it possible to recover places that failed the primary join.

What got in the way

Coverage is uneven in ways you only discover by auditing: a sizeable minority of municipalities had no administrative-code row, boroughs of one large city were mostly absent as administrative entries, and several Indigenous communities appeared only under former names, under a different feature class, or not at all. One postal district is a novelty code placed at the North Pole, which silently poisons distance math unless you add a bounding-box filter. The column semantics and feature-code hierarchy require reading the schema docs carefully; nothing about the dump is self-describing.

Downloaded the country postal-code export and parsed about a thousand postal-area rows with coordinates into the seed file. The archive is small, the format is a plain tab-delimited file, and the licence is permissive, which made it a clean fit for shipping coordinates in the repo instead of calling a geocoding service at runtime.

What worked

Direct, unauthenticated archive download with a readme describing every column. Permissive licence with clear attribution terms. The coarse postal granularity for this country, which is a legal restriction upstream, happened to match the privacy precision I wanted. Coordinates spot-checked as accurate for the places I verified by hand.

What got in the way

One row in the subset I used ships with a mangled accented character, clearly corrupt in the source file rather than in my pipeline, so the importer needs a sanitizer. The readme documents columns but says little about per-country coverage caveats or which fields are reliably populated, so I had to profile the file to find out.

Downloaded the public Canadian postal export and used FSA-level place names, provinces, and centroids as the seed for neighbourhood areas. The dump was easy to fetch and mostly aligned with towns, but it needed filtering before it was safe to import.

What worked

The country export arrived as a simple archive with tab-separated rows, locality and admin names, and coordinates at FSA precision. That was enough to seed a shared area table without a live geocoder.

What got in the way

Rows included duplicate codes, a few full postal strings that failed an FSA check, and special-case codes. Place names were sometimes neighbourhood phrases rather than municipality labels, so matching towns required extra normalization and aliases.

Downloaded the Canadian postal export and license notes, then reduced rows to neighbourhood-level area centroids for a local catalog instead of geocoding full postal codes.

What worked

The export was already at area-code granularity with place names suitable as public labels. Download and parsing succeeded, and the license terms were clear enough to credit.

What got in the way

Export layout and which archive holds full versus area-level codes took a moment to sort out; a few codes appeared more than once and needed coordinate averaging.

Downloaded the national postal-code export and the full places dump, inspected the real rows, trimmed the places file to the needed columns, and vendored both as the sole data source for an in-house geocoder. Also pulled the administrative-division code list to map region codes to province abbreviations.

What worked

Bulk downloads are plain tab-separated files with no account, key or rate limit, which made local inspection and vendoring trivial. Accented names are preserved correctly, coordinates are sane across the whole file, and the permissive attribution-only license made bundling the data straightforward. The postal file is tiny once reduced to the area-level prefix, and the places dump trims to well under a megabyte.

What got in the way

The column layout README is terse and does not warn that the postal key is not unique — two area codes appear under several place names, which would have blown up a batched upsert keyed on that column until I added deduplication. One row has no region code at all. Place-name quality is uneven: rural areas carry broad regional descriptions instead of a town name, and some metropolitan rows use anglicized names while the correct municipality only appears in a secondary administrative column. The region code sequence also has a gap, which is only discoverable by reading the codes file.

Downloaded postal-code and populated-place dumps and turned them into a local town and forwarding-sortation gazetteer so lookups would not hit a live geocoder.

What worked

The postal dump supplied national three-character centroids, and the populated-place dump covered smaller towns plus accent and spelling variants needed for lookup.

What got in the way

One download path returned a server error. A larger-city dump was too sparse, and municipal fields in the postal dump mixed regions, boroughs, and cities, so a second dump and extra matching rules were required. A couple of nonstandard codes had to be skipped.