# Closed Overture places in 31 cities, November 2026

How many Overture places have closed? This checks every named place in Overture's 2026-11-18.0 release inside 31 cities in eight countries against official registers that keep a business on file after it closes, and puts OSM venues through the same check for comparison. The blog post [A third of Overture places we checked had closed](https://mapricot.com/blog/overture-places-closed/) reports the results, and [`results.json`](results.json) has every number. It extends [the October check](../overture-2026-10/README.md#closed-places), which covered San Francisco and Chicago and led the export to drop Foursquare-only and Microsoft-only places.

## Method

- **Places.** Every named Overture place in each city's box ([`cities.py`](cities.py), [`extract.py`](extract.py)), 4.4 million in all. Places Overture marks `permanently_closed` are counted on their own, not with the rest. A place's source is its record's datasets joined, which in this release is always one. OSM venues are named nodes and ways tagged `amenity`, `shop`, `tourism`, `leisure`, `craft`, `office`, or `healthcare`, from BBBike's and openstreetmap.fr's extracts, downloaded on 20 November 2026 ([`osm.py`](osm.py)).
- **Registers** ([`fetch.py`](fetch.py), [`prepare.py`](prepare.py), [`registers.py`](registers.py)), each an official list that keeps closed businesses:

  | Cities | Register | Closed means |
  | --- | --- | --- |
  | San Francisco | Registered business locations | a location or business end date, or closed by the city |
  | Chicago | Business licences | cancelled or revoked, or expired over 120 days before the release |
  | Chicago (check) | Food inspections, by the health department | last inspection found it out of business; open means inspected in the last year |
  | Philadelphia | Business licences (not rentals) | closed or revoked, or inactive or expired over 120 days before |
  | New York | DCWP premises licences (tobacco, electronics, laundries, garages, and so on) | closed, surrendered, revoked, or out of business, or expired over 120 days before |
  | Washington, DC | Basic business licences | closed, cancelled, revoked, or inactive, or expired over 120 days before |
  | Vancouver | Business licences, 2024 on | no licence issued for 2025 or 2026 |
  | Toronto | Business licences and permits (premises only) | a cancel date |
  | 12 French cities | SIRENE establishments with INSEE's geocoding | closed (état F), on its date |
  | 4 Mexican cities | INEGI's DENUE, November 2019 and May 2026 | in 2019, gone by 2026 |
  | 6 Brazilian cities | Receita Federal's CNPJ establishments (trade names) | baixada, on its date |
  | Singapore | ACRA's entities | deregistered (no date given) |
  | Melbourne | City of Melbourne's yearly business census, 2012–2024 | missing from the 2024 census |

- **Matching** ([`measure.py`](measure.py), [`names.py`](names.py)). A place matches a register entry with the same name within 150 m, or in the same postcode where the register has no points (Toronto, Singapore, Brazil). "Same name" means the same words in any order, or the same letters without spaces, after dropping accents, punctuation, legal forms (Inc, SARL, Ltda), and articles. A looser rule, the October check's, also counts names that share most of their words. It matches more places to more entries, so more of them match at least one open entry, and it lowers the medians by about three points, for Overture and OSM alike. The results use the strict rule, and `results.json` has both.
- **Closed.** A matched place has closed when every register entry it matches had closed by the release, 18 November 2026. An entry that closed after the release counts as open, and so does a place matching both a closed and an open entry, such as a shop that changed owners. "Closed over a year" needs every closing date to be before 18 November 2025.
- **Across cities.** The share closed is of matched places. The summary is the median and mean of the city shares, each city counting once, with a 95% interval for the mean from 10,000 bootstrap resamples of the cities. City shares carry 95% Wilson intervals. Groups with fewer than 30 matched places in a city are left out of that city's summary.

## Checks

- **Independent register.** Chicago's food inspections and its licences are kept by different departments. On the 7,734 places both can check, they agree on 90%, and 91% of the places the licences call closed also had an out-of-business last inspection.
- **By hand.** Eleven randomly drawn places the method called closed, in San Francisco, Philadelphia, Toronto, and Melbourne, were looked up on the web. Three were confirmed closed, one looked open, and seven couldn't be settled, since listings rarely say a place has closed. That's too few to measure anything, so the independent register is the check that counts.

## Caveats

- Only places a register can match are counted: about a third of each French city's Overture places, but 1% in New York, whose register covers only a few trades, and 4–21% in the other cities. Matched places lean towards registered storefronts. Unmatched ones may be open, closed, or never real.
- Registers record legal and licensed names, which can differ from a storefront's name, and their points are geocoded addresses.
- How far back a register goes sets the level. Vancouver's open data starts in 2024, so few closures show there, for Overture and OSM alike. That's why OSM goes through the same check in each city: compare a city's Overture share with its OSM share, not with other cities.
- Singapore's register is of companies at their registered address, and Mexico's compares two snapshots of a directory, not closing dates.

## Rerun

In an empty directory, with DuckDB, pandas, pyarrow, pyosmium, and pyproj installed (`pip install duckdb pandas pyarrow osmium pyproj`), run these from this directory's path. It takes about two hours, most of it downloading.

```sh
python3 fetch.py                       # registers and OSM extracts, about 12 GB
python3 prepare.py fr; python3 prepare.py mx; python3 prepare.py br
python3 extract.py 2026-11-18.0        # Overture places from S3
python3 osm.py pbf/Paris.osm.pbf paris # once per extract, naming the cities in it
python3 measure.py san-francisco chicago chicago-food ...   # every city in cities.py
python3 summarize.py                   # results.json
```
