# TasteBranch data sources

Last updated: 2026-11-15. Status: early preview. The data set is not complete yet.

## How to read this file

- **Verified** means checked in the build session against the source, or against recorded web search results.
- The build session could not open any outside web page directly, except files on `raw.githubusercontent.com`. Web fetch was blocked for every other host. Facts and links were therefore verified by web search. Each search (query and result links) is saved in the project repository under `research/evidence/`.

## Sources

| Source | Licence | Page where the licence was read | What was taken | Status |
|---|---|---|---|---|
| Natural Earth, 1:110m countries and 1:10m populated places | Public domain | `LICENSE.md` in https://github.com/nvkelso/natural-earth-vector (opened directly; it carries the Natural Earth terms of use and links to https://www.naturalearthdata.com/about/terms-of-use/) | Country outlines, UN subregion names, city names, city coordinates, country names. Used for `data/world.js` and for the coordinates of each node's origin city. | Read directly. |
| English Wikipedia | Text is CC BY-SA 4.0 and GFDL. Commercial reuse is allowed under those terms (attribution, share-alike). | https://en.wikipedia.org/wiki/Wikipedia:Copyrights | Facts only, as shown in web search results: titles, makers, years, places, and coordinates from infoboxes. Article addresses are used as links and as the named source of a bridge or cross-link. No sentence is shown in the app; every sentence in the app is written for TasteBranch. Short quotes from search result text are kept in the project repository only, as proof for each fact. | Licence read through web search on 2026-11-14 (page not opened directly). |
| Wikidata | CC0 for the main, Property, Lexeme and EntitySchema namespaces. Text in other namespaces is CC BY-SA 4.0. | https://www.wikidata.org/wiki/Wikidata:Licensing | Nothing was taken directly. Wikidata could not be reached. Some coordinates shown by Wikipedia infoboxes are also held in Wikidata. | Licence read through web search on 2026-11-14. |
| MusicBrainz | Core data is CC0. Supplementary data (annotations, tags including genres, ratings, statistics, edit history) is CC BY-NC-SA 3.0, which is non-commercial, so it must not be used. | https://musicbrainz.org/doc/About/Data_License | Nothing taken. MusicBrainz could not be reached. | Licence read through web search on 2026-11-14. |
| OpenStreetMap | ODbL. The credit "© OpenStreetMap contributors" and a link to the licence are required when a work made from the data is shown in public. | https://osmfoundation.org/wiki/Licence/Attribution_Guidelines and https://wiki.openstreetmap.org/wiki/Legal_FAQ (the page https://www.openstreetmap.org/copyright did not come up in the search) | Map links for San Francisco places open openstreetmap.org at the place's coordinates. The footer carries the credit and the licence link. Planned: coordinates for places that have none on Wikipedia, looked up with `tools/geocode-osm.js` (OpenStreetMap Nominatim). Nothing taken yet. | Licence read through web search on 2026-11-14. |
| Foursquare Open Source Places | Apache 2.0 | https://docs.foursquare.com/data-products/docs/fsq-places-open-source | Nothing. The data set is gated and could not be reached, so it was not used. | Licence read through web search on 2026-11-14. Not used. |
| Official sites of venues and makers | Not a data set | Not applicable | Facts only: name, street address, and proof that the place operates in 2026. Site addresses are used as links. Nothing else was copied. | Verified by web search, item by item. |
| Fraunces typeface (Google Fonts) | SIL Open Font License 1.1 | https://www.fontsquirrel.com/license/fraunces (copy of the licence); font page https://fonts.google.com/specimen/Fraunces/about | The font is loaded from Google Fonts when the page opens. | Licence read through web search on 2026-11-14. |

## Coordinates of San Francisco places

Coordinates are accepted only from Wikipedia or Wikidata, from OpenStreetMap, or from the place's own page. Coordinates that a search found only on another map listing page (for example MapQuest) are not used; those places are held back until an OpenStreetMap lookup gives their coordinates. Proof that a place is open is never taken from Yelp or another review site.

## Sources that are not used

MovieLens, Yelp Open Dataset, Amazon Reviews, Goodreads dumps, IMDb, and any source marked research-only or non-commercial. Luma, Partiful and Eventbrite were not scraped and are not linked. Researchers were told not to take data from Yelp or TripAdvisor.

## How a bridge and a cross-link are verified

A bridge is a link between a place and a taste area. It is shown only when it passed two tests.

1. A sceptic read the bridge and every web search the researcher ran, with no new searches and no facts from memory. It kept the bridge only when every claim in the linking fact is in the recorded result text. It corrected or removed any part the text does not support.
2. The build then checked by program that the source link is in the result links of a recorded search, and that the supporting quote is in the result text of a recorded search.

The same two tests apply to every cross-link between two steps. Where the recorded text does not name the city at the far end of a bridge, the map shows the country only.

The file `research/verification-bridges.md` in the project repository lists every bridge and cross-link with its source link, the search that showed it and the quote. No page was opened directly; web fetch was blocked for every outside host, so the proof is the recorded search result.

## How each item was verified

The file `research/verification.md` in the project repository lists every node and every San Francisco place with the search that showed its link, the evidence that the place is open in 2026, and where its coordinates came from. It also lists every dropped item and the reason.

## Images

There are no photos of people and no images taken from other sites. All drawings (tree, map, bridge) are made by the app.