The data layer
A newsroom can only cite what it can check. This is every source behind Watch, how each number is produced, what it does not cover, and how to reference it in a story.
Sources
| Dataset | Source | Cadence | Coverage |
|---|---|---|---|
| Titles, artwork, cast | TMDB | Weekly | 24,096 titles carrying any US offer — subscription, rent or buy |
| Streaming availability | JustWatch, via the TMDB API | Weekly | US only, 157 on-demand services |
| Charts | Netflix’s published Top 10 TSV | Weekly, Tuesdays | US films and TV, 10 each |
| Game schedules | ESPN public scoreboard | Daily | NFL and college football |
| Network → service carriage | Ours, hand-maintained | Reviewed 2026-08-27 | 29 networks, 16 carriers |
| Collections | Ours, computed | On re-cluster | 114 clusters over the catalogue |
Method
Availability. Every title is queried against TMDB’s
watch/providers endpoint, which returns JustWatch’s US offer data.
Provider names are canonicalised — TMDB returns Paramount+ under four spellings and
Peacock, AMC+ and MGM+ under two each — then live-TV bundles are separated from
on-demand libraries, because a channel line-up answers a different question than a library.
Change detection. The feed is a snapshot with no arrival or expiry dates, so change is derived by diffing consecutive refreshes into an event log. That is the only honest route to “new this week” from this source.
Collections. Each description is embedded into a 384-dimension space, clustered, and each cluster named by a language model. They are not genres; they are what the corpus actually grouped into.
Sports. ESPN supplies the game and its broadcaster, including local affiliates. We map that network to the services carrying it using our own table.
Known limits
Machine-readable
The catalogue index that powers search and personalisation is a single JSON document, and the derived datasets are stable paths:
/data.json— services, collections and every title with its slug, year, type, services and artwork/charts.json— the current Top 10 with source and week/sports.json— upcoming games, networks, resolved services and carriers/sitemap.xml— every indexable page
Advertising and neutrality
Two kinds of unit appear. House promotions for this product, and sponsored units carrying a real service’s public price and a link to its own signup page. Nothing invents a campaign or reuses a brand’s creative, no affiliate tag is attached, and no commercial relationship exists — they show what the inventory would look like.
Ordering is not for sale. Providers on a title page are sorted by tier — subscription, then rent, then buy — and alphabetically inside each tier. That order is computed before any sponsorship state is known and cannot be influenced by it. Sponsored placements only ever appear in visually distinct, labelled containers, never inside the availability list.
What is stored about you
Your selected services, your saved list, and a capped local log of what you searched and which titles you wanted but could not watch. All of it lives in your browser’s own storage, none of it is transmitted, and there is no account, cookie or third-party tracker. Clearing site data removes it entirely.
How to cite it
Always cite the retrieval date. Availability changes constantly, and a claim about where something streams is only true as of a date.