Method

Each day, English Wikipedia’s most-read list supplies around 45 concepts. Each is resolved through Wikidata into its article title in 36 language editions, and each of those articles’ views for that day are fetched from the Wikimedia Pageviews API.

For every concept, in every edition:

share = article views in edition ÷ that edition’s total views that day

asymmetry = highest qualifying share ÷ share in English

Concepts are ranked by asymmetry, then by the peak edition’s absolute views — so a large peak outranks a small one at a similar ratio.

Pageviews carry no country dimension. A Swede reading the English article is counted in English Wikipedia, invisibly. So this site never claims that Sweden read something — it claims that Swedish Wikipedia did, measured against its own daily traffic. That statement is exactly true and requires no inference about who the readers were or where they live.

A corollary: English Wikipedia is a global edition, not an Anglophone one. Traffic from everywhere lands there, which dilutes its share and is part of why English share is consistently low.

This is also why there are no maps, flags or country names anywhere on this site. They would reintroduce exactly the inference the method rules out.

The daily concepts come from English Wikipedia’s most-read list. Stories that never charted in English are out of scope — which is a real limitation, not an oversight.

The narrowing is what makes the comparison possible. A survey of eight editions found only 26 of 313 charting concepts appeared in two or more editions, so starting from every edition’s own list would spend most of the effort on articles that can never have a cross-language comparison. Leading with English guarantees every row has a comparison and a denominator.

The cost is concrete. On 9 October 2026, Japanese Wikipedia spent 1.13% of its entire day — 307,768 views — on a person with no article in any other edition. That story cannot appear here.

A figure counts toward the metric only if the article drew at least 2,000 views in that edition, and the floor applies to English too. Without it the ratios become noise: on one sampled day an actor drew 16,970 Spanish views against 710 English ones, producing a meaningless 238×.

Two biases pull in opposite directions, and neither is hidden:

• Share-of-edition rewards small editions. A 93,000-views-a-day edition is easier to dominate than a 194-million one. This is why edition size sits next to every share on the concept page rather than being tucked away.
• The floor pushes the other way. Requiring 2,000 views means a small edition must spend around 2% of its entire day on one article to qualify, while a large edition clears it trivially.

The two partly cancel. Figures below the floor are still shown, marked with an asterisk, so you can see the number without it driving a label.

• Pageviews are not interest. A link from a popular page inflates counts, and a reader who opens an article and leaves counts the same as one who reads it.
• Redirects split counts. Views arriving via a redirect title are not enumerated here, so every figure is a mild undercount — more so for concepts known by several names.
• Bot filtering is good, not perfect. All figures use the API’s agent=user filter, which excludes known bots and spiders. Undetected automated traffic remains.
• Edition size is not population. No per-capita claim is made or possible.
• Absence is a finding. When an edition has no article for a concept it is counted as absent and shown as such, never as a zero. A zero here means the article exists and recorded no views.

Wikimedia publishes a day’s pageviews about a day after it ends, so the newest complete day is yesterday. Every page names the date it is showing rather than implying it is live.

A scheduled job fetches one day each morning and commits it. Nothing is fetched when you load a page: every figure comes from a snapshot stored with the site and read when it was built.

A concept’s shape label comes from counting how many editions over-index — reach at least twice English’s share while clearing the floor:

• under 2× asymmetry — broadly global
• 2 or fewer over-indexed — single epicentre
• 5 or fewer — regional cluster
• more than 5 — widely over-indexed

That over-indexed count is shown on every card as an evidence count, because it says how much the label rests on: a 4× backed by one edition and a 4× backed by nine are different claims.

The label describes shape, never cause. Explaining why an edition spiked would need a model or a news feed, and both would make the figures unreproducible. No language model is used anywhere in this project — the labels, the ranking and this page are all arithmetic and hand-written text.

• Wikimedia Pageviews API — per-article daily views and per-edition daily totals. Every measured number on this site comes from here.
• Wikimedia Feed API — English Wikipedia’s most-read list, used to identify each day’s concepts. Its own view counts are not used: they disagree with the Pageviews API by as much as a factor of five, so mixing them would put two definitions of “a day’s views” in one row.
• Wikidata — resolves one concept to its article title in every edition, so no searching or guessing is involved.

Wikipedia content is available under CC BY-SA 4.0. This site reproduces article titles and view counts only.