Methodology

The Fad Index

Each subject gets a monthly series of English Wikipedia pageviews from the Wikimedia REST API. Pageviews measure curiosity, not purchases. We normalize nothing across subjects; the index is a set of comparable series, not a single score.

Lifecycle stages are computed from the 12-week slope: Rising (above +15% per month), Cooling (below −15%), Peak (within 10% of the trailing 12-month maximum), Dead (below 25% of peak for 12+ weeks), otherwise Holding.

Known limits. Article-title identity matters: a redirect or a differently titled article can understate interest. Bots are partially excluded by the API. A fad can be huge offline and quiet online. Price premium is only added when a permitted comps source exists; it is currently not published.

The shop census

Shops start from the Overture Maps places dataset (open data, CDLA-Permissive 2.0), filtered to toy, hobby, game, comic and collectible stores, national chains removed. Every candidate is then checked against its own website: we fetch the homepage plus contact/about pages, and extract facts with a model that must return a verbatim quote for each one. Any fact whose quote cannot be found in the fetched text is dropped automatically.

Depth gate. A shop page exists only when we have at least three quoted facts and the site shows a physical storefront. A city guide exists only when at least four such shops are in that city. Thin pages do not get published; noindex is not a cleanup tool.

Hours are not published unless a shop's own structured data states them, because stale hours are worse than no hours.

Corrections

If a shop has moved, closed, or changed hands, tell us through the contact form and we will fix it the same week. Closures require two sources.