How it works

How they were chosen

About 85 candidate rankings were measured two ways over the twelve months to October 2026: English Wikipedia pageviews of each ranking’s article (what people read) and Google Trends search interest, worldwide (what people look up). Each signal was scaled from 0 to 1 and the score is their average.

Then three rules: skip lists that almost never change (there would be nothing to monitor), allow at most six per category (three in ten, so no one subject crowds out the rest), and use only sources that can be read without breaking their terms. Where the best-known version had no such source, the closest clean stand-in is used and labelled. The table lists the 20 chosen, highest score first.

Where the data comes from

How changes are detected

• Every few minutes the most overdue ranking is fetched. A new snapshot is stored only when the order actually changes.
• A fetch that is incomplete or malformed is rejected, so it can never invent an exit or erase good data. Failures show on the status page.
• Wikipedia-based lists use the newest revision that is at least an hour old, so vandalism reverted within the hour never shows up.
• When too much of a list changes at once, the new order is held until a later check confirms it.
• When a season or year rolls over, the list starts afresh instead of reporting every entry as a drop-out.
• History from before a ranking was first checked is rebuilt from its source’s own archive: Wikipedia’s page history, and race-by-race results for F1. The archive is sampled daily for sports in season and weekly for lists, so a change found this way is dated to the sample that first showed it.

The same data is available as JSON at /api/rankings and as an Atom feed of changes.