Open stats

How we count

• A view is a browser loading a page and its counting pixel, a tiny image that is never cached. Pages are cached at the edge, so the pixel is the part that reaches us.
• A visitor is one network and browser on one day. We hash the IP address (the /64 block for IPv6) with the browser's user agent and a salt that is replaced every UTC day. The salt and the hashes are deleted when the day ends, and nothing that identifies a person is stored.
• Repeats don't count. One visitor counts once per page per day, and for at most 60 pages a day in total.
• A read is a second pixel placed after the story and loaded lazily, so the browser only fetches it as the reader gets near the end. Browsers start loading a little early, so read-through is a close estimate.
• A directory click is counted through the link's ping, which the browser sends when a reader follows the link. Browsers that turn pings off (Firefox, Brave) send nothing, so the click count is a minimum.
• A save is a reader pressing “Save to ChatGPT / Claude” on a story. The link passes through our server, which counts it and sends the reader on to the assistant. Only a click the browser reports as the reader's own is counted, once per story per assistant per day. We record which assistant was picked and never see what happens there.
• Not counted: crawlers, scripts and link previews; requests that did not come from one of our pages; search results, the 404 and other utility pages. Older browsers that don't send fetch metadata are left out too, which keeps every figure on the low side.
• Nothing is sampled or adjusted. Days run on UTC, and every total is a plain sum of daily counts.