Views Links · metric contract v1

How Views Links counts

Every figure on a Views Links dashboard is defined here. When a rule changes, the version number changes with it and the dashboard shows which version produced each figure.

The figures

Total redirect requests: GET requests that resolved an active link and were redirected, automated traffic included. Excludes HEAD requests, unknown addresses, paused, expired, deleted and disabled links, and our own monitoring. Best effort: if recording a click fails, the visitor is still sent on and the loss is counted in our operations logs, never filled in later.
Filtered clicks: The total redirect requests that pass the automation filter below. An estimate of human activity, not proof of a person.
Excluded automated requests: Total redirect requests minus filtered clicks, with a coarse reason for each. Excluded requests are still redirected.
Daily unique click estimate: Distinct visitor keys among filtered clicks, per link, per UTC day. The key is a keyed hash of the visitor’s network address and a coarse browser description (device type, browser family, OS family) under a secret that changes every day and is deleted after 48 hours. It is not a person: a shared network or VPN can merge people, and one person on two networks counts twice. Across several days we show the SUM of daily estimates and call it that, never “unique visitors”.
Attributed arrivals, playback starts, qualifying views, click-to-view rate: Not available. They need a Views watch page with a player we can measure, which does not exist yet. A link to an external site shows “playback measurement unavailable”, never zero views. Click-to-view rate, when it exists, will be a different number from Verified Rate and will be labelled as such.

The automation filter, version 1

Browser prefetch or prerender (prefetch): the browser said it was loading the link ahead of time, not because someone opened it.
Known crawler or link-preview bot (crawler): the request identified itself as a crawler, a link-preview fetcher (messaging apps and social networks build previews this way) or a scripted client.
No user agent (no-user-agent): the request carried no browser description at all.
Same visitor key repeating on one link (repeat-burst): more than 30 requests from one visitor key on one link in one UTC day; the ones past that point are excluded.

No filter is perfect. A preview service that disguises itself as a browser is counted; a person using an unusual client can be excluded. Sharing an IP address or using a VPN never excludes anyone on its own, and nobody is ever blocked from the destination by this filter.

Where clicks came from

Referrer: the site in the browser’s Referer header, host only. Many apps and browsers send none, so “Direct / unknown” does not mean someone typed the address. For reliable sources, make one link per place you share.
Country: approximate, from the network location our hosting provider reports. Never city level.
Device: inferred from how the browser describes itself.
Days: clicks are stored in UTC and grouped by calendar day in the time zone the dashboard shows, with daylight-saving changes handled by the database. Unique estimates are always per UTC day.
Destination versions: changing where a link goes creates a new version. Each click keeps the version it was sent to.

What we store, and for how long

Each click: time, link, destination version, classification, inferred device, referring site, approximate country. Kept 30 days.
Daily visitor keys and the secret that makes them: 48 hours. After that, nobody, including us, can recompute them.
Daily totals: 13 months.
Never stored: visitors’ IP addresses, full browser descriptions, cookies on the visitor, or anything that follows a visitor from one link to another.
Destination addresses belong to the link owner and are treated as private: query values are hidden in public previews and moderation, and left out of exports unless the owner asks for them.