The traffic view knows a returning visitor: new versus returning, how many visits each made, and the path a returning visitor took before starting or buying a forge
Why. Owner, 2026-11-11: "i want you to lean into observability too. i want fingerprinting so we know if there are repeat users." The traffic view counts a visit as one visitor hash per day, so the same person on Monday and Thursday is two strangers. To sell, the owner needs to know who comes back, how often, and what they did the visit before they bought. Goes with the funnel in #11259.
What.
A stable first-party visitor id. On the first page view the forge sets a random id in a first-party cookie (HttpOnly, Secure, SameSite=Lax, 13 months, on the forge's own host only). It carries nothing about the person and is never sent to anyone else. This is the primary signal: it is exact, where a browser fingerprint is a guess.
A fallback for visitors without the cookie (cookies off, a new private tab): a salted hash of coarse request facts the server already sees — user-agent family and major version, OS, language, and the IP's network prefix (/24, /48 for v6) — with a salt that lives on the forge and rotates every 30 days. It is marked "probable" wherever it is shown, never merged silently with a cookie id. No canvas, font, audio or WebGL probing and no client script that enumerates the device.
What is stored: per visitor id, first seen, last seen, days seen, visit count, entry referrer of the first and the latest visit, and the ordered pages of each visit (path only, no query string). Raw IPs and full user-agent strings are not stored. Rows older than 13 months are deleted by a maintenance duty.
What it shows (Settings → Traffic and the Billing funnel, admins only): new versus returning visitors per day; a returning-visitor list (id shortened, visits, first and last seen, first referrer, furthest step reached: saw plans, clicked a plan, started a forge, paid); and for any forge started or paid, the visits that led to it. GET /api/v1/admin/traffic/visitors?days=30 answers the same; the traffic probe proves the counts against their rows.
When a visitor becomes someone: starting or buying a forge links the visitor id to that forge's name on the host's side, so the owner sees "this buyer first came from <referrer> six days ago, three visits".
Respect: a request carrying Sec-GPC: 1 or DNT: 1 gets no cookie and no stored visitor row (it still counts in the anonymous daily total). Scrapers stay counted apart and get no id. The privacy page says in three plain lines what is kept, for how long, and that it never leaves this forge.
A setting, per forge: traffic.returning — on for the top forge, off by default on every new forge, so a hosted forge's owner chooses it for their own visitors. Nothing a child collects reaches its host.
Acceptance. Tests: two visits with the same cookie on different days are one returning visitor with two visits; a cookie-less repeat from the same network and browser is "probable"; a GPC request stores no row; a scraper gets no id; the visitor who starts a forge is linked to it; rows past 13 months are removed; a new forge has the setting off. No raw IP or full user agent appears in the store (asserted). The returning-visitors row fits one line of the first view.
The traffic view knows a returning visitor: new versus returning, how many visits each made, and the path a returning visitor took before starting or buying a forge
Why. Owner, 2026-11-11: "i want you to lean into observability too. i want fingerprinting so we know if there are repeat users." The traffic view counts a visit as one visitor hash per day, so the same person on Monday and Thursday is two strangers. To sell, the owner needs to know who comes back, how often, and what they did the visit before they bought. Goes with the funnel in #11259.
What.
HttpOnly,Secure,SameSite=Lax, 13 months, on the forge's own host only). It carries nothing about the person and is never sent to anyone else. This is the primary signal: it is exact, where a browser fingerprint is a guess.GET /api/v1/admin/traffic/visitors?days=30answers the same; the traffic probe proves the counts against their rows.Sec-GPC: 1orDNT: 1gets no cookie and no stored visitor row (it still counts in the anonymous daily total). Scrapers stay counted apart and get no id. The privacy page says in three plain lines what is kept, for how long, and that it never leaves this forge.traffic.returning— on for the top forge, off by default on every new forge, so a hosted forge's owner chooses it for their own visitors. Nothing a child collects reaches its host.Acceptance. Tests: two visits with the same cookie on different days are one returning visitor with two visits; a cookie-less repeat from the same network and browser is "probable"; a GPC request stores no row; a scraper gets no id; the visitor who starts a forge is linked to it; rows past 13 months are removed; a new forge has the setting off. No raw IP or full user agent appears in the store (asserted). The returning-visitors row fits one line of the first view.