Journal · 9 April 2026
A session is a timeout wearing a suit. Web analytics inherited it from server logs: if the next request arrived after thirty idle minutes, a new visit was declared. Mobile operating systems do not live that way. An app can sit in memory, a user can glance at a lock-screen widget, and a webview can die without anyone meaning to leave. When a dashboard still reports “sessions up 12%,” it is often reporting a change in how the OS interrupted people, not a change in care.
In the Signal Atelier we ask students to pick one Tuesday and list every time the product was truly used. Then we compare that list to the session table. The gap is rarely small. Background fetches, push-driven opens that last four seconds, and double-counted cold starts after a crash all inflate the number. A growth lead who celebrates sessions is celebrating the clock.
Timeouts make it worse. Thirty minutes is a web habit. On a navigation app, a person may keep the product open on the passenger seat for an hour without touching it; that is still one journey. On a banking app, three separate authentications in a morning are three intents, even if they fall inside one vendor session. Identitycachegrid treats session as a debugging aid, never as a north star.
What we read instead is closer to the product’s job. Completed first funded transfer. Class attended to the last minute. Article finished, not merely opened. Those events still need honest identity: a reinstall should not mint a new engaged person if the same account returns. That is why the house name puts cache before grid. You cannot grid behaviour you have not held still.
If your board pack still leads with sessions, try a fortnight of silence on that row. Replace it with a count of first-value actions and a small note on how the OS version mix shifted. The room will feel less exciting. The argument will be easier to defend when finance asks what the number is made of.