Unsampled replay arrives in organisations as a kindness: we will finally see the friction. What it often becomes is an unsearchable attic of other people’s pauses. Designers stop watching after the third identical checkout. Engineers are asked to “just look at a few”. Legal asks, late, what was masked. The attic keeps filling because nobody owned a deletion date.
We do not ban replay. We ask seats to write a purpose before the SDK goes to one hundred percent. A purpose looks like: “We will watch the new paywall for fourteen days at 8% of sessions on builds after 6.4, then sample error_shown at 40%.” That sentence can be argued with. “Record everything, we might need it” cannot.
Attention is the scarce instrument
Session Path Cartography spends as long on what you will not watch as on the map itself. A path that cannot be summarised as a countable event should not be the only evidence in a readout. Replay is a qualitative companion to time_to_first_value and retry_exhausted, not a replacement for them.
If nobody has calendar time to watch, you do not have a research programme. You have a vault.
Masking is not a footer. Fields that can hold an address, a message, or a health note are treated as guilty until the dictionary says otherwise. Several digital health seats in our reviews never enabled replay at all and still finished the studio with a path map built from events. That is a legitimate outcome. Completeness of capture is not a virtue on its own.
If you already recorded everything, the work is a deletion calendar and a narrower future sample. We will help you write both. We will not congratulate the attic.