Replay Sampling

Replay sampling means recording only a portion of sessions instead of all of them — to contain data volume and stay within the quota and capacity ceiling.

Recording every session both costs storage and consumes quota, so at high traffic recording everything is neither possible nor necessary. Sampling solves this, but how it is done decides everything. Random sampling is simple but risky: because troubled sessions are usually a minority, a blind draw may miss exactly the ones you needed to see. Smart sampling, instead of rolling dice, keeps the more valuable sessions — for instance those carrying a friction signal or an important event — in priority for recording. A more advanced layer is custom capture rules: specifying that only sessions on a certain URL, or sessions in which a given event occurred, are recorded. The big trap of sampling is that any metric drawn from recorded sessions is biased and must not be taken as the true rate across all users; sampling is for “seeing the right examples,” not for “counting precisely.”