Before/after vs causal
Before/after shows what moved around your change; a causal estimate nets out what moved anyway. We label which.
A before/after compares your metrics in the window before a change to the window after. It is honest and useful, but it is correlational: it cannot separate your work from a seasonal bump or an algorithm update that happened at the same time.
A causal estimate uses a synthetic control to subtract what would have happened anyway, so the remaining lift is attributable to your change with stated assumptions. It needs enough matched peers (a donor count of 30+) to be trustworthy.
The product always labels which one you are looking at. When we cannot support a causal claim, we say before/after only - we never quietly upgrade a correlation into a proof.