What I would test, on sites I do not work on

Each teardown takes one surface of a live storefront, names the funnel stage, states a hypothesis, sketches a variant, and picks a primary metric and a guardrail.

These are written from the outside, with no access to anyone's analytics. The team almost certainly knows things I do not, and may have already tested the idea. The value here is the line of inquiry and how I would set the test up, not the conclusion.