Optimizely runs the test. Something has to tell you what to test.

For most teams, no, and we would rather say so plainly. Optimizely is a mature experimentation platform with a genuinely excellent statistics engine, feature flagging, and audience targeting at scale. If you have a full experimentation program with a steady hypothesis backlog and the traffic to power it, keep it. Mimetic solves the problem upstream: knowing which of the hundred things on your site is actually costing you money, and whether the numbers you are about to trust are even being recorded correctly.

Two things. First, hypotheses grounded in observed defects rather than opinion, which is usually the scarcest input in an experimentation program. Second, measurement verification. A surprising share of inconclusive tests are inconclusive because the event was firing twice, the consent banner was blocking the tag, or the variant never rendered on mobile Safari. Mimetic checks that layer before you spend three weeks of traffic on a test that could not have produced a clean answer.

No, and testing everything is a common way to waste traffic. A checkout that fails in an in-app browser or a purchase event that double-counts is a defect, not a design question. You fix defects and verify they are gone. Reserve experiments for genuine judgement calls where a reasonable person could argue either way, which is exactly what Optimizely is good at.

Mimetic is primarily a diagnostic. When it does serve a change, patches are bounded and validated, and the holdout arm is assigned server-side so a client can never select its own arm. But if your priority is high-volume client-side experimentation with tight flicker control, a dedicated platform is the right tool and we will say so.