Screenshot tests feel tangible, which makes them tempting. But a visible difference is not automatically a meaningful hypothesis. Start by naming the behaviour you expect the new sequence to change.
Choose one learning goal
A test might ask whether earlier evidence of trust helps cautious visitors, or whether showing the primary task first improves relevance. It should not attempt to answer both at once.
Protect the comparison
Record campaign, release and seasonal changes during the experiment. Check whether audience composition shifted, and allow enough observations for the result to stabilise.
Keep the losing test useful
A variant that does not improve conversion can still challenge an assumption. Capture what happened, what remains uncertain and which future concept the result rules out.