Run controlled content and conversation experiments

Prioritize one-variable tests, hold editorial quality constant, document confounders, and avoid optimizing short-term engagement at the expense of trust.

23 minute field note