Every seedling carries two potential outcomes: the height it would reach if left alone (Y0) and the height it would reach if treated (Y1). The true effect is Y1 − Y0 — but you only ever observe one of them per plant. That is the fundamental problem of causal inference.
Here vigor secretly drives Y0: hardy seedlings grow tall no matter what. When you decide who gets treated, you become the confounder. Route the vigorous into the treated plot and the observed difference credits the treatment for growth the plants already had.
The naive estimate splits into two pieces: the true effect plus a selection bias equal to how different the groups were to begin with. Tilt the gate and watch that bias swing the needle far from the truth you set.
The brass lever assigns by coin flip. Randomization makes the groups exchangeable: any imbalance is now pure luck, zero on average, and it averages away over repeated trials. It is the one gate a confounder cannot rig.
The trial's measuring frame jammed mid-season. The lesson continues without it — everything you have learned so far is safe.