Variable rewards deliver payoffs on an unpredictable schedule so that the anticipation of a reward — not the reward itself — drives repeated, often compulsive, engagement. The principle is the most robust finding in the study of reinforcement: Skinner showed that behaviors reinforced on a variable-ratio schedule (a reward after an unpredictable number of responses) are performed faster and extinguish more slowly than behaviors on any fixed schedule. It is why a slot machine is more compelling than a vending machine, and why a feed that sometimes contains something good outperforms one that reliably does.
The neuroscience explains the pull. Dopamine neurons respond most strongly not to a reward but to a reward that is better than expected — the reward-prediction-error signal described by Schultz and colleagues. Maximal uncertainty about whether the next pull will pay off produces the largest anticipatory dopamine response. Products engineer that uncertainty deliberately: pull-to-refresh mimics a slot lever, feeds are ranked so the good post arrives unpredictably, and notifications arrive on no discernible pattern. The user experiences it as “just checking”; the mechanism is the same one casinos are regulated for.
The diagnostic separates honest design from manipulation. A tool that reliably gives you what you came for gets more useful the more predictable it is; a variable-reward product depends on you not being able to predict the payoff, because predictability would let you stop. When the inconsistency is the feature — when the design would collapse if the reward were dependable — you are looking at conditioning, not service.