Reward prediction error exploitation engineers surprise into rewards to amplify their psychological impact. The underlying neuroscience is well established: Schultz, Dayan and Montague (1997) showed that dopamine neurons fire not to reward itself but to the difference between reward received and reward predicted — the prediction error. An expected reward produces little signal; an unexpected or larger-than-predicted one produces a strong one. Designs that deliberately break an expectation — paying out when least anticipated, or more than trained — manufacture that spike, and Berridge and Robinson’s work on “wanting” versus “liking” explains why the anticipation itself, not the enjoyment, sustains the behavior.
Genuine delight and honest surprise are legitimate. This becomes manipulative when unpredictability is deliberately tuned to keep the prediction-error signal firing for the purpose of maximizing compulsion — engineered variance, near-miss teases, and coming-reward hints calibrated to sustain anticipation rather than to reward anything real. It is a more sophisticated relative of ordinary variable rewards (T18.1): where variable reward simply randomizes payoff, prediction-error exploitation specifically targets the expectation-violation that drives the neural response, often layered onto frequent micro-rewards and precise dopamine scheduling.
Because it operates on a neural mechanism below conscious awareness and is deliberately disguised as natural variation, it is hard to detect and correspondingly hard to resist, placing it in the High band despite an emerging rather than universal prevalence. The defense begins with recognition: an outsized reaction to a small unexpected reward is a manufactured prediction-error spike, not a genuine gain. Naming the mechanism, pricing each payout by its real value, and adding deliberate friction and delay deny the system the ability to deliver surprises on its own cadence — and treating near-miss teases as cues to stop rather than continue breaks the anticipation loop.