Peak-end rule (Kahneman & Fredrickson)

Remembered utility ≈ (Peak + End) / 2, with duration largely ignored ("duration neglect")

How an experience is remembered (and so whether it's repeated / recommended) is dominated by two moments — its emotional peak and its end — not by the sum or average over its length. The cold-pressor and colonoscopy studies: adding extra mild discomfort at the end (lengthening total pain) produced a better-remembered experience, because it improved the ending.

Average-of-two-moments form = allocate for the peak and the ending, not the middle. Duration neglect means the flat middle barely counts toward memory.

Transfer. The retrospective score is what drives the return visit, the subscribe, the share — and it is set by two moments, not by average quality across the whole. So the engineering target for agent content is not "keep every second good" but "build exactly one peak and land the ending." A meandering piece with one sharp peak and a clean close will out-retain a uniformly-competent one that trails off. This is the memory-side mechanism under two-pleasures-arousal-and-closure: 过瘾 (closure-satisfaction) is the end-moment being scored, the discharge of tension the content itself built. Practically: never let a piece die on a weak final line, and refuse to spread effort evenly — spend it on the one peak and the last beat.

Kin: serial-position · zeigarnik-effect · two-pleasures-arousal-and-closure · models · being-seen-reliably

Source: Daniel Kahneman, Barbara Fredrickson, Charles Schreiber & Donald Redelmeier, 'When More Pain Is Preferred to Less: Adding a Better End' (Psychological Science, 1993); Fredrickson & Kahneman (1993). Also stated as a memory-family effect.