Peak-end rule¶

Predict the remembered value of an extended experience from just two moments — its most intense point and its final moment, averaged — while the duration and the integral of momentary affect over the interior receive near-zero weight.

Core Idea¶

The peak-end rule is the empirical finding, established by Kahneman, Fredrickson, Schreiber, and Redelmeier (1993), that retrospective evaluation of an extended hedonic experience is determined approximately by the average of two moments — the most intense point (the peak, whether pleasant or painful) and the final moment (the end) — with the actual duration of the experience and the integral of momentary affect over that duration receiving near-zero weight.

The mechanism follows from how episodic memory encodes a continuous affective trajectory. Ongoing hedonic experience is generated moment by moment, but memory does not store its integral: it samples a sparse, salience-weighted record. Two sample points are systematically privileged. The peak is privileged because intensity drives attentional capture and encoding strength — the most painful or most pleasurable moment leaves a proportionally deeper memory trace. The end is privileged because retrieval is launched from the terminus, giving the final moment a recency advantage that places it reliably in the summary regardless of its absolute intensity. Interior moments that are neither extremely intense nor terminal are encoded weakly and contribute little to the retrospective verdict. The result is a structural dissociation: the summary judgment used to decide whether to repeat an experience tracks a two-anchor read of the intensity trajectory, not the time-averaged quality of the experience that was actually lived through.

The practical signature of the rule is duration neglect. In Kahneman's cold-water paradigm, subjects who held one hand in painfully cold water for 60 seconds, then held the other in the same water for 60 seconds followed by 30 seconds of slightly warmer (but still uncomfortable) water, subsequently preferred to repeat the longer trial — preferring strictly more total pain because the ending was less intense. The end dominated the integral. In Redelmeier and Kahneman's colonoscopy study (1996), patients whose procedure was extended by several minutes of low-level discomfort at the close rated the overall experience as less unpleasant and showed higher willingness to return than patients whose procedure ended at higher pain — even though the extended-trial group received more total pain. These results expose the architecture: the remembered experience and the experienced experience are calibrated to different statistics, and it is the remembered version that drives future choice.

Structural Signature¶

Sig role-phrases:

the extended affective trajectory — a continuous, instant-to-instant intensity signal (pleasant or aversive) over an arbitrary stretch of time, the full object being summarized
the lived integral — the time-integral of momentary affect, the statistic experienced utility tracks in real time
the peak anchor — the global intensity maximum, privileged because intensity drives attentional capture and encoding strength
the end anchor — the terminal moment, privileged because retrieval is launched from the terminus, giving it a recency advantage regardless of its absolute intensity
the two-anchor average — the remembered verdict, approximately the mean of peak and end, the quantity that actually drives repeat-or-avoid choice
duration neglect — the near-zero weight on episode length and on the discarded interior, the rule's conspicuous engineered omission and its litmus for being inside the regime
the experienced/remembered dissociation — the remembered statistic decoupled from the lived integral, so a strictly worse episode can be chosen again
the order sensitivity — because the end is recency-privileged, identical multisets of momentary affect in different orderings are remembered differently, a claim the order-blind integral cannot make

What It Is Not¶

Not a description of the lived experience. The rule governs remembered utility — the retrospective verdict that drives repeat-or-avoid choice — not experienced utility, which still tracks the moment-by-moment time-integral. The lived episode and the remembered one are calibrated to different statistics; the rule is a claim only about the second.
Not merely the recency effect. Recency names a general advantage for the last items in a sequence; the peak-end rule adds a second privileged anchor (the intensity peak) and the conspicuous duration-neglect commitment. It is a two-anchor average with near-zero weight on length, not a bare end-of-list boost.
Not a demonstration of irrationality. Subjects who choose to repeat the strictly-worse episode are not failing to compute; their choice correctly tracks remembered utility, which is what memory makes available. The cold-water "paradox" dissolves once experienced and remembered utility are held apart — it is a mechanical consequence of which statistic the judgment encodes, not a lapse.
Not a claim that duration is unimportant to experience. Duration neglect is about the memory of the episode: more time really is more total pain in real time. The rule says only that the extra time barely registers in the retrospective summary unless it alters the peak or the end — not that the lived integral is unaffected.
Not an unconditional law that a gentler-ending always wins. The peak-end average is an approximation, and adding a mild tail improves the remembered verdict only when the new terminus replaces a harsher one in the summary. Where lengthening an episode while holding peak and end fixed does move its remembered value, the regime is not peak-end-governed and the two-anchor read should not be trusted.
Not generic trajectory compression in any bounded summarizer. An integrating instrument that records the time-integral, a stateless logger with no privileged terminus, or a non-mnemonic data-compression scheme has no reason to discard the interior or over-weight the end, and so would not produce peak-end weighting. That sparse-anchor-summarization shape belongs to the parent sampling/compression/salience pattern; the named rule's specific calibration (peak + end, averaged, duration neglected) requires an episodic-memory architecture.

Scope of Application¶

The peak-end rule lives across the applied content areas of one substrate — the human episodic-memory readout of an affective trajectory — as studied in cognitive psychology and retrospective-judgment research; its reach is within that domain. The substrate-neutral "bounded summarizer keeps anchors plus an endpoint" shape belongs to its sampling/compression/salience parents, not to the named rule.

Medical procedures — the foundational paradigm (Redelmeier and Kahneman 1996): colonoscopy patients whose procedure was extended by minutes of low-level discomfort at the close rated the whole experience as less unpleasant and were likelier to return, despite strictly more total pain.
Customer experience and hospitality — deliberately engineered endings (a gift on departure, a memorable last course) lift trip ratings out of proportion to cost, since the peak and the close dominate the review.
Live-performance and event design — concerts, theme parks, and conferences engineer a climactic moment plus a strong close, allocating budget to peak and end rather than to average quality across the run.
User-experience research — onboarding and feature launches are judged less by the steady-state grind than by the highest point of delight and the last screen before quitting, driving the discipline of designing a strong final state.

Clarity¶

Naming the peak-end rule forces a distinction that retrospective-judgment research routinely collapses: experienced utility — the moment-by-moment affect integrated over the episode — versus remembered utility, the summary verdict that actually governs whether the experience is repeated. As long as these are treated as one quantity, the cold-water and colonoscopy results read as paradoxes — subjects "irrationally" choosing more total pain. Once the two statistics are held apart, the paradox dissolves into a mechanical fact: choice is driven by the remembered version, and the remembered version is calibrated to a different function of the trajectory (a peak-end average) than the lived version (a time-integral). The rule converts an apparent failure of rationality into a measurement question — which statistic of the experience does this judgment actually encode?

It also sharpens what within memory research had been a diffuse appeal to "recency." The rule decomposes the retrospective summary into named, separately manipulable anchors — peak intensity, end intensity, and the conspicuous near-zero weight on duration — and so lets a researcher ask the load-bearing question directly: of all the moments in an extended episode, which ones are written into the summary and which are discarded? Duration neglect, in particular, stops being a curiosity and becomes a sharp prediction — that lengthening an episode changes its remembered value almost not at all unless the added time alters the peak or the end. The practitioner gains a precise target: to move the remembered verdict, intervene on the two anchors, not on the average quality of the interior.

Manages Complexity¶

An extended hedonic episode is, in full, an unbounded object: a continuous affect signal varying instant to instant over an arbitrary stretch of time, with no obvious low-dimensional reduction — to predict the retrospective verdict from first principles one would seemingly need the entire trajectory, every momentary value and the duration over which they accrued. The peak-end rule collapses that infinite-dimensional record to three numbers: the intensity at the peak, the intensity at the end, and the sign of each (pleasant or aversive). The remembered verdict — the quantity that decides whether the episode is repeated — is approximately the average of the two anchors, and everything else about the trajectory, conspicuously its length and the integral of affect over the interior, drops out at near-zero weight. So a researcher or designer confronting any extended experience need not reconstruct or model the whole curve; the work reduces to locating two moments and reading off their intensities, with the qualitative outcome (will this be remembered as better or worse, chosen again or avoided) following from how those two compare.

This also fixes where intervention has leverage and where it is wasted. Because the interior and the duration carry almost no weight, effort spent improving average quality across the middle, or shortening or lengthening the episode, moves the remembered verdict scarcely at all; effort spent on the peak or the ending moves it directly. The sprawl of "how do I make this experience be remembered well?" — across every possible reshaping of a complex trajectory — contracts to two control points. And the family of seeming paradoxes (preferring strictly more total pain, rating a longer ordeal as milder) stops being a list of separate anomalies to explain case by case: each is the same two-anchor read predicting against the time-integral, and once the analyst tracks peak and end rather than the area under the curve, the direction of each "paradox" is read off rather than re-derived.

Abstract Reasoning¶

The peak-end rule licenses a tight family of inferences, all flowing from the same two-anchor read of the trajectory.

Diagnostic — infer the remembered verdict, and the lived experience it conceals, from two intensities. Given only the peak and the end of an episode, the analyst predicts the retrospective rating without reconstructing the interior: a procedure that ends mild will be remembered as milder than one that ends sharp, whatever the totals. Run in reverse, the rule reads behind a surface judgment to expose the dissociation: when a person reliably chooses to repeat an experience that was, moment for moment, worse — more total pain, more total time aversive — infer that remembered utility (peak-end average) and experienced utility (the time-integral) have come apart, and that choice is tracking the former. A flat or even inverted relationship between total affect and willingness-to-return is therefore not noise to be explained away but the rule's fingerprint. And from a known peak plus end, the sign and rough size of the gap between the two utilities can be estimated: the longer the aversive interior that the two anchors omit, the larger the hidden discrepancy.

Interventionist — to move the remembered verdict, act on the anchors, never on the average. The rule converts "make this experience be remembered well" into two specific control points with signed predictions. Add a mildly-good (or merely less-bad) tail after the peak and remembered value rises, even though total affect worsens — the counterintuitive prescription that extending an aversive episode with a gentle close improves its memory, since the new terminus replaces a harsher one in the summary while duration's near-zero weight absorbs the added time. End sharply at the peak and the episode is remembered as worse than one tapered down from the same maximum. Conversely, the rule predicts where effort is wasted: money and labor poured into lifting the quality of the interior, or into shortening or padding the middle, move the verdict scarcely at all, because those moments were never written into the summary. Each intervention is a falsifiable claim about the direction the remembered rating should shift, and the magnitude of the shift measures how much of the verdict the two anchors actually carry.

Boundary-drawing — when the rule binds, and when it lifts. It applies to an extended hedonic episode that is retrospectively summarized by an episodic-memory readout and then used to drive a repeat-or-avoid choice; the peak-end statistic is the right predictor exactly there. It does not govern the moment-to-moment experience as it is lived (that still tracks the integral), nor real-time choices made during the episode, nor evaluations by an integrating instrument or a stateless record that has no privileged terminus — strip the salience-weighted sparse encoding and the privileged retrieval-launch endpoint and the rule dissolves. A diagnostic boundary-test follows: if lengthening an episode while holding peak and end fixed changes its remembered value appreciably, the regime is not peak-end-governed (duration is doing work), and the two-anchor read should not be trusted. Duration neglect is thus both a prediction and a litmus for whether one is inside the rule's regime.

Order-of-events / predictive. Because the end is privileged by recency-at-retrieval, the rule makes the temporal placement of intensity decisive, not merely its presence: the same painful or pleasurable moment contributes differently depending on whether it falls at the terminus or in the discarded interior. So the rule predicts a sequencing principle — front-load or middle-bury the worst, finish on the gentlest available note; place the most intense delight at the climax and reserve a strong moment for the close, treating the middle as expendable for memory. Two episodes with identical multisets of momentary affect but different orderings are predicted to be remembered differently, with the one whose maximum and whose final moment are more favorable winning the retrospective comparison — a claim about order that the time-integral, being order-blind, cannot make.

Knowledge Transfer¶

Within cognitive psychology and retrospective-judgment research the peak-end rule transfers as mechanism, and its celebrated "cross-domain" applications are reach across content areas of one substrate — the human episodic-memory readout of an affective trajectory — rather than across structurally distinct substrates. Medicine (the colonoscopy and cold-water paradigms), hospitality and customer experience (engineered endings, a memorable last course), live-event and theme-park design (a climactic peak plus a strong close), and user-experience research (a strong final screen) are different settings feeding the same memory architecture, and the rule carries across them intact: the same two-anchor read, the same near-zero weight on duration and interior, the same intervention map. So the diagnostic ports without translation (predict the remembered verdict from peak and end alone; read a choose-to-repeat-the-worse pattern as the experienced/remembered dissociation), and so does the prescription — to move the remembered verdict, act on the two anchors, never on the average quality of the middle; taper an aversive episode to a gentle close even though it adds total pain; sequence the worst into the discarded interior and finish on the best available note. These are not separate findings to be relearned per field; they are one mechanism reasoned about in different content areas, exactly as the seed notes the "three substrates the criterion asks for collapse into one — human retrospective judgment."

Beyond an episodic-memory readout the picture is the third category, and the seam is sharp. A genuinely substrate-independent shape does recur as co-instances: any evaluation system with bounded encoding capacity and a privileged retrieval-launch point will summarize an extended signal by a few high-salience anchors plus an endpoint, with characteristic neglect of duration and interior. That sparse-anchor-summarization-of-a-trajectory pattern is the parent — carried by primes around sampling, compression, and salience-weighted memory (and it would unify peak-end with recency, primacy, and the von Restorff effect) — and where the cross-domain lesson is "a bounded summarizer compresses a long signal to a handful of anchors," that lesson belongs there. But the peak-end rule's own calibration does not travel: the specific weighting (peak + end, averaged, with duration at near-zero weight), the recency-at-retrieval-launch privileging of the terminus, and the salience-driven privileging of the intensity maximum are properties of an agent whose episodic memory encodes a sparse, salience-weighted sample of a continuous affect signal. Strip that architecture and the rule dissolves: an integrating instrument that records the time-integral, a stateless logger with no privileged terminus, or a neural-net agent without episodic readout would not produce peak-end weighting — they have no reason to discard the interior or to over-weight the end. So calling a generic data-compression scheme or a sensor's summary statistic "a peak-end rule" would import the peak/end/duration-neglect calibration that has no referent outside episodic memory — analogy at the level of "compress a trajectory," mechanism only at the level of the parent. The honest move is therefore layered: within human retrospective judgment the two-anchor mechanism and its full intervention map travel across every applied setting; the abstract "bounded summarizer keeps anchors plus an endpoint" lesson belongs to the sampling/compression/salience parents wherever a non-mnemonic system compresses a signal; but "peak-end rule," as named — peak and end, averaged, duration neglected — is reserved for the episodic-memory case whose specific architecture produces that specific calibration (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

Kahneman, Fredrickson, Schreiber, and Redelmeier's 1993 cold-pressor study ("When more pain is preferred to less") is the founding demonstration. Each participant underwent two trials of holding a hand in painfully cold water. In the short trial the water was kept at about 14°C for 60 seconds. In the long trial the same 60 seconds at 14°C were followed by an extra 30 seconds during which the water was gradually warmed to roughly 15°C — still unpleasant, but perceptibly less so. Asked afterward which trial they would repeat, a majority chose the long one, even though it delivered strictly more total pain. The less-painful ending, not the larger integral of suffering, drove the choice.

Mapped back: The hand immersion is the extended affective trajectory; the long trial's greater cumulative pain is the lived integral. The warmed final 30 seconds is the end anchor that lowers the two-anchor average, while the extra half-minute barely counting is duration neglect. Preferring the strictly-worse trial is exactly the experienced/remembered dissociation — choice tracking the remembered statistic, not the lived one.

Applied / In Practice¶

Redelmeier and Kahneman carried the rule into the clinic with colonoscopy patients, then commonly examined without sedation. In their 1996 study, patients reported their pain in real time during the procedure and rated it retrospectively afterward; the retrospective rating was well predicted by the average of the peak pain and the pain in the final moments, and scarcely at all by how long the procedure lasted. A later randomized trial deliberately extended some procedures by leaving the scope in place for a few extra minutes of low, minimal discomfort — a gentler ending. Patients in the extended group remembered the whole ordeal as less unpleasant and were more willing to return for follow-up, despite enduring more total discomfort.

Mapped back: The procedure is the extended affective trajectory; its worst instant is the peak anchor and its closing minutes the end anchor, whose mean is the two-anchor average predicting the rating. The retrospective rating ignoring procedure length is duration neglect, and the extend-with-a-gentle-close manipulation is the rule's interventionist prescription — improving memory while worsening the lived integral.

Structural Tensions¶

T1: Experienced utility versus remembered utility (which statistic is "the experience"). The rule's whole force is that an extended episode has two distinct valuations — the time-integral of momentary affect that was actually lived, and the peak-end average that memory writes down — and these can diverge to the point of inversion, a subject choosing to repeat strictly more total pain. The tension is that neither statistic is simply "correct": the lived integral is what was suffered, but the remembered verdict is what governs whether the episode is repeated, so choice is anchored to the version that is, by construction, a distortion of the version that mattered in real time. An account that fuses them reads the cold-water result as irrationality; one that holds them apart must then say which utility a decision ought to track. Diagnostic: Is the question "how did this feel while lived" (integral) or "will this be chosen again" (peak-end average) — and which one is the intervention meant to serve?

T2: Improving the memory versus improving the experience (the gentle-tail prescription cuts both ways). The rule's signature intervention — taper an aversive episode to a milder close — raises remembered value precisely by adding total pain, since duration's near-zero weight lets the new, gentler terminus replace a harsher one in the summary while the extra aversive minutes barely register. This is a genuine double edge, not a free lunch: the same manipulation that makes a colonoscopy more likely to be returned for also makes the patient endure more discomfort. A designer optimizing the remembered verdict and a clinician minimizing suffering are pulled in opposite directions by the identical move, and the rule cannot itself adjudicate which objective is legitimate. Diagnostic: Does the proposed ending improve what the person will remember and choose, or what they will undergo — and if these conflict, which is the design actually accountable to?

T3: Peak-end prediction versus duration neglect as litmus (the rule marks its own regime). Duration neglect is doing double duty: it is both the rule's boldest empirical claim (lengthening an episode changes its remembered value almost not at all) and the test for whether one is even inside the rule's regime (if holding peak and end fixed while stretching the episode does move remembered value, duration is doing work and the two-anchor read should not be trusted). The tension is that the same observation that most dramatically confirms the rule is also the boundary condition that can disconfirm its applicability, so a practitioner cannot assume the peak-end statistic is the right predictor without first checking that duration has genuinely dropped out. Applying the two-anchor read where duration still matters imports the rule's confidence into a regime it does not govern. Diagnostic: When length is varied with peak and end held fixed, does remembered value stay flat (inside the regime) or shift (duration is live, and peak-end does not apply)?

T4: Two anchors versus bare recency (bounding the concept downward). The peak-end rule is repeatedly collapsed into "the recency effect" — the last moments matter most — but this loses half of it. Recency names only the end anchor's retrieval-launch advantage; the rule adds a second, independently privileged anchor, the intensity peak, and the conspicuous commitment that duration and the interior carry near-zero weight. The tension is that recency and the peak-end average usually point the same way (a strong finish helps), so the extra structure is easy to overlook until a case pulls peak and end apart — a sharp peak buried mid-episode with a mild close, versus a mild interior climaxing at the terminus. An analysis that keeps only recency mispredicts exactly there, and misses that placement of the maximum, not just the ending, is written into the summary. Diagnostic: Does the account track a second privileged anchor at the intensity peak, or only the final moments — and does the case require distinguishing them?

T5: Order sensitivity versus order-blind totals (what the integral cannot say). Because the end is privileged by recency-at-retrieval and the peak by encoding strength, two episodes built from the identical multiset of momentary affects — the same suffering rearranged — are predicted to be remembered differently, the one whose maximum and whose final moment are more favorable winning. This is a claim about ordering that the time-integral, being a sum, is structurally incapable of making. The tension is that the lived-experience statistic and the remembered-experience statistic disagree not only in magnitude but in what variables they are even functions of: totals versus sequence. A designer who sequences the worst into the discarded interior and finishes on the best available note is exploiting a degree of freedom — order — that is invisible to any order-blind account of the episode's quality. Diagnostic: Do two arrangements of the same momentary affects differ in remembered value here — and if so, is the analysis using a statistic (peak-end) that can see order, or one (the integral) that cannot?

T6: Autonomy versus reduction (its own calibration or an instance of bounded summarization). "Peak-end rule" is a named, canonically measured finding with a specific calibration — peak and end, averaged, duration neglected — earning its own study through the cold-water and colonoscopy paradigms. Yet a more general shape recurs beyond human memory: any evaluation system with bounded encoding capacity and a privileged retrieval-launch point will summarize an extended signal by a few high-salience anchors plus an endpoint, with characteristic neglect of the interior. That sparse-anchor-summarization pattern is the parent — carried by primes around sampling, compression, and salience — and it, not the named rule, is what travels to a non-mnemonic summarizer. The rule's precise weighting requires an episodic-memory architecture; an integrating instrument or stateless logger would not produce it. Diagnostic: Resolve toward the parents (sampling, compression, salience-weighted summarization) when asking what recurs outside human memory; toward the named rule when predicting a person's retrospective verdict from a specific peak and end.

Structural–Framed Character¶

The peak-end rule sits toward the structural end of the spectrum but stops short of the pole — best read as mixed-structural, though its substrate is a mind rather than a lithosphere. On four of the five criteria its structural credentials are strong. Its evaluative_weight is low: the rule is a neutral empirical finding about how episodic memory encodes an affective trajectory, and the entry is emphatic that it is not a demonstration of irrationality — a subject who repeats the strictly-worse episode is correctly tracking remembered utility, so the rule names a mechanism, not a defect to convict. It is not human_practice_bound in the constitutive sense: no institution, tradition, or human practice makes the rule true — a colonoscopy patient's memory obeys peak-end weighting whether or not a psychologist is measuring it, so the phenomenon runs observer-free (its substrate is an agent's episodic-memory architecture, not a human convention that dissolves when removed). Its institutional_origin is none: Kahneman and colleagues discovered and formalized the weighting through the cold-water and colonoscopy paradigms, they did not invent it; the two-anchor calibration is a fact about cognition that predates its measurement. And within its proper range cross-setting reuse falls on the import_vs_recognize recognition side: medicine, hospitality, live-event design, and UX are content areas of one substrate (the human episodic-memory readout), so moving among them recognizes the same mechanism intact rather than borrowing a frame.

What keeps it off the structural pole is vocab_travels, which it fails at the level of its specific calibration. The rule's distinctive content — peak and end, averaged, with duration and the interior at near-zero weight, the terminus privileged by recency-at-retrieval and the maximum by encoding strength — is pinned to an episodic-memory architecture and does not float free: an integrating instrument that records the time-integral, a stateless logger with no privileged terminus, or a non-mnemonic compression scheme has no reason to discard the interior or over-weight the end, so calling any of them "a peak-end rule" imports a calibration with no referent outside episodic memory. The portable structural skeleton is a bounded summarizer with a privileged retrieval-launch endpoint compresses an extended signal to a few high-salience anchors plus its endpoint, neglecting duration and the interior — and that is exactly what the peak-end rule instantiates from its parent primes (sampling, compression, and salience-weighted memory, which would also unify it with recency, primacy, and von Restorff), not what makes "peak-end rule" itself travel: the cross-domain reach — any bounded summarizer of a trajectory — belongs to those parents, while the named rule's specific two-anchor weighting is episodic-memory cargo that stays home. Its character: a real, evaluatively neutral, recognized-across-content-areas memory mechanism, structural in its sparse-anchor-summarization skeleton but calibrated in a peak-plus-end-minus-duration form that pins it to the episodic-memory substrate, leaving it mixed-structural rather than a free-floating prime.

Structural Core vs. Domain Accent¶

This section decides why the peak-end rule is a domain-specific abstraction and not a prime, and it carries the case for its domain-specificity in one place.

What is skeletal (could lift toward a cross-domain prime). Strip the psychology and a thin relational structure survives: a bounded summarizer with a privileged retrieval-launch endpoint compresses an extended signal to a few high-salience anchors plus its endpoint, neglecting duration and the interior. The portable pieces are abstract — an extended signal too rich to store whole, a finite encoding capacity that forces a sparse sample, a salience weighting that privileges extrema, and a retrieval process launched from the terminus that privileges the endpoint. That skeleton is genuinely substrate-portable — any evaluation system with bounded encoding and a privileged endpoint recurs to it — which is exactly why the entry instantiates it as sampling (the sparse salience-weighted sample of a continuous signal), compression (the infinite-dimensional trajectory reduced to a few numbers), and salience-weighted memory (which would also unify peak-end with recency, primacy, and von Restorff). But it is the core the entry shares, not what makes the peak-end rule distinctive.

What is domain-bound. Almost everything that makes the concept the peak-end rule in particular is episodic-memory furniture, and none of it survives extraction. It requires an agent whose episodic memory encodes a sparse, salience-weighted sample of a continuous affect signal; the two privileged anchors are specifically the intensity peak (privileged because intensity drives attentional capture and encoding strength) and the end (privileged because retrieval is launched from the terminus, a recency-at-retrieval advantage); the summary is their average; and the signature commitment is duration neglect — near-zero weight on episode length and on the discarded interior — with the resulting experienced/remembered dissociation that lets a strictly worse episode be chosen again, plus the order sensitivity the order-blind integral cannot produce. The decisive test: remove the episodic-memory architecture and the affective-hedonic content — replace the summarizer with an integrating instrument that records the time-integral, a stateless logger with no privileged terminus, or a non-mnemonic compression scheme — and the specific peak-plus-end-minus-duration calibration vanishes, because such systems have no reason to discard the interior or over-weight the end. What is left is the looser sparse-anchor-summarization shape, not the peak-end rule.

Why this does not clear the prime bar. A prime is a relational structure whose vocabulary travels and whose cross-domain transfer is recognition of the same mechanism, not analogy. The peak-end rule's transfer is bimodal. Within human retrospective judgment it travels intact as mechanism — its celebrated "cross-domain" reach (medicine, hospitality, live-event and theme-park design, UX) is really reach across content areas of one substrate, the human episodic-memory readout, so the two-anchor prediction, the duration-neglect litmus, and the full intervention map (act on the anchors, taper to a gentle close, bury the worst in the interior) re-apply without translation. Beyond an episodic-memory readout it travels only by analogy: to call a generic data-compression scheme or a sensor's summary statistic "a peak-end rule" imports the peak/end/duration-neglect calibration that has no referent outside episodic memory. And when the bare structural lesson is needed cross-domain — that a bounded summarizer compresses a long signal to a handful of anchors plus an endpoint — it is already carried, in more general form, by the sampling, compression, and salience-weighted-memory parents the entry instantiates. The cross-domain reach belongs to those parents; "peak-end rule," as named, carries episodic-memory baggage — its specific two-anchor, duration-neglecting calibration — that does not and should not travel.

Relationships to Other Abstractions¶

Current abstraction Peak-end rule Domain-specific

Parents (3) — more general patterns this builds on

Peak-end rule is part of Duration Neglect Domain-specific

Duration Neglect is a strict constituent of the Peak-End Rule: the full rule combines a positive peak-and-end weighting with near-zero weight on episode length.
Peak-end rule is part of Compression Prime

Peak-End Rule contains Compression because it reduces an extended affective trajectory to a sparse two-anchor representation while discarding most interior detail.
Peak-end rule presupposes Evaluation Prime

Peak-End Rule presupposes Evaluation because it specifies how an extended experience is converted into a retrospective verdict that guides repeat-or-avoid choice.

Children (1) — more specific cases that build on this

Service Recovery Paradox Domain-specific is part of, typical Peak-end rule

Peak-End Rule is a typical constituent of Service Recovery Paradox because an exceptional recovery can become the overweighted positive ending in the customer's retrospective evaluation.

Not to Be Confused With¶

Recency effect. The general memory advantage for the last items in a sequence, driven by retrieval launched from the terminus. It is exactly the peak-end rule's end anchor in isolation — but the rule adds a second, independently privileged anchor (the intensity peak) and the conspicuous duration-neglect commitment. Recency and peak-end usually point the same way (a strong finish helps), so the difference only shows when a sharp peak is buried mid-episode with a mild close. Tell: does the account track a second privileged anchor at the intensity maximum, or only the final moments? Only the peak-end rule predicts from the buried peak.
Primacy / serial-position effect. The advantage for the first items in a list (primacy) and the U-shaped curve combining first and last (serial position). These are effects over ordered discrete items in free recall, weighting by list position; the peak-end rule weights a continuous affect trajectory by intensity extremum and terminus, and privileges no beginning. Tell: is the phenomenon better recall of early/late items in a list to be reproduced (primacy/serial position), or the retrospective valuation of an experience from its peak and end (peak-end)?
Von Restorff (isolation) effect. The tendency for a distinctive item — one that stands out from its neighbors — to be remembered better. It shares the salience-drives-encoding mechanism that privileges the peak, but it is about categorical distinctiveness within a set, not about the intensity maximum of an affective trajectory or the privileging of the endpoint. Tell: is the remembered item privileged for being unlike its context (von Restorff), or for being the most intense affective moment and the final one, averaged (peak-end)?
Focusing illusion. Kahneman's finding that when people judge well-being they overweight a single salient attribute they are prompted to attend to ("nothing in life is as important as you think it is while you are thinking about it"). It is a distortion of attention across features of a judgment, not a compression of a temporal trajectory to two moments. Tell: is one salient aspect dominating an evaluative judgment (focusing illusion), or two salient moments in time dominating the memory of an extended episode (peak-end)?
Duration neglect. Often taken as a synonym for the rule, this is actually one component of it — the near-zero weight on episode length. The full peak-end rule is duration neglect plus the positive claim that the remembered verdict is the average of peak and end; duration neglect alone says only what does not count, not what does. Tell: does the statement specify the two anchors that carry the verdict (the whole rule), or merely that length barely registers (duration neglect, the rule's signature but partial consequence)?
The bounded-summarizer parents it instances (sampling, compression, salience-weighted memory). The broad, substrate-neutral pattern the entry composes — a capacity-limited summarizer with a privileged retrieval endpoint reduces an extended signal to a few high-salience anchors plus its terminus — not confusable peers but the parents. The peak-end rule is the episodic-memory instance whose specific calibration (peak + end, averaged, duration neglected) requires an affective, recency-launched memory architecture the bare primes lack. Tell: strip the episodic-memory substrate and what remains is generic sparse-anchor summarization that also fits a non-mnemonic logger — at which point you are using sampling/compression, not the peak-end rule, which would not even predict its specific weighting there. Treated fully in a later section.

Neighborhood in Abstraction Space¶

Peak-end rule sits in a crowded region of the domain-specific corpus (38^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Attention, Memory & Automaticity (13 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12