Expectation Violation¶

Version: v4 · 2026-08-30 · History
Prime #: 1360
Origin domain: Psychology Cognitive Science
Subdomain: prediction and attention → Psychology Cognitive Science
Related primes: Baseline Deviation

Core Idea¶

An expectation violation occurs when something is already committed to what will happen, an event arrives that is incompatible with that commitment, and the incompatibility is large enough that the system does something about it. Four parts are load-bearing in that order: a model in force before the event; a realization that speaks to the same question the model answered; a mismatch above the level at which the system would otherwise absorb it; and a registered response — surprise, slowed processing, a shift of attention, model revision, or repair. ^[1]

What makes the prime worth naming is not the mismatch but what the response reveals. Expectations are mostly tacit. A listener cannot recite the harmonic rules they hold; a user cannot list the interface conventions they rely on; a team cannot enumerate the assumptions its runbook encodes. The breach converts that tacit content into observable behaviour, which means a violation is less a fact about the world than a readout of a model that had no other way of being seen. ^[2]

This identity is much narrower than ordinary usage. Everyday speech calls anything unusual, unwelcome, or rule-breaking a violated expectation. Here nothing counts unless a specific anticipatory commitment was in force and the event contradicted it; the rarity of the event, its unpleasantness, and whether anyone was entitled to expect otherwise are all beside the point. ^[3]

Structural Signature¶

The pattern is: an anticipatory model held by a bearer, plus a commensurable realization that contradicts it, crossing a registration threshold, yields a response whose shape and size expose the model's content.

Recurring features:

Priority in force. The commitment exists before the event rather than being reconstructed after it; a model fitted in hindsight can be illustrated by an outcome but never breached by one. ^[4]
Commensurability. The realization must fall in the space the model spoke about, at the grain the model used. An event on a dimension the model never addressed is irrelevant rather than contrary.
A registration threshold. Below some tolerance the mismatch is assimilated as noise, rounding, or acceptable variance and nothing downstream changes; it is the threshold, not the raw size of the discrepancy, that decides whether a violation exists at all. ^[5]
A bearer. Something holds the expectation — a perceptual system, a controller, a reader, a community sustaining a convention — but nothing requires that bearer to be deliberative or able to report what it anticipates.
A costly response. The reaction spends something measurable: reading time, latency, an interrupt, attention, repair work. That expenditure is what makes the breach detectable from outside the bearer.
Valence neutrality. The breach may be delightful, alarming, or flat. Punchlines, plot reversals, and outages instantiate one shape, because incompatibility rather than harm is the defining relation.
Open resolution. Assimilating, revising the model, repairing the world, and disputing the event are all admissible endings. The structure specifies that a response occurs, never which one.

What It Is Not¶

This abstraction does not claim that expectations are conscious, verbal, or available for report. Most models that get breached were never formulated: a reader's syntactic anticipations, a driver's sense of what the next car will do, a household's unspoken rule about who speaks first at dinner. What is required is that the commitment was operative, not that anyone could have written it down beforehand. ^[6]

Despite the word, the prime carries no verdict. "Violation" here does not mean wrongdoing, and it does not imply that the expectation was reasonable, accurate, or entitled to be met. A prejudiced person's anticipations are breached by exactly the same structure as a careful forecaster's. The abstraction describes a mismatch and what follows from it; whether the model deserved to survive is a separate judgement resting on other grounds entirely. ^[7]

Nor is this a claim about rarity. Frequency in the world and contradiction of a model are different variables that routinely come apart. A thoroughly ordinary event breaches a strong local commitment — your own front door standing open — while an event with a vanishing base rate breaches nothing if no model ever ruled it out. Improbability is not the criterion; contradiction is.

It also does not require felt surprise. The response can be a longer fixation, an extra clock cycle, a logged exception, or a clarifying question, none of which resemble emotion. And it does not require that the model be corrected afterwards. Absorbing the breach and carrying on is a real response with real downstream effects, not a failure to instantiate the pattern.

Finally, the abstraction supplies no scale. There is no general unit in which the magnitude of a breach can be stated, and comparisons of how badly two different bearers were caught out are not licensed by the structure by itself.

Broad Use¶

Perception and language. Sentence comprehension slows measurably where a continuation contradicts the reading a parser has already committed to, and the slowdown localizes to the word that does the contradicting. Visual systems behave the same way when an object fails to appear where motion made it due. Developmental looking-time methods rest entirely on this logic: infants who cannot report anything look longer at outcomes their implicit physics rules out, and the looking time is treated as the readout. ^[8]

Music. Tonal, metric, and timbral expectations are acquired by exposure and breached by cadences that resolve elsewhere, entries that arrive off the grid, and instruments that enter where none was implied. Composition is largely the management of these breaches, which is why the structure appears here in its most deliberately engineered form.

Narrative and comedy. Reversals, unreliable narrators, and punchlines all install a reading and then contradict it. Comic timing is threshold work: the setup must build a commitment strong enough that the contradiction registers, and the interval before the breach is what lets the commitment consolidate.

Interaction design. Mode errors, controls that look actionable and are not, destructive operations where a reversible one was implied, and responses arriving long after a user's internal deadline are breaches of conventions the user could not have articulated before they were broken.

Social interaction. Greetings unreturned, questions met with silence, physical distance closed further than a setting allows: the repair work that follows — apology, explanation, renegotiation — is the response that makes an invisible script legible to everyone present.

Operations and monitoring. Forecast-based alerting, canary analysis, and anomaly detection all pair a model of record with an incoming realization and act when the gap crosses a band. Practitioners tune these systems by moving the threshold rather than the model, which is the same lever the structure names. ^[9]

Scientific and engineering practice. An anomaly exists only relative to a theory that made a contrary commitment; runtime assertions, invariants, type checks, contract tests, and regression baselines are expectations deliberately written down so that a machine can register their breach.

Markets and institutions. An earnings surprise is defined against a published consensus that serves as the model of record. A breach of contract is an expectation written into an enforceable document precisely so that the response is not left to the parties' discretion.

Clarity¶

The most useful thing this abstraction does is relocate the object of interest. Confusion arises because breaches are experienced as properties of events — the outcome was shocking — when they are always properties of a pairing. Once that relocation is made, several persistent muddles dissolve at once. Two observers reacting differently to the same occurrence are no longer disagreeing about the world; they are exhibiting different models, and the productive question stops being who is right about the event and becomes what each of them expected. ^[10]

It also separates two things that get conflated under the single word "surprising": how improbable an outcome is and how sharply it contradicts a commitment in force. These come apart constantly, and keeping them apart is what makes it possible to say that a routine occurrence can be diagnostically explosive while a genuinely rare one can be inert.

A third confusion it clears concerns blame. When something goes wrong, the available framings are usually "the system failed" or "I was mistaken," and choosing between them tends to be settled by status rather than evidence. The abstraction supplies a third and more accurate description: the model and the world came apart at a specific place, and locating that place is a prior step to deciding which of them should move.

Manages Complexity¶

The saving is that you stop tracking confirmations. An environment produces overwhelmingly more events consistent with a model than inconsistent with it, and every consistent one can be discarded without loss, because it adds nothing to what the model already asserts. Only the contradictions carry information about where the model stops being right. ^[11]

That reduces monitoring from an impossible obligation — observe everything, in case some part of it matters — to a bounded one: know what is currently committed, and watch the comparatively small number of places where a live commitment could be contradicted. Attention budgets, alert budgets, and review budgets are all finite; this is the compression that makes them viable rather than nominal.

It also removes the need to keep a separate description of the norm. The accumulated record of registered breaches is an operational specification of a model's boundary, expressed in exactly the terms that turned out to matter, and it stays current without anyone maintaining it. Where a written standard and a breach log disagree, the breach log is the one describing the model actually in force.

Abstract Reasoning¶

The abstraction licenses a specific diagnostic procedure, run in order. First, name the bearer and state one thing its model ruled out before the event; if that cannot be done, there is no candidate violation to analyse. Second, check commensurability: did the model address this dimension at this grain? Third, check whether anything downstream actually changed — if the mismatch was absorbed with no cost, it fell below the registration threshold and was assimilated, not breached. Fourth, read backwards: the location, size, and shape of the response bound the content of the model, which is the inference the whole structure exists to support. ^[12]

The controlling counterfactual is whether a bearer holding a different model, or none, would have responded the same way to the same event. If it would, the reaction is driven by the event's intrinsic properties and the violation framing is doing no work.

Two further moves follow. Breaches can be manufactured, which turns the abstraction into an elicitation instrument: deliberately contradict a commitment and the resulting response reports on a model nobody could have described. And the absence of breaches is itself evidence, though weak and ambiguous, since a model that never registers one is either accurate, not actually in force, or tuned loosely enough that nothing can contradict it.

Knowledge Transfer¶

What carries across substrates is the four-role skeleton, the response-as-readout inference, and the threshold trade-off between missing breaches and drowning in them. Also portable is the drift result: repeated breaches at the same place erode the commitment until the identical event stops registering, which is why any long-running detector must be tested rather than trusted. These transfer intact from perception to alerting to institutional practice. ^[13]

What does not carry is nearly everything that makes the human case vivid. Felt surprise is one implementation of the response role, not the role itself, and importing the affect into a description of a monitoring system produces anthropomorphic nonsense with no predictive content. Valence does not carry: a breach that is sought after in music and catastrophic in aviation has identical structure. Timescale does not carry either, ranging from milliseconds in a parser to decades in a legal convention, so intuitions about how quickly revision should follow are domain-local.

Magnitude is the most dangerous non-transfer. Each substrate scales its breaches in its own units, and there is no exchange rate between them, so claims that one violation is larger than another across domains are unsupported by the shared structure and should be treated as rhetoric.

Examples¶

Formal/abstract¶

Consider a cadence in common-practice tonal harmony. An extended passage establishes a key, and a dominant chord in that key installs a strong commitment about what comes next: the tonic. A composer instead supplies the submediant. Voice-leading is satisfied, the chord is consonant, nothing sounds like an error — and yet listeners with no theoretical vocabulary register the substitution immediately, hearing the phrase as unfinished and continuing to attend where they would otherwise have relaxed.

Every role is filled. The model was installed by exposure to a corpus of style rather than by instruction, and was in force before the chord sounded. The realization is commensurable, occupying precisely the slot the commitment addressed. The mismatch crosses a threshold that small deviations do not: an ornament or a slight stretch of tempo at the same moment is absorbed without comment. The response is reorientation, and it costs something — the phrase must now be extended to reach closure.

The controlling test is available. Play the same chord to a listener unexposed to the idiom, or in a passage that never established a key, and the response disappears while the acoustic event is unchanged. The breach was never in the sound.

Mapped back: This case shows the diagnostic payoff in its cleanest form. Before the substitution, the listener could not have stated the rule they were holding; the response externalizes a model that introspection could not reach. It also demonstrates valence neutrality, since the breach is the reason the passage is worth hearing rather than a defect in it, which is precisely why incompatibility and not harm is the defining relation. ^[14]

Applied/industry¶

A payments team maintains a seasonal forecast of authorization volume, published as the model of record and used to set an alert band. On a Tuesday morning the realization arrives forty percent below the forecast interval. The band is crossed, an on-call engineer is paged, and the response begins.

Investigation splits along the two admissible resolutions. Either the world departed from something that should be restored — a degraded acquirer, a bad deploy, an expired certificate — or the model is stale and should be revised, because a promotion ended and the elevated baseline it produced was never a permanent feature of the system. The breach does not say which; the attribution has to come from evidence gathered afterwards.

The diagnostic yield shows up in the postmortem rather than the page. Reconstructing why the forecast asserted what it did surfaces assumptions nobody had recorded: that a single acquirer would carry the bulk of traffic, that promotional lift decays over a particular window, that retries are counted as authorizations. None of this was written anywhere before the band was crossed.

The failure mode is equally visible. Each unexplained dip that gets waved through as noise widens the band a little, and after enough of them the alert cannot fire — leaving a dashboard that looks calm for exactly the same reason a vacuous model looks calm.

Mapped back: The forecast is an anticipatory commitment made explicit and mechanically checkable, which is what lets a machine register the breach at all. Threshold placement, the repair-or-revise fork, the readout of assumptions in the postmortem, and the drift toward a model that can no longer be contradicted are the same structural features the musical case shows, now carrying operational cost. ^[13]

Structural Tensions¶

T1 — Threshold placement trades readout against noise. Lower the registration band and more mismatches become breaches, so more of the model is exercised and reported — at the cost of promoting genuine noise to signal and exhausting whatever responds. Raise it and only gross contradictions register, leaving the model's finer commitments permanently invisible. No setting is correct independent of what the readout is for: a fraud monitor and a first-time user of the same interface want opposite bands, and the same event is a breach for one and unremarkable for the other.

T2 — Repair and revision are opposite verdicts on one signal. A breach can mean the world departed from something worth restoring, or that the model was wrong and should change. The structure does not adjudicate. A bearer that always repairs hardens a stale model and keeps paying to force reality back into it; a bearer that always revises dissolves every standard it holds, because a model amended to accommodate each contradiction eventually predicts whatever happens and can no longer be breached at all. Attribution must come from outside the violation.

T3 — Probing erodes the expectation it measures. Manufacturing a breach is the sharpest way to read out a tacit model, but the act also teaches the bearer that breaches occur at that place, weakening the very commitment under study. The second exposure is always quieter than the first, and a probe run often enough widens the model until the response vanishes. The measurement is therefore most informative exactly once, and diagnostic yield falls as repeatability rises, so a probing programme has to keep moving to new commitments rather than returning to proven ones — the instrument consumes its own subject.

T4 — A relation experienced as a property of the event. Bearers report that an occurrence was shocking, as though the breach were located in the world rather than between the world and a model. This mis-frames disagreement systematically. Parties argue about whether the event was outrageous when what actually differs is which commitment each held, and the argument is unresolvable in those terms. Cross-cultural contact, contract disputes, and code review all reward surfacing both models instead of adjudicating the occurrence, yet the phenomenology pushes hard the other way every time, because the felt location of a breach is stubbornly external.

T5 — Silence is under-determined. A model that registers nothing may be accurate, may not actually be in force, or may be tuned so loosely that no realization could contradict it. These are radically different conditions and they are indistinguishable from outside without a deliberate probe. Organizations read long quiet periods as evidence of control, and the reading stays unfalsifiable until an event arrives large enough to breach even a badly widened band — by which point the model has been vacuous for a long time.

T6 — Sharp models read out best and resist correction hardest. A commitment that rules out a great deal produces informative breaches, because whatever contradicts it says a lot. The same sharpness makes its bearer brittle in precisely the environments where it is wrong, and a model confident enough to be diagnostically valuable is confident enough to reclassify inconvenient realizations as measurement error rather than register them. Diagnostic power and openness to correction are produced by the same property, sharpness of commitment, and they pull in opposite directions, so nothing about tuning one settles how to tune the other.

Structural–Framed Character¶

Expectation Violation sits at the midline of the structural–framed spectrum — mixed-framed, aggregate 0.5, with every diagnostic at 0.5. The skeleton underneath is four-place: an active prediction, script, schema, or learned regularity; a realization comparable with what that model anticipated; a mismatch salient enough to register rather than be assimilated as noise; and a response — surprise, processing difficulty, attention shift, model revision, or repair. Music, narrative, user interfaces, social scripts, anomaly detection, and perception all preserve those prediction–breach–response roles.

Human-practice-bound at 0.5 does most of the explanatory work. The prime requires an active anticipatory model, and conventions and social scripts, sustained by communities, are among the models that can be breached. What keeps it off the framed pole is that human surprise is one realization, not the whole structure: anomaly detectors and perceptual systems breach without deliberation.

Vocabulary travels partially at 0.5 — schema, script, and surprise carry the accent of fields that study expectation; anomaly detection restates the roles in its own lexicon. Evaluative weight is 0.5: violation and breach imply something that should have happened, though the prime only describes a mismatch and its response. Institutional origin is 0.5, that register belonging to the study of expectation and convention rather than to a formal regularity. Import-versus-recognize is 0.5: in music or narrative one imports the breach-and-response framing; in perception one recognizes a loop already running.

Used well, the prime requires a nameable breached model: without an active prediction there is only baseline deviation, and a signed residual used for update or routing is prediction error.

Substrate Independence¶

Expectation Violation is a highly substrate-independent prime — composite 4 / 5 on the substrate-independence scale. The relational core makes no reference to feeling: a model in force before the event, a realization commensurable with the question that model answered, a mismatch large enough to cross the registration threshold instead of being absorbed as noise, and a response that spends something measurable and thereby exposes what the model held. Harmonic surprise, plot reversal, interface convention, social script, machine anomaly detection, and low-level perceptual prediction all fit without strain, human surprise being one realization rather than the entire shape. It stays below the ceiling because the statement still requires a bearer carrying a forward model; a system that anticipates nothing has no way to instantiate the pattern at all.

Composite substrate independence — 4 / 5
Domain breadth — 4 / 5
Structural abstraction — 4 / 5
Transfer evidence — 4 / 5

Relationships to Other Abstractions¶

Current abstraction Expectation Violation Prime

Parents (1) — more general patterns this builds on

Expectation Violation is a kind of Baseline Deviation Prime

Expectation violation is baseline deviation specialized to an active predictive reference and a registered violation response.

Children (2) — more specific cases that build on this

Language expectancy theory Domain-specific is a kind of Expectation Violation

The proposed strict upward parent is prime:expectation_violation.
Syncopation Domain-specific is a decomposition of Expectation Violation

Syncopation is the metric-substrate form in which an expected accent position is displaced and registered against the maintained prediction.

Hierarchy path (1) — routes to 1 parentless root

Expectation Violation → Baseline Deviation → Comparison → Self Checking

Neighborhood in Abstraction Space¶

Expectation Violation sits in a sparse region of abstraction space (67^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely rather than landing on a neighbor.

Family — Cue-Triggered Deferred Action (27 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-10

Not to Be Confused With¶

Expectation Violation must first be separated from Baseline Deviation, its parent and the neighbour it is most often collapsed into. Baseline deviation requires only that an observation be read against a declared reference and flagged as departing from it. The reference can be entirely retrospective and inert: a historical mean, a control group, a specification sheet, a value recorded last quarter. Nothing about it anticipates. Expectation violation specializes that relation by demanding that the reference be actively predictive at the moment the realization arrives, and that the departure trigger something. A quarterly report noting that spend exceeded the prior-year figure is deviation; a forecasting system that had committed to a range and now reroutes work because the range was breached is violation. The practical test is whether the reference was doing anticipatory work before the event or was fetched afterwards to characterize it.

The confusion with Prediction Error runs the other way and is subtler, because both involve a model that anticipates. Prediction error is a signed residual — a quantity with a magnitude and a direction — produced so that it can be consumed downstream for updating, routing, or diagnosis. It is arithmetic that happens continuously and mostly at zero interest. Expectation violation requires that a mismatch cross a salience threshold and provoke a registered reaction, and it does not require that the mismatch be signed, scalar, or even measurable. A gradient-descent step consumes millions of prediction errors per second, none of which is a violation of anything; a reader hitting an impossible continuation registers a breach without any quantity being computed. Related but distinct is Predictive Coding, which is an architectural commitment about how a system is built — predict the input, propagate only the residual — rather than a claim about an event. A predictive-coding system is a plausible mechanism for producing violations, not an instance of one.

Reward Prediction Error narrows the prediction-error family further by fixing the currency: the residual is specifically about received versus expected value, and its function is to teach policies and update value estimates. Expectation violation is indifferent to value. A breach that carries no reward implication at all — a chord that resolves elsewhere, a sentence whose subject turns out to be the object, a colleague using an unexpected greeting — is fully within scope. When an event is both, the two abstractions describe different aspects of it: the reward residual explains what the bearer will do more or less of next time, while the violation explains what became visible about the model that was in force.

Tension and Release shares the machinery but adds commitments that expectation violation does not carry. Tension and release requires deliberate construction: instability is built on purpose, sustained long enough for anticipation to accumulate, and then resolved in a way whose payoff is earned by the buildup. That is a single coupled cycle with an author and an intended arc. Expectation violation needs no author, no sustaining phase, and no resolution: most breaches are accidents, and many are never resolved at all. A composer withholding a cadence for sixteen bars is doing both; a power supply failing in a way nobody anticipated is doing only the latter.

Salience is often mistaken for the same phenomenon because breaches tend to be salient. But salience is a bottom-up, relational score by which an item stands out from its surround before attention is deliberately allocated, and it can be driven entirely by contrast — a bright patch, a loud onset, a moving object among still ones — with no anticipatory model anywhere in the account. Salience explains why something is picked up; violation explains why something contradicts. High-salience events routinely breach nothing, and low-salience breaches are exactly the ones that damage systems, because the response they trigger is small enough to be waved through.

Finally, two neighbours describe what happens to expectations over time rather than what a breach is. Habituation is the progressive decline of a response to a repeated stimulus that proves inconsequential; it removes the response while the stimulus continues, which is one route by which something that used to be a violation stops being one. Normalization of Deviance describes the standard itself drifting outward as small departures, repeatedly observed without immediate consequence, get reclassified as acceptable. Habituation attenuates the reaction; normalization of deviance moves the commitment. Both are processes acting on the terms of a violation, not alternative accounts of the violation itself.

Solution Archetypes¶

No catalogued solution archetypes reference this prime yet.

References¶

[1] Meyer, Wulf-Uwe, Rainer Reisenzein, and Achim Schützwohl. "Toward a Process Analysis of Emotions: The Case of Surprise". Motivation and Emotion, 1997. Sets out the schema-discrepancy account in which an active schema, a discrepant event, a discrepancy signal crossing a threshold, and an interruption of ongoing processing with measurable cost are separate consecutive stages. registry ↩

[2] Garfinkel, Harold. Studies in Ethnomethodology. Prentice-Hall, 1967. Establishes the breaching demonstration, in which contradicting a tacit background expectancy makes its content observable through the repair work that follows. registry ↩

[3] Reisenzein, Rainer, Gernot Horstmann, and Achim Schützwohl. "The Cognitive-Evolutionary Model of Surprise: A Review of the Evidence". Topics in Cognitive Science, 2019. Reviews evidence that surprise is evoked by schema-discrepancy alone, with the novelty and the valence of the eliciting event having no independent effect. registry ↩

[4] Fischhoff, Baruch. "Hindsight is not equal to foresight: The effect of outcome knowledge on judgment under uncertainty". Journal of Experimental Psychology: Human Perception and Performance, 1975. Shows outcome knowledge inflates the judged prior probability of what happened, so a model reconstructed after the event no longer excludes it and cannot be breached by it. registry ↩

[5] Green, David M., and John A. Swets. Signal Detection Theory and Psychophysics. Wiley, 1966. Separates the observer's decision criterion from sensitivity, establishing that whether a discrepancy is registered depends on a threshold held independently of the magnitude of the discrepancy. registry ↩

[6] Reber, Arthur S. "Implicit learning of artificial grammars". Journal of Verbal Learning and Verbal Behavior, 1967. Shows learners classify grammatical strings well above chance while unable to state the regularity they are applying. registry ↩

[7] Burgoon, Judee K. "Interpersonal Expectations, Expectancy Violations, and Emotional Communication". Journal of Language and Social Psychology, 1993. Treats a violation as an enacted deviation from what was expected that may run positively or negatively, so the relation itself carries no verdict on the expectation that was held. registry ↩

[8] Baillargeon, Renée, Elizabeth S. Spelke, and Stanley Wasserman. "Object permanence in five-month-old infants". Cognition, 1985. Establishes the violation-of-expectation method in which looking time to an impossible outcome is treated as the readout of a model the infant cannot report. registry ↩

[9] Chandola, Varun, Arindam Banerjee, and Vipin Kumar. "Anomaly Detection: A Survey". ACM Computing Surveys, 2009. Describes scoring each instance against a model of normal behaviour and states that the analyst selects anomalies by applying a domain-specific cutoff threshold to that score. registry ↩

[10] Bruner, Jerome S., and Leo Postman. "On the perception of incongruity: A paradigm". Journal of Personality, 1949. Shows an identical stimulus is registered differently according to the expectation in force, locating incongruity in the pairing rather than in the event. registry ↩

[11] Popper, Karl R. The Logic of Scientific Discovery. Hutchinson, 1959. Establishes the asymmetry by which confirming instances add nothing while a contradicting instance bounds where a model stops holding. registry ↩

[12] Kutas, Marta, and Kara D. Federmeier. "Thirty Years and Counting: Finding Meaning in the N400 Component of the Event-Related Brain Potential (ERP)". Annual Review of Psychology, 2011. Shows the latency and graded amplitude of the response to a discrepant word read back the content and the strength of the prediction that preceded it, which is the backwards inference the diagnostic depends on. registry ↩

[13] Vaughan, Diane. The Challenger Launch Decision: Risky Technology, Culture, and Deviance at NASA. University of Chicago Press, 1996. Documents normalization of deviance, in which repeated unpunished departures move the standard until identical events stop registering. registry ↩a ↩b

[14] Huron, David. Sweet Anticipation: Music and the Psychology of Expectation. MIT Press, 2006. Argues tonal expectations are acquired by statistical exposure to a corpus, so an identical acoustic event breaches nothing for an unexposed listener, and treats the resulting responses as hedonically various rather than aversive. registry ↩