Skip to content

Working-Memory Updating

The executive function of monitoring a capacity-limited short-term store and revising its contents in real time — adding, replacing, transforming, and evicting items by relevance to the current task, distinct from passively maintaining them.

Core Idea

Working-memory updating is the cognitive executive function of actively monitoring the contents of a capacity-limited short-term store and revising those contents in real time as new task-relevant information arrives — adding incoming items, replacing stored items that have become outdated, transforming held representations, and evicting items whose relevance has lapsed — so that what is currently maintained reflects what the current task currently requires. Miyake and colleagues (2000) identified it as one of three separable executive functions alongside inhibition and set-shifting, confirmed by factor-analytic unity-and-diversity studies, and it is operationalised through n-back tasks, keep-track tasks, and letter-memory tasks that require participants to revise a maintained set rather than merely hold it.

The structural commitments are specific to the human cognitive-buffer system: a capacity of approximately four chunks, a second-by-second timescale governed by phonological and visuospatial subsystems, decay without rehearsal, and a dorsolateral prefrontal substrate indexed by neuroimaging and disrupted by frontal lesions. The updating function is logically distinct from passive maintenance (the mere retention of items over brief delays, which requires no revision), from inhibition (suppressing prepotent responses or intrusive representations), and from shifting (reallocating processing resources between task sets). Updating specifically requires an online monitoring process that evaluates the relevance of incoming information against the current task state and executes replacement or eviction decisions accordingly — the cognitive analogue of a cache manager that must balance currency of held information against capacity constraints, with the critical addition that relevance, not merely recency, governs eviction.

Structural Signature

Sig role-phrases:

  • the capacity-bounded active store — a short-term buffer of roughly four chunks, on a second-by-second timescale with phonological and visuospatial subsystems and a dorsolateral-prefrontal seat
  • the relevance-tagged arrival stream — incoming items, each evaluated for task-relevance against the current task state
  • the monitoring process — the online executive that assesses the ongoing relevance of held and incoming items against what the task currently requires
  • the update policy — the revision step: add, replace, or transform held items so contents reflect current task needs (the active operation distinguishing this from passive maintenance)
  • the relevance-driven eviction — the load-bearing commitment: items leave by relevance, not recency or decay, so a recently-arrived but now-irrelevant item can be dropped while an older relevant one is kept
  • the decay process — the background erosion of held items without rehearsal, the maintenance failure mode held distinct from eviction
  • the executive separability — the function set apart from inhibition and shifting by factor analysis, so a deficit localises to revision rather than to suppression or set-switching

What It Is Not

  • Not passive maintenance. Updating is the active revision of buffer contents — adding, replacing, transforming, evicting — not the mere retention of items across a delay, which requires no revision. The two are separable: a person can hold four items well yet fail to swap a stale one for a fresh one, so a pure hold-over-delay task is outside the updating regime.
  • Not eviction by recency or decay. What leaves the buffer is governed by relevance to the current task state, not by arrival order or the passage of time. A recently-arrived but now-irrelevant item can be the one dropped while an older, still-relevant item is kept — so "why did this item leave?" is answerable only inside the monitoring/relevance account, and a loss attributable to time alone belongs to decay instead.
  • Not undifferentiated "working memory" or "bad memory." Updating is one of three executive functions separable by factor analysis from inhibition (suppressing intrusions) and shifting (reallocating between task sets). A dynamic-memory failure localises to a specific sub-process — revision, suppression, set-switching, or maintenance/capacity — rather than collapsing into a single monolithic store.
  • Not Bayesian belief updating. Despite the shared word, this is capacity-bounded revision of items in an active buffer, not revision of probability weights over hypotheses under evidence. The constraints here are slots, decay, and task-relevance; the constraints there are priors and likelihoods, with no four-item buffer or eviction at all.
  • Not the substrate-free buffer-with-update pattern. Stripped of the ~four-chunk capacity, the second-by-second phonological loop, and the dorsolateral-prefrontal seat, the residue — a maintained active state revised as relevant information arrives, with capacity-pressured eviction — simply is the buffering / state / eviction-policy family, which the receiving disciplines already name cache, scratchpad, or situation report. "Organisational working memory" is metaphor when it imports the cognitive specifics (no organisation has a four-item limit); the portable structure travels under the parent, not this construct.

Scope of Application

Working-memory updating lives across the subfields of cognitive psychology that assay the human cognitive buffer — second-by-second, with its phonological and visuospatial subsystems and dorsolateral-prefrontal seat; its reach is bounded to that substrate (the capacity-bounded-state-revision structure travels far further under its buffering, state/state-transition, and eviction-policy parents — to caches, routing tables, agent state, situational awareness — which the receiving disciplines already name differently).

  • Developmental psychology — tracking the maturation of the updating function across childhood with n-back, keep-track, and letter-memory paradigms.
  • Aging research — age-graded decline in dynamic-memory tasks read as a monitoring/revision rather than maintenance failure.
  • ADHD research — updating deficits as a component of the executive-function profile, separable from inhibition and shifting.
  • Clinical neuropsychology — frontal-lesion and disorder-linked updating impairment indexed against the same dorsolateral-prefrontal substrate.
  • Education — working-memory-training and remediation interventions targeting the revision process rather than raw storage capacity.

Clarity

The decisive clarity updating brings is prying apart two things that the umbrella term "working memory" silently fused: the passive maintenance of items across a delay and the active revision of what is held. Before the function was isolated, a low score on a complex span task could implicate either — a person might fail because they cannot keep four items alive or because they cannot swap a stale item for a fresh one. By naming updating as a distinct executive function, separable by factor analysis from inhibition and set-shifting, the field gains a vocabulary in which "this task taxes maintenance" and "this task taxes updating" are different, testable claims. That is what lets the n-back and keep-track families be designed as updating assays specifically — they deliberately demand revision rather than mere retention — and it lets diagnostic taxonomies and training interventions target the revision process rather than the store's raw capacity.

It also sharpens the eviction question in a way that the storage view obscures. The clarifying insight is that what leaves the buffer is governed by relevance to the current task state, not simply by recency or decay: an item is dropped because monitoring has judged it no longer task-relevant, and a recently-arrived item can be the one evicted while an older, still-relevant one is kept. Holding "relevance-driven eviction" distinct from "decay" turns a vague sense that working memory "loses things" into the precise question a practitioner can now pose — is this a monitoring/relevance failure or a maintenance/decay failure? — which point to different deficits and different remedies.

Manages Complexity

The behavioral phenomena that updating is recruited to explain are heterogeneous: failures on n-back as the load climbs, the drift and intrusions of keep-track performance, the way a held set degrades when distractors arrive, age-graded and ADHD-linked decline in dynamic-memory tasks, the dissociation between people who can hold information and people who can revise it. Approached as a storage problem alone, each of these invites its own account — too few slots, too fast a decay, too weak a rehearsal loop — and the picture fragments into capacity stories that do not, by themselves, predict which task taxes a person and which does not. Updating compresses that field by recasting the buffer as a managed process with a small set of governing parameters an analyst can track: how many chunks the store holds at once (the roughly-four capacity), how fast held items decay without rehearsal, and — the parameter the storage view omits — what the monitoring process uses to decide what stays and what is evicted, namely relevance to the current task state rather than recency. Given those, the qualitative behavior of a task largely follows. The analyst reads off the branch directly: a task that merely demands retention over a delay loads maintenance and capacity; a task that demands swapping stale items for fresh ones loads the monitoring-and-revision process; and a given person's failure is diagnosable as one or the other rather than as a single undifferentiated "bad memory."

The deeper compression is that "working memory," an umbrella that silently fused holding-and-revising, is split into separable functions — updating set beside inhibition and shifting by factor analysis — so that complex, dynamic-memory behavior decomposes into a few executive contributions instead of one monolithic store. Within that decomposition the eviction question, which under pure decay has no determinate answer (why did this item leave and not that one?), becomes parameterized: track what monitoring has judged task-relevant, and the eviction reads off — a recently-arrived but now-irrelevant item can be the one dropped while an older, still-relevant one is kept. What would otherwise be a sprawl of task-specific capacity puzzles collapses onto a buffer-with-relevance-driven-revision model whose handful of parameters (capacity, decay, relevance-governed eviction, monitoring) let an analyst predict which tasks tax which sub-process and localize a deficit without re-deriving the cognitive architecture for each new paradigm.

Abstract Reasoning

Working-memory updating licenses reasoning that treats the short-term store as a managed buffer with a few governing parameters — capacity around four chunks, decay without rehearsal, and a monitoring process that evicts by relevance to the current task state rather than recency — so the cognitive scientist reasons from which parameter a task or person loads to the behavior, and back from a failure to the sub-process that produced it.

Diagnostic (localize a deficit to monitoring/revision versus maintenance/decay). The defining inference goes from a dynamic-memory failure back to the function that failed, refusing the undifferentiated "bad memory" reading. A low score on a task that demands swapping stale items for fresh ones is read as a monitoring-and-revision failure; a low score on a task that merely demands holding items over a delay is read as a maintenance/capacity failure — and because updating is separable from inhibition and shifting by factor analysis, an updating deficit is further distinguished from a failure to suppress intrusions (inhibition) or to reallocate between task sets (shifting). The eviction pattern is itself diagnostic: an item that left the buffer is read as having been judged no longer task-relevant, so observing that a recently-arrived item was dropped while an older one was kept signals relevance-driven eviction rather than decay. The inference runs task failure → the specific executive sub-process (and item-loss → relevance judgment), never task failure → "too few slots" by default.

Interventionist (load a parameter or train the process, predict the effect). Because the buffer's behavior is set by its parameters, manipulations have forecasts. Raise the rate at which task-relevant items arrive and demand revision, and updating load is predicted to rise, degrading performance for someone weak in the revision process specifically while leaving a strong-maintenance/weak-updating profile's mere-retention performance intact. Design an n-back or keep-track task and the prediction is that it assays updating rather than storage, because it deliberately requires revision; design a simple span task and the prediction is that it taxes maintenance. Target a training or remediation at the revision process and the prediction is improvement on updating-loaded tasks specifically, distinct from any change in raw capacity. Each manipulation pairs a change in which parameter is stressed with a predicted, sub-process-specific change in performance.

Boundary-drawing (revision, not retention; relevance, not decay; the cognitive buffer's specifics). The concept fixes its scope by three lines. It is logically distinct from passive maintenance — mere retention requires no revision, so a pure hold-over-delay task is outside the updating regime. It is governed by relevance, not recency or decay — so "why did this item leave?" is answerable only inside the monitoring account, where eviction follows a relevance judgment, and a loss attributable to time alone belongs to decay instead. And its quantitative commitments are those of the human cognitive buffer — roughly four chunks, a second-by-second timescale, phonological and visuospatial subsystems, a dorsolateral prefrontal substrate — which bound where the construct's parameters literally apply. Those boundaries tell the practitioner which deficits and remedies the updating account governs and which belong to maintenance, inhibition, or shifting.

Predictive / branch-ordering. From the parameter set the analyst forecasts task and person behavior before testing: a retention-only task loads maintenance and capacity; a swap-stale-for-fresh task loads monitoring-and-revision; load climbing past the roughly-four capacity predicts failure first where revision is also demanded; and eviction is predicted from what monitoring has marked irrelevant, so the item dropped is forecast by task-relevance rather than by arrival order. Which task taxes which sub-process, and which item leaves the buffer, both read off the buffer-with-relevance-driven-revision parameters.

Knowledge Transfer

Within cognitive psychology the construct transfers as mechanism, because the managed-buffer parameters (capacity ~four chunks, decay without rehearsal, relevance-governed eviction, online monitoring) and the separability of updating from inhibition and shifting apply unchanged across the subfields that use it. Developmental psychology, aging research, ADHD research, clinical neuropsychology, and education all assay the same function with the same paradigms (n-back, keep-track, letter-memory) and read the same diagnostic split — monitoring/revision failure versus maintenance/decay failure — off the same parameters. The vocabulary (executive function, updating versus maintenance, relevance-driven eviction, monitoring, n-back load) and the Miyake unity-and-diversity framing carry intact across that cluster because the substrate is constant: the human cognitive buffer, second-by-second, with its phonological and visuospatial subsystems and dorsolateral-prefrontal seat.

Beyond that cognitive substrate the entry is a clear shared abstract mechanism (B), and the boundary is unusually clean because the general pattern is already catalogued as primes. What genuinely generalises is capacity-bounded state revision — a maintained active state, an arrival stream of relevance-tagged candidates, an update-and-eviction policy, a monitoring process — which the catalogue carries as state / state-transition, buffering, and the eviction-policy family. That pattern really recurs across substrates as co-instances: database transaction state, agent state and scratchpads in software, a BGP routing table updated from advertisements, an air-traffic controller's tracked sector picture, organisational situational awareness. So when the cross-domain lesson is wanted — "hold a bounded active state, revise it as relevant information arrives, evict by relevance under capacity pressure" — it is carried by the buffering / state / eviction primes, not by "working-memory updating," whose distinctive cargo (the ~four-chunk capacity, the prefrontal substrate, the dual-task interference signatures, the n-back and keep-track measures) is human-cognitive furniture that does not transfer. Tellingly, the receiving disciplines already name the shared content differently — cache, state, scratchpad, situation report — which is the surest sign that what travels is the parent pattern, not this construct.

The failure mode to mark is the metaphor that smuggles the cognitive specifics across. "Organisational working memory" or "interface working memory" are analogy (A) when they import the construct's measurement-and-substrate content: an organisation does not have a four-item limit, a second-by-second phonological loop, or a dorsolateral-prefrontal bottleneck, so those properties do not constrain it, and treating them as if they did over-reads the construct. The legitimate version of that move drops the cognitive specifics and keeps only the buffer-with-update structure — at which point it simply is the buffering/state parent, and should be named as such. The clean boundary, then: literal transfer of working-memory updating across the cognitive subfields wherever the human buffer is being assayed; and beyond cognition, the capacity-bounded-state-revision structure travels under its buffering, state/state-transition, and eviction-policy parents (to caches, routing tables, agent state, situational awareness), not under the named cognitive function. (See Structural Core vs. Domain Accent.)

Examples

Canonical

The keep-track task (used in Miyake and colleagues' 2000 study isolating updating as a distinct executive function) is the cleanest demonstration. A participant is given a few target categories to track — say animals and metals — then reads a stream of words one at a time: table, dog, gold, chair, cat, iron, blue. The instruction is to report, at the end, the most recent word seen from each target category — here "cat" for animals and "iron" for metals. Performing this requires continuous revision: when "cat" appears it must replace "dog" in the buffer, even though "dog" is still an animal, because it has been superseded; when "iron" appears it replaces "gold." Non-target words (table, chair, blue) are never admitted, even though "blue" arrived last. The buffer holds one slot per target category and its contents are rewritten as newer exemplars arrive — mere passive retention of the whole list would fail the task.

Mapped back: The one-slot-per-category buffer is the capacity-bounded active store; the incoming word stream is the relevance-tagged arrival stream that the monitoring process checks against the target categories. Replacing "dog" with "cat" is the update policy, and dropping "dog" when superseded while never admitting the last-arriving "blue" is the relevance-driven eviction — eviction by task-relevance, not recency, exactly the commitment the storage view misses.

Applied / In Practice

Clinical neuropsychologists use updating assays like the n-back to localize memory complaints in ADHD, aging, and frontal-lobe patients. In an n-back task, a patient watches a stream of stimuli and responds whenever the current item matches the one presented n steps earlier, which forces constant revision of a running buffer rather than static holding. A patient who performs adequately on a simple digit-span (pure retention) task but falls apart as n-back load rises is diagnosed with a monitoring-and-revision deficit rather than a reduced storage capacity — a distinction that points to different remediation. This dissociation matters clinically: ADHD and dorsolateral-prefrontal dysfunction characteristically impair the dynamic updating process while sparing passive maintenance, so pairing a maintenance measure with an updating measure lets the clinician attribute the deficit to the correct executive sub-process.

Mapped back: The running n-back buffer is the capacity-bounded active store, and matching-then-revising each item exercises the update policy under the monitoring process. A patient who holds a digit span but fails n-back reveals a deficit in revision, not storage — the diagnostic payoff of the executive separability that sets updating apart from maintenance, inhibition, and shifting.

Structural Tensions

T1: Active revision versus passive maintenance (the split the umbrella term fused). The construct's central contribution is prying "working memory" into two separable things — holding items across a delay versus revising which items are held. This is genuinely load-bearing: a person can maintain four items well yet fail to swap a stale one for a fresh one, so a low complex-span score implicates one or the other, not an undifferentiated store. The tension is that the two run on the same buffer and co-occur in almost every real task, so isolating updating requires paradigms deliberately engineered to demand revision over retention, and the everyday notion of "memory" keeps collapsing them back together. Any task that merely holds items is outside the updating regime, yet looks superficially like the same faculty. Diagnostic: Does this task demand revising what is held (updating), or merely retaining it across a delay (maintenance) — and is a failure being pinned to the right one?

T2: Relevance-driven eviction versus recency and decay (what governs what leaves). The construct's sharp commitment is that items leave the buffer by relevance to the current task state, not by arrival order or the passage of time — a recently-arrived but now-irrelevant item can be dropped while an older, still-relevant one is kept. This is exactly what the storage-and-decay view cannot explain, and it makes "why did this item leave?" answerable. The tension is that in practice relevance, recency, and decay are entangled — the most recent item is often the most relevant, and unrehearsed items both decay and lose relevance — so distinguishing a relevance-driven eviction from a decay-driven loss requires engineering cases where they diverge. Attributing a loss to decay when monitoring evicted it (or vice versa) points to the wrong deficit and the wrong remedy. Diagnostic: Did this item leave because monitoring judged it no longer task-relevant, or because it decayed from disuse — and does the task separate the two?

T3: Executive separability versus unity (isolable yet correlated). The Miyake unity-and-diversity finding is that updating, inhibition, and shifting are separable by factor analysis — which licenses localizing a deficit to revision rather than suppression or set-switching — yet they also share common variance, a unity underlying the diversity. The tension is that the construct is trading on separability (updating is its own function with its own assays) while the same evidence base shows the three functions are correlated and partly common, so a "pure updating deficit" is an idealization. Over-reading separability treats updating as fully modular when it shares a general executive resource; over-reading unity collapses the very distinction that makes the diagnostic split useful. The construct lives precisely in the tension the unity-and-diversity framing names. Diagnostic: Is the deficit here specific to updating (diversity), or a manifestation of a shared executive resource that also loads inhibition and shifting (unity)?

T4: Process-pure assay versus impure task loading (localization assumes a purity the paradigms lack). The diagnostic payoff — this task assays updating, that one assays maintenance — depends on the assays being process-pure, but n-back, keep-track, and letter-memory all load maintenance, inhibition, and monitoring together to varying degrees. An n-back requires holding the running set (maintenance), suppressing lures (inhibition), and revising (updating) at once, so a failure is only presumptively an updating failure. The tension is that the construct's clean localization ("attribute the deficit to the correct sub-process") rests on tasks that are impure mixtures of the very functions it separates, so pairing a maintenance measure with an updating measure narrows but never fully isolates the culprit. Treating any single paradigm as a clean updating meter over-reads it. Diagnostic: Is this task a reasonably pure assay of updating, or does its failure also admit a maintenance, inhibition, or monitoring explanation the design has not ruled out?

T5: Raw capacity versus the monitoring process (which to target, and whether it transfers). The construct relocates the lever for dynamic-memory performance from the store's raw capacity to the relevance-driven revision process — so remediation should target monitoring-and-revision, not slot count. This is a genuine reframing with clinical stakes. But the tension is twofold: capacity and revision efficiency are hard to disentangle behaviourally (a bigger effective store eases revision, and better revision mimics more capacity), and training the revision process is exactly where working-memory-training claims are most contested — gains often stay task-specific (better n-back) without transferring to the broader executive competence the intervention was meant to lift. Targeting the process is the theoretically correct move and the one whose real-world payoff is least secure. Diagnostic: Is the intervention improving the revision process in a way that generalizes, or only lifting performance on the trained paradigm without touching capacity or transferring?

T6: Autonomy versus reduction (a cognitive function or an instance of buffering and state revision). Working-memory updating is a named, well-instrumented cognitive construct — the ~four-chunk capacity, the phonological and visuospatial subsystems, the dorsolateral-prefrontal seat, the n-back and keep-track measures — and within cognitive psychology it transfers as mechanism intact across development, ageing, ADHD, and clinical work. But strip those specifics and the residue — a bounded active state revised as relevant items arrive, with capacity-pressured relevance-driven eviction — simply is the buffering / state / eviction-policy family, which the receiving disciplines already name cache, scratchpad, routing table, or situation report. That the shared content already has other names is the surest sign the parent pattern is what travels. "Organisational working memory" is metaphor when it imports the cognitive specifics (no organisation has a four-item limit or a prefrontal bottleneck). The tension is between a cognitive function that earns its own name through substrate-specific measurement and the recognition that its cross-domain structure belongs to the buffering/state parents. Diagnostic: Resolve toward buffering/state/eviction-policy when the lesson is bounded-state revision in a cache, agent, or routing table; toward the named cognitive function when diagnosing the human buffer with its capacity, substrate, and paradigms.

Structural–Framed Character

Working-memory updating sits at the mixed-structural position on the structural–framed spectrum — well onto the structural side, alongside working memory capacity and the other cognitive-psychology entries, held off the pole by human-buffer vocabulary and a mind-bound substrate, and, like its sibling, unusually reducible because its portable pattern is already catalogued as primes. Four of the five criteria point structural. Its evaluative_weight is nil: monitoring and revising a buffer's contents is neither good nor bad, and the construct's payoff is a neutral diagnostic split (monitoring/revision failure versus maintenance/decay failure), not a verdict. Its institutional_origin is none: updating is a fact of the human executive system — an online monitoring process evicting by relevance from a capacity-limited store — identified and factor-analytically separated (Miyake), not an artifact of any convention (the n-back and keep-track paradigms are assays, not the function). And it is not human-practice-bound in the constitutive sense: the buffer is revised in real time in a person's cognition whether or not any clinician administers a task — remove every experimenter and relevance-driven eviction still operates, so nothing dissolves when the scholarly practice is withdrawn. Within its range cross-context reuse is recognition, not import: the managed-buffer parameters (capacity, decay, relevance-governed eviction, monitoring) and the separability from inhibition and shifting carry as the same mechanism across developmental, aging, ADHD, clinical, and educational subfields. The one qualification is that its substrate is a mind rather than inert nature, so it runs on minds rather than fully observer-free — but a mind is a natural substrate, which keeps it structural, merely narrower than isostasy.

What holds it off the structural pole is vocab_travels, which it fails: the operative vocabulary — executive function, updating versus maintenance, relevance-driven eviction, monitoring, n-back load, dorsolateral-prefrontal seat — is irreducibly human-cognitive furniture; "organisational working memory" is metaphor once it imports the four-item limit, the phonological loop, or the prefrontal bottleneck, none of which constrain a non-cognitive substrate.

The portable structural skeleton is capacity-bounded state revision — a maintained active state, an arrival stream of relevance-tagged candidates, an update-and-eviction policy under capacity pressure, an online monitor — and, as the entry establishes, that skeleton is precisely what updating instantiates from its parent primes buffering, state / state-transition, and the eviction-policy family, not what makes "working-memory updating" itself travel. The tell is that the receiving disciplines already name the shared content differently — cache, scratchpad, routing table, situation report — the surest sign the parent pattern is what carries. The cross-domain reach belongs to those parents; the distinctive cargo — the ~four-chunk capacity, the second-by-second phonological/visuospatial subsystems, the prefrontal seat, the dual-task interference signatures, the n-back/keep-track measures — is human-cognitive furniture that stays home. Its character: an evaluatively neutral, discovered-in-a-mind capacity-bounded-state-revision mechanism whose buffer-with-relevance-eviction skeleton is genuinely portable via buffering/state/eviction-policy, but whose distinctive capacity, substrate, and paradigms pin it to the human cognitive buffer — mixed-structural, and notably reducible, but not a prime.

Structural Core vs. Domain Accent

This section decides why working-memory updating is a domain-specific abstraction and not a prime, and it carries the case for its domain-specificity — there is no separate section for that.

What is skeletal (could lift toward a cross-domain prime). Strip the human cognitive buffer and a thin relational structure survives: a maintained active state is revised in real time as relevance-tagged candidates arrive — items added, replaced, transformed, and evicted by relevance under capacity pressure, with an online monitor deciding what stays. The portable pieces are abstract — a bounded active store, an arrival stream evaluated for relevance, an update-and-eviction policy, and a monitoring process. That skeleton is genuinely substrate-portable, which is why the entry attributes it to the catalog primes updating instantiates: buffering (the maintained active store), state / state-transition (the revised contents), and the eviction-policy family (relevance-governed dropping under capacity pressure). That capacity-bounded-state-revision core is what the construct shares with a cache, a routing table, or an air-traffic controller's tracked picture — not what makes it working-memory updating.

What is domain-bound. The distinctive content is human-cognitive furniture and none of it survives extraction intact: the ~four-chunk capacity; the second-by-second timescale with its phonological and visuospatial subsystems; decay without rehearsal; the dorsolateral-prefrontal seat indexed by neuroimaging and disrupted by frontal lesions; the executive separability from inhibition and shifting established by Miyake's factor-analytic unity-and-diversity work; and the n-back, keep-track, and letter-memory paradigms that assay it. These are the worked vocabulary, the instruments, and the empirical cases the field studies — developmental maturation, age-graded decline, ADHD profiles, frontal-lesion impairment, working-memory training. The decisive test: apply "working-memory updating" to an organisation or an interface and none of these constrain it — no four-item limit, no second-by-second phonological loop, no prefrontal bottleneck — so importing them is over-reading, and dropping them leaves the bare buffer-with-update structure, which simply is the buffering/state parent. Tellingly, the receiving disciplines already have their own names for that content — cache, scratchpad, routing table, situation report — the surest sign the cognitive specifics are what stay home. The construct is constituted by the human buffer the prime bar asks it to shed.

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. Working-memory updating's transfer is bimodal. Within cognitive psychology it transfers as mechanism intact — the managed-buffer parameters (capacity ~four chunks, decay without rehearsal, relevance-governed eviction, online monitoring) and the separability from inhibition and shifting carry unchanged across developmental, aging, ADHD, clinical, and educational subfields, which assay the same function with the same paradigms and read the same diagnostic split, so this is recognition, not analogy. Beyond cognition it does not port under its own name: database transaction state, agent scratchpads, a BGP routing table, situational awareness are genuine co-instances of capacity-bounded state revision, but they instantiate the parent primes, and the receiving disciplines already name them cache, state, or situation report — the parent pattern is what travels, not this construct. Any "organisational working memory" that smuggles the four-item limit or the prefrontal bottleneck across is mere analogy that over-reads the construct's substrate onto a system those properties do not constrain. Crucially, when the cross-domain lesson — "hold a bounded active state, revise it as relevant information arrives, evict by relevance under capacity pressure" — is genuinely wanted, it is already carried, in more general form, by buffering, state/state-transition, and the eviction-policy family. So the cross-domain reach belongs to those parents; working-memory updating is the human-cognitive instance that specializes them with a chunk-counted, prefrontal, second-by-second buffer, and its distinctive cargo is exactly the part that does not travel. It clears the domain-specific bar comfortably across the mind sciences but sits below the prime bar — indeed, like its sibling, sits noticeably close to home — because its only substrate-spanning content is already held by the primes it instantiates.

Relationships to Other Abstractions

Local relationship map for Working-Memory UpdatingParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Working-MemoryUpdatingDOMAINDomain-specific abstraction: Working Memory — presupposesWorking MemoryDOMAINPrime abstraction: State and State Transition — is a kind ofState and StateTransitionPRIME

Current abstraction Working-Memory Updating Domain-specific

Parents (2) — more general patterns this builds on

  • Working-Memory Updating is a kind of State and State Transition Prime

    Working-memory updating specializes state transition to relevance-governed additions, replacements, transformations, and evictions in a capacity-limited cognitive store.

  • Working-Memory Updating presupposes Working Memory Domain-specific

    Working-Memory Updating operates on the Working Memory architecture's active content set, replacing and transforming items by current task relevance.

Hierarchy paths (3) — routes to 3 parentless roots

Not to Be Confused With

  • Working-memory capacity. The sibling construct: capacity is the finite size of the active store (how many chunks fit at once, the supply term); updating is the executive function that revises those contents — monitoring, replacing, evicting by relevance. One is the store's size, the other the management of what fills it. Tell: is the question how much can be held simultaneously (capacity), or how the held set is kept current as relevance shifts (updating)? A large store does not guarantee good revision.

  • Passive maintenance / short-term storage. The mere retention of items across a delay, requiring no revision — what simple span assays. Updating is the active revision of buffer contents; a person can hold four items well yet fail to swap a stale one for a fresh one. Tell: does the task demand only holding items until recall (maintenance), or continuously rewriting which items are held as new relevant ones arrive (updating)? A pure hold-over-delay task is outside the updating regime.

  • Inhibition and set-shifting. The other two executive functions Miyake's factor-analytic work separates from updating: inhibition suppresses prepotent or intrusive responses; shifting reallocates processing between task sets. Updating is specifically revision of buffer contents. A dynamic-memory failure localizes to one of these, not to an undifferentiated executive. Tell: is the failure in revising held items (updating), in suppressing an intrusion (inhibition), or in switching between task rules (shifting)? Separable functions, correlated but distinct.

  • Bayesian belief updating. The shared word is a trap: this is capacity-bounded revision of items in a ~four-slot active buffer (constrained by slots, decay, and task-relevance); Bayesian updating is revision of probability weights over hypotheses under evidence (constrained by priors and likelihoods), with no buffer or eviction at all. Tell: is what changes the contents of a limited store (working-memory updating), or the credences assigned to hypotheses given data (Bayesian updating)? Different objects, different constraints, same verb.

  • Cache eviction by recency (LRU). The computing policy in which the least-recently-used item is dropped when a cache fills — eviction governed by recency/time. Working-memory updating's load-bearing commitment is that eviction is governed by relevance to the current task state, so a recently-arrived but now-irrelevant item can be dropped while an older relevant one is kept. Tell: does the buffer drop the oldest/least-recently-touched item (LRU/recency), or whatever monitoring judges no longer task-relevant regardless of arrival order (updating)?

  • The parent primes it instances (buffering, state / state-transition, and the eviction-policy family). The substrate-neutral core — hold a bounded active state, revise it as relevant information arrives, evict by relevance under capacity pressure — which is what travels to caches, routing tables, agent scratchpads, and situational awareness. Tellingly, those disciplines already name it cache, state, or situation report — the surest sign the parent, not this construct, is what carries. Tell: strip the four-chunk capacity, the phonological loop, and the prefrontal seat and the residue simply is these parents (treated more fully elsewhere); "organisational working memory" importing a four-item limit is metaphor for them.

Neighborhood in Abstraction Space

Working-Memory Updating sits in a crowded region of the domain-specific corpus (34th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Memory Encoding & Retrieval Effects (22 abstractions)

Nearest neighbors

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