Miller's Law (7 ± 2)¶

Miller's finding that immediate human recall is capped at a small integer of chunks (~7, later corrected nearer four) — separating a bounded item count from unbounded content per item, so the only way to hold more is to recode the unit, not widen the channel.

Core Idea¶

Miller's law is George A. Miller's 1956 empirical generalisation, reported in "The Magical Number Seven, Plus or Minus Two" (Psychological Review), that the immediate span of human short-term recall and absolute judgment converges on approximately seven distinguishable items, plus or minus two, across a striking range of paradigms — digit span, letter span, word span, span of absolute judgment on unidimensional perceptual scales, and span of attention to simultaneously presented items. The finding's central conceptual contribution is the distinction between the item count that constrains short-term recall (approximately 7, a small fixed integer) and the information content each item can carry (in bits, essentially unlimited). Miller's operative unit is the chunk — whatever recoded unit the encoder has constructed — so the string "1 4 9 2 1 7 7 6 1 9 4 5" presents twelve digit-chunks to an unprepared encoder but three date-chunks to someone who recognises 1492, 1776, and 1945; the span limit applies to chunks, not to raw symbols or bits, which means encoding strategy (chunking) can dramatically expand the amount of information held within the fixed item-count budget. Subsequent research, particularly Cowan's (2001) work controlling for covert rehearsal, has revised the estimate of the true focal-attention limit to approximately four chunks; the 7 ± 2 figure reflects uncontrolled rehearsal inflating apparent span. The law nevertheless remains the canonical anchor for the broader structural claim that human immediate memory is bounded by a small integer and that the productive response to this bound is recoding (chunking) rather than attempting to expand the limit. In design practice the law is widely invoked as a heuristic to group navigation menus, instruction sequences, form fields, and category lists within a 5–9-item range so that users can hold the set in active mind; Miller himself cautioned against this direct prescriptive reading, but the heuristic has become standard in user-interface, information-architecture, and instructional-design practice. The law also motivated the Baddeley-Hitch working-memory architecture — the search for why the limit exists led to the multicomponent model of working memory with its phonological loop, visuospatial sketchpad, and central executive — making Miller's 1956 paper the historical launch point for the entire working-memory research programme.

Structural Signature¶

Sig role-phrases:

the focal-attention store — the immediate stage of human cognition that holds a small integer number of independently active items at once
the chunk — the unit the limit counts: whatever recoded grouping the encoder has constructed, not a fixed information quantum
the count-content separation — the load-bearing decomposition of item count (bounded, small) from information content per item (bits, essentially unbounded)
the small-integer cap — the empirical figure: ~7 ± 2 chunks, varying with dimensionality and confusability, the budget the count is held against
the chunking lever — the sole throughput move: since the count cannot be enlarged, more is held only by recoding into denser chunks, not by widening the channel
the non-graceful overrun — once the cap is crossed, recall loss, error, and abandonment rise sharply rather than degrading smoothly — the signature of a capacity ceiling
the rehearsal correction — the apparent 7 was inflated by uncontrolled covert rehearsal; the true focal-attention limit is nearer four — a refinement only the count-versus-content framing makes well-posed

What It Is Not¶

Not a limit on how much one can know. The law caps the count of independently active items in the immediate focal-attention stage, not the contents of long-term memory or the total a person can learn. A chess master holds the same number of chunks as a novice; expertise expands what fits without touching the budget, so reading 7 ± 2 as a ceiling on knowledge misplaces it entirely.
Not a limit of seven bits or units of information. The bounded quantity is chunks (~7), and the information content per chunk is essentially unbounded — twelve raw digits become three date-chunks for a prepared encoder. The cap is on item count, not on the bits each item carries, which is exactly the count/content separation that makes chunking productive.
Not a precise or universal constant. The figure varies with dimensionality and confusability, and the apparent seven was later traced to uncontrolled covert rehearsal inflating the count; the true focal-attention limit is nearer four. "7 ± 2" is a memorable anchor, not an exact invariant of the cognitive system.
Not a literal design prescription. The "limit menus and lists to 5–9 items" heuristic is a looser pop reading that Miller himself cautioned against extrapolating. The law is an empirical generalization about immediate human recall, not a rule that interface element counts must fall in that range — the binding constraint, when it applies, is the user's cognitive cap, not a property of the artifact.
Not the other "Miller's law." This is George A. Miller's 1956 capacity finding, distinct from the unrelated communication principle ("assume the speaker is true and imagine what it could be true of") sometimes given the same name. They come from different papers and concern entirely different things.

Scope of Application¶

Miller's law lives across cognitive psychology and the applied design fields that all bind on the same human-cognition substrate — the focal-attention component of working memory; its reach stays within that substrate, since the number itself does not travel and the portable carrier (recode units to fit whatever budget holds) is the parent chunking, not this law by name.

Cognitive-psychology canon — the home turf: the capacity-limit finding that anchored a generation of span research and motivated the search for the mechanism behind the limit, with the Baddeley-Hitch multicomponent model its direct descendant.
Educational design — chunking in lecture and worked-example sequencing, feeding Sweller's cognitive load theory as the rule for fitting new material to the working-memory budget.
UX and information architecture — the menu-length, breadcrumb-depth, and form-field-grouping heuristic (cited accurately by some, loosely by others) for keeping a set holdable in active mind.
Telephony and serial-encoding design — digit-string, area-code, and postal-code lengths landing in the 5-9-symbol range partly for span reasons.
Test and assessment design — the digit-span subscale standard in IQ, dementia screening, and developmental-cognitive assessment.
Cognitive aging and clinical neurology — span change as a sensitive marker of frontal-system dysfunction, dementia onset, and developmental disorders.

Clarity¶

Naming Miller's law makes a structural fact visible that intuition does not supply: the immediate stage of human cognition is bounded by a small integer, not by a continuous capacity that scales with effort or task difficulty. That reframing reclassifies a large class of design and learning failures — an overlong menu, an instruction sequence that won't stick, a form users abandon — as budget overruns rather than as failures of motivation, intelligence, or interest, which points the practitioner at the right lever. And because the budget is a fixed small count, the law makes chunking a first-class move: if the cap cannot be enlarged, increased throughput must come from re-coding the unit, not widening the channel.

Its most load-bearing clarification is the separation of item count from information content — the bounded quantity Miller actually measured (chunks, ~7) from the unbounded one he did not (bits per chunk). This is exactly what keeps the law from being read as a limit on how much one can know: a chess master and a novice both hold roughly the same number of chunks, but the master's chunks each carry vastly more position-information, so expertise expands what fits the budget without touching the budget's size. The sharper question the law licenses is therefore not "how many things can a person remember?" but "what is the chunk here, and can it be recoded to pack more into the same fixed count?" — and the law's own history shows the discipline this demands, since the apparent 7 was later found to be inflated by uncontrolled rehearsal to a true focal-attention limit nearer four, a correction that only the item-count framing makes askable.

Manages Complexity¶

A broad family of capacity findings — digit span, letter and word span, the span of absolute judgment on a perceptual scale, the span of immediate apprehension, the point at which a menu or instruction sequence starts shedding recall — would, taken separately, each be its own measured limit with its own units and its own task. Miller's law compresses that family to a single structural fact: the immediate stage holds a small integer number of independently active chunks. The high-dimensional question "how much can a person take in at once, across all these paradigms?" reduces to tracking one quantity, the chunk count, against one small budget. An analyst no longer carries a separate ceiling for digits versus tones versus menu items; the same item-count cap predicts the qualitative profile across them, and the only thing that varies from paradigm to paradigm is what counts as a chunk.

The decisive move that makes this a compression rather than a slogan is the separation of two quantities that the undifferentiated notion of "capacity" fuses: item count, which is bounded and small, and information content per item, which is essentially unbounded. Holding those apart converts a confusing space of results into a two-parameter account the analyst reads off directly. The bounded parameter (chunk count) is fixed regardless of expertise or material; the unbounded one (bits per chunk) is where strategy and expertise live — so the chess master and the novice are predicted to hold the same number of chunks while the master's each carry far more, and the date-recoded digit string is predicted to fit because three chunks fall under the cap where twelve did not. That separation also fixes the only productive lever: since the count cannot be enlarged, increased throughput must come from re-coding the unit (chunking), not from widening the channel — which is the single design move the law licenses for overlong menus, instruction sequences, and form-field groups, reclassified as budget overruns rather than failures of effort. And the same item-count framing is what made the law's own correction askable: the apparent seven was traceable to uncontrolled rehearsal padding the chunk count, the true focal-attention limit nearer four, a refinement that only the count-versus-content decomposition makes a well-posed question. The move is from a scatter of paradigm-specific capacity limits to one small-integer cap on chunks plus one unbounded content dimension, with chunking as the sole throughput lever and "what is the chunk here" as the one question the analyst must answer per case.

Abstract Reasoning¶

Miller's law equips the cognitive or design analyst with moves that all turn on a single act of measurement — counting chunks against a small fixed budget — and on the count/content separation that makes the count meaningful. The diagnostic move reclassifies a failure by its cause: an overlong menu that sheds recall, an instruction sequence that won't stick, a form users abandon, are read not as failures of motivation, intelligence, or interest but as budget overruns — the active item count exceeded the small-integer cap — and the inference is confirmed by the non-graceful signature, since a capacity ceiling predicts that error rates and abandonment rise sharply once the budget is crossed rather than degrading smoothly. The analyst reasons FROM "recall collapses past a handful of items, abruptly" TO "the immediate stage is a small-integer-bounded store, and this presentation overran it," and the corrective lever is fixed in advance.

That lever is the interventionist move, and the law constrains it to one option. Because the cap is a count that cannot be enlarged by effort, increased throughput must come from re-coding the unit — chunking — not from widening the channel: group the navigation items, sequence worked examples into recodable units, batch the form fields. The analyst predicts that the same raw material, recoded into fewer chunks, fits where it did not before, and reasons that no amount of motivation or training will raise the underlying count, so a design that demands holding more than the budget allows will fail regardless of user quality. The predictive move exploits the separation directly: holding the chunk count fixed, expertise lives entirely in bits-per-chunk, so the analyst predicts a chess master and a novice hold the same number of chunks while the master's each carry far more position-information — the master's advantage is denser chunks, not a larger budget — and predicts the date-recoded digit string fits because three chunks fall under the cap where twelve did not. The boundary-drawing move fixes what the law does and does not bound: it caps the count of independently active items, not the content each carries or the total one can know, so it must not be read as a limit on knowledge or long-term memory; it applies to the immediate focal-attention stage, and the figure itself is conditional on what counts as a chunk and on whether covert rehearsal is controlled. The law's own history is the exemplar of this discipline: the apparent seven was traceable to uncontrolled rehearsal padding the chunk count, with the true focal-attention limit nearer four — a correction that is a well-posed question only under the count-versus-content framing, since one can ask "how many chunks, rehearsal removed?" but not "how many bits is the limit?" The single operative question the law forces per case is therefore "what is the chunk here, and can it be recoded to pack more into the same fixed count?" — and every move above is an answer read off that question.

Knowledge Transfer¶

Within cognitive psychology the law transfers as mechanism, across every setting that rests on the same human-cognition substrate — the focal-attention component of working memory. The small-integer cap on independently active chunks, the load-bearing count-versus-content separation, and the single throughput lever (recode the unit, do not try to widen the channel) all carry intact. In the cognitive-psychology canon it anchored a generation of capacity-limit work and motivated the search for the mechanism behind the limit — the Baddeley-Hitch multicomponent model is a direct descendant. In educational design it informs chunking in lecture and worked-example sequencing and feeds Sweller's cognitive load theory. In UX and information architecture it is the menu-length, breadcrumb-depth, and form-grouping heuristic. In telephony and serial-encoding design it shaped digit-string lengths. In test and assessment it is the digit-span subscale of IQ and dementia screening, and in cognitive aging and clinical neurology span change is a sensitive marker of frontal-system dysfunction. The crucial point — which the entry stresses — is that in every applied case the binding constraint is the human user's cognitive cap, not a structural property of the designed artifact: the same human substrate is present throughout, so this is breadth of setting on one substrate, and the chunk-counting diagnostics and the recoding lever move without translation.

Beyond the human mind the honest reading is shared abstract mechanism, not the named law, with a sharp note that the number itself does not travel at all. The portable carrier is chunking — the strategy of recoding low-level items into higher-level units to fit a budget — which is a more general structural prime (and the one Miller's law exists to motivate). When Miller's-law-style reasoning is exported to other domains, what travels is the chunking move, not the figure: encoding schemes, numbering plans, and UI groupings all choose their own caps based on their own task profiles, and computers, ecosystems, organizations, and markets do not have a 7 ± 2 cap — they have different capacity limits with different mechanisms. So the cross-domain lesson ("recode units to fit whatever the budget is") should carry chunking (with attentional_capacity / cognitive_load for the budget itself), not "Miller's law" by name. What stays home-bound is everything specific to the law: the integer ~7 (itself later corrected to a true focal-attention limit nearer four once uncontrolled rehearsal was removed — a refinement only the count-versus-content framing makes a well-posed question), the focal-attention organ it measures, and the chunk as the unit of the human immediate store. Stripped of vocabulary the law is "people can hold roughly seven things in mind at once," a content-specific fact about human cognition, not a structural commitment of broad scope. So invoking "Miller's law" or "7 ± 2" for a non-human buffer (a cache, a register file, a queue) is at best instrumental citation and at worst (A) analogy — it borrows a memorable integer that was never about that substrate. The discipline is to carry chunking wherever throughput must come from recoding against a fixed budget, to let each substrate set its own cap, and to reserve "Miller's law" for the human focal-attention span it actually measures (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

George Miller's 1956 paper "The Magical Number Seven, Plus or Minus Two" assembled span results across disparate paradigms — digit span, letter span, and the span of absolute judgment on unidimensional perceptual scales such as pitch or loudness — and found them all clustering near seven distinguishable items. Its defining conceptual move is the worked recoding example. Present the string 1 4 9 2 1 7 7 6 1 9 4 5 to an unprepared listener and it is twelve independent digit-chunks — past the span, so recall fails. Present it to someone who recognizes 1492, 1776, and 1945 and it collapses to three date-chunks, comfortably inside the budget. The raw material is identical; only the unit changed. Twelve items over budget become three under it, and the same fixed count of three now carries four digits apiece.

Mapped back: The listener's immediate recall is the focal-attention store; the date is the chunk, the recoded unit the limit counts. That twelve digits and three date-chunks are the very same symbols is the count-content separation — content held fixed while item count drops from twelve to three. Seven is the small-integer cap the counts are read against, and the digit-to-date recoding is the chunking lever: throughput rose by re-coding the unit, not by widening the channel.

Applied / In Practice¶

The North American Numbering Plan is a long-running deployment of the chunking lever. A local number is seven digits, written and spoken 3–4 (for example 867-5309); the full number prepends a three-digit area code, grouped again. Seven raw digits sit near the top of immediate span and are error-prone to hold as an undifferentiated string, so the hyphenation recodes them into two chunks that fit comfortably under the cap — a caller holds "867" and "5309" rather than seven separate digits. The grouping does not widen anyone's memory; it lowers the chunk count of the same material, which is the only move the law licenses. The same logic drives how credit-card numbers are printed in four groups of four and how postal codes are visually spaced.

Mapped back: The caller trying to hold the number is the focal-attention store; "867" and "5309" are the chunks. The digits are unchanged whether written 8675309 or 867-5309 — the count-content separation — but the grouping cuts the item count from seven toward two or three, under the small-integer cap. Hyphenation is the chunking lever applied by the designer: recoding the unit rather than expecting callers to enlarge their span.

Structural Tensions¶

T1: The memorable number versus the structural claim (the part everyone cites is the part that does not travel). "Miller's law" is famous for its integer — 7 ± 2 — and that figure is exactly what carries no weight outside the human focal-attention store: it was itself corrected to a true limit nearer four once uncontrolled rehearsal was removed, and a cache, a register file, or a queue has its own capacity with its own mechanism and no reason to land near seven. What does the structural work is the count-versus-content decomposition and the recoding lever, not the number. The tension is that the law's memorability rides on the very quantity that is least portable and least exact, so the more the seven is quoted, the more the durable content (chunking against a fixed budget) is obscured behind a suspiciously round anchor. Diagnostic: Is the claim doing work here the recode-against-a-budget structure, or just the borrowed integer seven?

T2: Fixed budget versus unbounded throughput (why "how much can you hold" has no answer). The count of active chunks is bounded and small; the information each chunk carries is essentially unlimited. Held together these make capacity simultaneously rigid and elastic — the item budget cannot be enlarged by effort, yet the amount of information held under it can balloon through recoding, so twelve raw digits and three date-chunks are the same symbols at wildly different loads. The tension is that "how much can a person hold?" is ill-posed until the chunk is specified: quote a fixed span and you imply a hard knowledge ceiling that does not exist; emphasize the unbounded content and you lose the real constraint that only a handful of units stay active at once. Both halves are true and neither alone describes the system. Diagnostic: Am I treating capacity as a fixed count or an elastic throughput — and have I said what counts as a chunk before answering?

T3: Empirical generalization versus design prescription (the heuristic Miller disowned). The law is a measured regularity about immediate human recall; the "keep menus and lists to 5–9 items" rule is a looser prescriptive reading that Miller himself cautioned against, because the binding constraint, when it applies, is the user's cognitive cap, not a property of the artifact. The tension cuts both ways: the heuristic is genuinely useful — grouping navigation, batching form fields, and sequencing worked examples really do reduce budget overruns — yet applied as a literal element-count rule it fires where no span limit is engaged (a scannable list the user never holds in mind) and misses where one is (a sequence that must be retained across steps). The same law both licenses the design move and forbids reading it as a fixed numeric constraint on interface counts. Diagnostic: Does this interface actually require the user to hold the set in active mind, or am I applying a 5–9 count rule to items that are merely displayed?

T4: Apparent span versus pure capacity (which number is the limit depends on rehearsal). The measured seven and the focal-attention four are not rival estimates of one quantity; they measure different things. Uncontrolled covert rehearsal pads the apparent count, so seven is the functional span a person can actually deploy, while four is the pure focal-attention limit that survives once rehearsal is stripped. The tension is that the organism's own augmentation strategy — silent rehearsal — is both a confound to be controlled for scientific purposes and a real capability the user brings to real tasks. Report four and you describe the mechanism but understate what people functionally hold; report seven and you describe behavior but attribute to raw capacity what rehearsal supplied. The correct number depends on whether you are modeling the store or predicting performance. Diagnostic: Do I need the rehearsal-controlled focal-attention limit (~4) or the functional span the user can actually sustain (~7) for the task at hand?

T5: Chunking as the sole lever versus its hidden precondition (recoding presupposes the schema). Because the count cannot be widened, the only throughput move is denser chunks — and the law treats this as the productive response. But recoding is not free: it works only when the encoder already possesses the higher-level unit in long-term memory. The date-string collapses to three chunks for someone who recognizes 1492, 1776, and 1945; the chess master's chunks are dense because of years of pattern learning the novice lacks. The tension is that the fix the law licenses presupposes exactly the expertise the struggling user often does not have, so "just chunk it" quietly offloads the burden onto prior knowledge. A design that relies on chunking assumes the user's schemas are already in place; where they are not, the only sanctioned lever is unavailable. Diagnostic: Does the user already hold the higher-level units this recoding requires, or am I assuming a schema they have not yet built?

T6: Autonomy versus reduction (its own named finding or the human instance of its parents). Miller's law is a canonically studied, historically pivotal result — it anchored a generation of span research and launched the Baddeley-Hitch working-memory program — with its own organ (the focal-attention store), its own unit (the chunk), and its own signature figure. Yet stripped of vocabulary it is "people can hold roughly seven things in mind at once," a content-specific fact about human cognition, and what actually travels cross-domain is the parent chunking (recode low-level items into higher-level units to fit a budget), with attentional_capacity / cognitive_load supplying the budget itself. Every other substrate sets its own cap by its own mechanism, so exporting "7 ± 2" to a non-human buffer borrows a memorable integer that was never about it. The tension is between a standalone law that earns its own study and the recognition that its portable cargo already belongs to chunking. Diagnostic: Resolve toward the parents (chunking, attentional_capacity) when reasoning about any fixed-budget recoding; toward Miller's law when diagnosing the human focal-attention span it actually measures.

Structural–Framed Character¶

Miller's law is mixed-structural on the structural–framed spectrum — a genuine, evaluatively neutral, recognized-in-nature regularity of the human mind, stopped short of the structural pole only by vocabulary (indeed by a number) that pins it to its home substrate. It is closely analogous in profile to how isostasy or the Baldwin effect is characterized: real mechanism, no verdict, no institutional origin, but non-portable operative terms. Four of the five criteria point structural. Its evaluative weight is nil — a capacity cap is neither good nor bad, "7 ± 2" praises and blames nothing, it merely reports how many chunks the immediate store holds. It is not human-practice-bound: the limit runs observer-free in every human mind whether or not any psychologist measures it — a person's immediate recall is bounded at the checkout counter and the campfire, not only in the lab; remove all the experimenters and the focal-attention store still holds only its handful of chunks. Its institutional origin is none in the constitutive sense: Miller named a fact the cognitive system already exhibits, the way one names rather than invents a natural regularity — the 1956 paper reports a property of the mind, not an artifact of a survey or a convention (the later correction toward four, once rehearsal was controlled, is exactly the mark of a real quantity being measured better). And within its proper range cross-setting reuse is recognition, not import: the same small-integer cap is recognized intact across digit span, tone span, menu recall, and clinical assessment, one human substrate throughout.

What holds it off the structural pole is the fifth criterion, vocab-travels, which it fails harder than most — and the entry stresses that the least portable thing is the very number everyone quotes. The operative vocabulary — the focal-attention store, the chunk, and above all the integer ~7 — does not float free of human cognition the way "growing quantity" or a differential equation does; a cache, a register file, or a queue has its own capacity set by its own mechanism and no reason to land near seven, so "Miller's law" exported to a non-human buffer borrows a memorable figure that was never about that substrate (import-by-analogy, not recognition). The portable structural skeleton is a single one: chunking — recoding low-level items into higher-level units to raise throughput against a fixed capacity budget, with attentional_capacity / cognitive_load supplying the budget. That skeleton genuinely travels wherever throughput must come from recoding against a cap, but it is exactly what Miller's law instantiates from those umbrella primes, not what makes "Miller's law" itself portable: the cross-substrate reach belongs to chunking (each substrate setting its own cap by its own mechanism), while the domain-accented specifics — the number seven, the focal-attention organ, the human immediate store — stay home. Its character: a real, evaluatively neutral, recognized-in-the-mind capacity regularity whose structural core is the substrate-general chunking move, but stated in a human-cognition vocabulary (and a signature integer) that pins the named law to its home domain, leaving it mixed-structural rather than a free-floating prime.

Structural Core vs. Domain Accent¶

This section settles why Miller's law is a domain-specific abstraction and not a prime, and — with no separate section for the point — carries the case for its domain-specificity as well.

What is skeletal (could lift toward a cross-domain prime). Strip the human mind away and a thin relational structure survives: a processing stage holds only a small fixed number of independently active units against a hard budget, so the only way to hold more is to recode low-level units into denser higher-level ones, not to enlarge the budget. The portable pieces are abstract — a capacity-limited store, a fixed count of active items, an essentially unbounded content per item, and a recoding lever that raises throughput within the cap. That skeleton is genuinely substrate-portable, and the entry names its catalog home precisely: chunking is the recode-to-fit move, with attentional_capacity / cognitive_load supplying the budget the count is held against. But that is the core Miller's law shares — the reason it exists is to motivate chunking — not what makes it Miller's law.

What is domain-bound. Almost everything distinctive is human-cognition furniture, and — unusually — the single most-quoted element travels least: the integer itself. The domain-bound content is the signature figure ~7 ± 2 (later corrected nearer four once covert rehearsal was controlled); the focal-attention store it measures; the chunk as the unit of the human immediate stage; the rehearsal correction that only the count-versus-content framing makes well-posed; and the applied apparatus (digit-span subscales, menu-length heuristics, numbering-plan grouping, cognitive-load sequencing). The decisive test: remove the human focal-attention organ and the number has nothing to attach to — a cache, a register file, or a queue has its own capacity set by its own mechanism and no reason to land near seven, so exporting "7 ± 2" to a non-human buffer borrows a memorable integer that was never about that substrate. Stripped of vocabulary the law is simply "people can hold roughly seven things in mind at once," a content-specific fact about human cognition rather than a structural commitment of broad scope.

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. Miller's law's transfer is bimodal. Within the human-cognition substrate — the cognitive-psychology canon, educational design, UX and information architecture, telephony, assessment, clinical neurology — the small-integer cap, the count-versus-content separation, and the chunking lever carry intact as mechanism, because every applied case binds on the same focal-attention store; the binding constraint is always the human user's cognitive cap, so this breadth is one substrate replayed, recognition of the same limit. Beyond the human mind — a cache, a queue, an organization, a market — it travels only by analogy: the number does not port at all, and each substrate sets its own cap by its own mechanism. And when the bare structural lesson is wanted cross-domain — recode units to fit whatever budget holds — it is already carried, in more general form, by chunking (with attentional_capacity / cognitive_load for the budget), the primes Miller's law was coined to motivate. The cross-domain reach belongs to those parents; "Miller's law," and above all its signature seven, carries human-cognition baggage that should stay home.

Relationships to Other Abstractions¶

Current abstraction Miller's Law (7 ± 2) Domain-specific

Parents (2) — more general patterns this builds on

Miller's Law (7 ± 2) is part of Constraint Prime

Miller's Law contains a hard small-integer Constraint on the number of independently active units admitted to immediate recall or judgment.
Miller's Law (7 ± 2) is a decomposition of Chunking Prime

Miller's Law decomposes to Chunking: recode lower-level elements into denser meaningful units to fit more information within a fixed item budget.

Hierarchy paths (4) — routes to 4 parentless roots

Miller's Law (7 ± 2) → Constraint

Show alternative paths (3)

Not to Be Confused With¶

The other "Miller's law" (the communication principle). George A. Miller's separate maxim — "to understand what another person is saying, assume it is true and try to imagine what it could be true of" — a charity principle for interpretation, from a different paper and about a different thing entirely. This capacity finding shares only the surname. Tell: is the claim about how many items immediate memory holds (the 7 ± 2 capacity law) or about how to interpret a speaker charitably (the communication principle)? Same author, unrelated content.
Cowan's limit (~4). Nelson Cowan's rehearsal-controlled estimate of the true focal-attention capacity, nearer four chunks. This is not a rival law but the corrected measurement of the same quantity Miller measured — the apparent seven was inflated by uncontrolled covert rehearsal padding the count. They measure different things: four is the pure store, seven the functional span rehearsal lets a person deploy. Tell: do you need the rehearsal-stripped capacity of the store (Cowan's ~4) or the functional span a person can actually sustain on a task (Miller's ~7)? Not two guesses at one number.
Chunking. The substrate-general strategy of recoding low-level items into higher-level units to raise throughput against a fixed budget — the parent prime Miller's law was coined to motivate and instantiates. Chunking is the portable move; Miller's law is the human-cognition finding that there is a small fixed count to chunk against (and names the figure). Tell: chunking is the recode-to-fit lever that travels to any capacity-limited store; "Miller's law" is the specific human focal-attention span — treated more fully in earlier sections — whose signature integer does not travel at all.
The Baddeley-Hitch working-memory model. The multicomponent architecture (phonological loop, visuospatial sketchpad, central executive) developed to explain why the immediate-capacity limit exists. Miller's law is the empirical finding of the limit; Baddeley-Hitch is the mechanistic model that the search for the limit's cause produced. Miller's 1956 paper is the historical launch point, not the architecture. Tell: is the claim a measured capacity regularity (Miller's law) or a proposed cognitive machinery with named subsystems (Baddeley-Hitch)?
Hick's law and Fitts's law (the sibling HCI "laws"). Other quantitative regularities of human performance routinely grouped with Miller's in design pedagogy — Hick's law relates choice reaction time to the number of alternatives, Fitts's law relates pointing time to target size and distance. All three get invoked as "the psychology laws of UX," but they bound different things: decision time (Hick), movement time (Fitts), immediate-recall count (Miller). Tell: is the constraint about how long to choose among options (Hick), how long to reach a target (Fitts), or how many items can be held in active mind at once (Miller)?
Cognitive load theory / cognitive load and attentional capacity (the budget umbrella). Sweller's instructional framework about the working-memory load imposed by learning material, and the general primes (attentional_capacity, cognitive_load) that supply the budget the chunk count is held against. Cognitive load theory is a downstream application built on the limit; the budget primes are what the limit is a specific measured value of. Tell: cognitive load theory is the design theory that uses the bound; Miller's law is the finding of the bound itself, and the budget primes are the substrate-general capacity it instantiates a human figure for.

Neighborhood in Abstraction Space¶

Miller's Law (7 ± 2) sits in a moderately populated region (46^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Cognitive Load & Processing Interference (8 abstractions)

Nearest neighbors

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