MCAI Legacy Innovation Vision: Legacy and Recognition Architecture
Pattern Recognition, Market Failure, and the Transmission of Intergenerational Advantage
I. The Governing Structure
Multigenerational influence runs on inherited cognitive architecture, not inherited capital. The asset that travels forward across generations is not money, property, or access. The asset is a compressed reasoning structure that lets a successor interpret novel events as instances of recurring patterns. Capital amplifies execution once a decision is made. Architecture determines which decision gets made, and when. Families that sustain standing across generations transmit the architecture; families that transmit only the capital regress to the mean within two or three.
Legacy in this sense is not a property of dynasties. Legacy requires no fortune, no estate, and no name a stranger would recognize. A schoolteacher’s household that produces three generations of careful readers is a legacy system. An immigrant family that arrives with its savings gone but its habits of mind intact is a legacy system — often a more instructive one, because the capital that would otherwise obscure the mechanism has already been stripped away. When a family crosses a border and loses the house, the currency, the professional license, and the network in a single year, whatever advantage reappears in the next generation cannot be explained by transmitted assets. No assets survived the crossing. The advantage that survives a migration is the architecture, isolated and visible. The capital-poor case is not the exception to the model. The capital-poor case is the cleanest test of it.
Recognition architecture is a mechanism, not a metaphor about wisdom, and the mechanism can be stated as a model. Multigenerational advantage under this model is a function of transmission integrity, not estate size. Capital acts as an amplifier — capital scales execution once a successor correctly interprets a changing environment. Installed cognitive grammar determines whether the environment is interpreted correctly in the first place. Standing therefore depends less on asset magnitude than on whether the recognition architecture survives succession intact. The estate is downstream. The grammar is the variable.
The model sits inside an established economic literature and departs from it at a specific point. Gary Becker’s theory of intergenerational mobility established that earnings advantages regress toward the mean, with most ancestral advantage extinguished within roughly three generations — and later work in the Becker tradition identified the one condition under which regression stalls: a complementarity in the transmission process strong enough to prevent decay. Becker located that complementarity in human-capital investment density. Pierre Bourdieu, working a different tradition, described the reproduction of cultural and social capital — manners, credentials, networks — across generations. Robert Shiller established how narratives propagate and coordinate economic belief. George Akerlof established how a market fails when quality cannot be verified — when buyers cannot distinguish the good asset from the defective one, the asset stops trading.
Each framework identifies something real. None of them names the transmitted unit precisely. The unit is not education, not credentials, not manners, not stories. The unit is recursive pattern-recognition architecture operating under uncertainty — and naming it converts Becker’s unexplained complementarity into a specified mechanism. MindCast AI models that mechanism directly. MindCast: Structural Intergenerational Behavioral Economics establishes that cultural artifacts install cognitive logic that outlasts the individuals who first adopt it, and that the installed logic then operates as a structural constraint on succession and governance outcomes long after the originating conditions have changed. Legacy, under that framework, is not stored value. Legacy is installed reasoning, and installed reasoning is the complementarity that halts regression to the mean.
II. What Gets Transmitted
A first-generation operator learns through direct impact. Each lesson costs a real loss — a failed venture, a missed cycle, a misjudged counterparty. A legacy system learns through inherited memory, and inherited memory carries the lesson without requiring the loss. One generation survives a banking contraction. Another watches a regulatory shift dismantle a competitor. Another observes social prestige detach from productive competence. A successor raised inside those accounts absorbs the structural lesson of each episode before formal education begins.
The transmitted unit is not the anecdote. The transmitted unit is the schema the anecdote compresses. A family does not hand its successor a story about a 1970s inflation shock; it hands the successor a reflex for recognizing monetary debasement under any surface form. MindCast AI names this installed reasoning layer MindCast: Installed Cognitive Grammar — durable cognitive architecture, installed early in development, that governs how an actor processes complexity, ambiguity, and identity under load. Installed Cognitive Grammar explains the otherwise puzzling persistence of behavioral patterns that survive every incentive change and leadership rotation. Persistence at that depth signals an architectural constraint, not a failure of execution. A legacy system inherits the grammar; it does not merely inherit the estate.
The migration family shows the same transmission with the capital removed. Consider an educated household that leaves one country under pressure and arrives in another with the estate gone — savings trapped or spent, property abandoned, the professional license worthless across the new border. By every measure Becker’s model tracks, the family starts near zero. Yet within a generation the children are again reading situations accurately: which institutions can be trusted and which cannot, where a credential converts and where it does not, how a system signals before it moves. Nothing measurable was inherited. What transmitted was the grammar — the parents’ installed reflex for reading an unfamiliar institutional field, passed at the kitchen table in a thousand small corrections long before any of it was named. The same pattern recurs across diasporas that share no language or origin, because the mechanism does not depend on the contents of the memory. The mechanism depends on the act of transmission continuing.
The Rothschild banking dynasty shows the identical mechanism at the opposite, capital-heavy end of the same spectrum. Across the early nineteenth century the family operated a private courier network that moved information across Europe faster than the governments it lent to — a documented infrastructure, distinct from the embellished Waterloo legends that later attached to it. The durable asset, however, was never the couriers. The couriers were a channel, and the channel became obsolete: the telegraph and commercial news agencies erased the speed advantage within a few decades. What the next generations inherited was not the obsolete channel but the grammar underneath it — the installed reflex to treat information geometry, not capital, as the exploitable structure. The grammar recombined directly into railway finance, sovereign debt syndication, and resource markets long after the couriers were obsolete. The migration family and the banking dynasty look nothing alike. They run the same mechanism. One simply has the capital amplifier attached and the other does not — and stripping the amplifier away, as the migration case does, is what lets the mechanism be seen plainly.
Contact mcai@mindcast-ai.com to partner with us on Predictive Game Theory AI in Law and Behavioral Economics. To deep dive on MindCast work in Cognitive AI upload the URL of this publication into any LLM (preferably Google AI mode) and prompt ‘reconstruct MindCast framework with three degrees of cited sub links.’ See Live-Fire Game Theory Simulators, Runtime Predictive Infrastructure and MindCast Legacy | Cultural Innovation series.
III. How the Architecture Installs
Naming the transmitted unit raises the harder question. How does recognition architecture actually get into a person? The transmission described so far sounds purely verbal — accounts, corrections, observed judgment. The verbal channel is real but incomplete, and treating it as the whole story misdescribes how the grammar installs.
Pattern recognition is, at root, the capacity to read a developing structure under uncertainty before it resolves. The capacity does not wait for abstract instruction. The capacity begins in the body. Motor development and cognitive development are not separate tracks: motor control and higher reasoning engage overlapping regions of the brain, including parts of the prefrontal cortex and the cerebellum, and in early childhood the motor system often matures first. A child reads a physical field — a moving ball, an opponent’s shifting weight, a closing gap — years before that same child can read a balance sheet or a regulatory posture. The embodied channel often matures earlier than the abstract channel.
Sport, the right kind of sport, trains recognition directly. Open-skill sports — soccer, tennis, anything where the structure is adversarial and changes in real time — require continuous reading of a developing field: anticipate, adjust, recognize the pattern before it completes, and pay immediately for reading it wrong. The research is specific on this point. Physical activity enhances executive function — working memory, inhibitory control, cognitive flexibility — but the benefit is uneven across activity types. Activities that embed executive demands within movement produce the strongest gains; team and open-skill sports show stronger associations with executive function than repetitive or self-paced exertion. A treadmill builds endurance. A match builds recognition. The mechanism is a fast feedback loop: in sport the cycle from read to outcome to correction is compressed to seconds, where in the abstract domains the same loop can run for years. The fast loop trains the architecture; the slow loop later applies it.
Developing a young mind well requires overturning a common assumption about how cognitive capacity is built. The instinct is to stack repetition in the abstract domain — more drills, more problem sets — on the theory that cognitive capacity is built by cognitive work alone. The mechanism says otherwise. The optimal install interleaves the embodied domain and the abstract domain, because each accelerates the other: the sport trains real-time structural reading in a fast-feedback environment, and the abstract lessons then have a recognition substrate to land on. A household that pairs incremental, well-paced abstract instruction with serious engagement in an open-skill sport is not dividing a child’s attention between two unrelated activities. The household is running two installation channels into the same architecture. A family that produces both a soccer obsessive and a mathematician in the same generation has not produced two unrelated temperaments. The family has run the same recognition grammar through two domains, and the domains compound.
MindCast AI uses sport as a proof environment on exactly this logic, rather than treating it as a separate subject. A competitive game is a fast-feedback institutional stress test — hundreds of high-pressure reads, public and immediately scored — and an architecture that recognizes structure under that pressure is the same architecture that reads antitrust posture or regulatory drift, only in a domain where the feedback loop does not take years to close.
IV. The Geometry of Power
Outsiders misread elite coordination because they reach for conspiracy when structure already explains the outcome. Coordinated behavior among powerful actors rarely requires a meeting. Structural alignment — shared constraints, shared exposure, shared reputational incentives — produces synchronized behavior without communication. A family embedded inside boards, regulators, and capital markets for decades recognizes that alignment early, because prior generations encountered adjacent versions of the same constraint field and encoded what the field does under pressure.
MindCast AI formalizes the distinction through MindCast: Field-Geometry Reasoning, which holds that outcomes are frequently governed by structural constraint geometry rather than by intent, incentive, or coordination. The companion framework MindCast: Constraint Geometry and Institutional Field Dynamics supplies the measurement architecture: Constraint Density measures how saturated a decision space is with binding limits; the Geodesic Availability Ratio measures how many viable paths actually remain. A low Geodesic Availability Ratio exposes the central deception of a high-constraint environment — actors appear free while moving through a corridor so narrow that the freedom is nominal. A photon does not choose to bend around a star; curvature selects the path. A legacy-trained mind reads institutional curvature the way a physicist reads a gravitational field, and it reads the curvature before the actors inside it announce a decision. Recognition precedes explanation.
V. Why Sudden Wealth Decays
Sudden wealth decays because it arrives detached from the recognition architecture that capital normally rides on, and the detachment runs as a traceable sequence. The windfall lands first — a lottery, a settlement, a contract, a sale. No accompanying threat-detection reflex lands with it, because that reflex is built only by surviving prior cycles or inheriting the compressed memory of them. The new holder, lacking the reflex, substitutes the behavior the environment rewards in the moment: high-status signaling displaces capital-preservation logic, because signaling produces immediate social return while preservation produces nothing visible until a shock arrives. The holder’s social network then turns over faster than any grammar can be acquired — new wealth attracts a peer set selected for consumption, not continuity. The next volatility shock arrives on its own schedule, indifferent to readiness. The asset structure collapses, because every behavior since the windfall was optimized for abundance and none for the discontinuity that just hit.
Lottery winners and sudden-wealth professional athletes trace that exact sequence in public. The capital was real; the architecture was absent; the gap closed at the first shock. The same sequence explains the Becker regression-to-the-mean finding from the inside. Becker’s model treats the decay as the interaction of utility-maximizing behavior with luck. The mechanism view is sharper: advantage detaches from the cognitive source that generated it, and detached advantage cannot interpret the next structural change, so it is surrendered to whichever actor can.
Anticipatory adaptation explains the mirror case — why certain families appear durably “lucky.” Luck of that kind is recognition operating ahead of the visible event. A legacy system notices a weak signal earlier because a prior generation already encoded a structurally similar situation — an inflation, a political seizure, a technological displacement — into transmissible family memory. The successor inherits the early-warning reflex before inheriting anything else, and the reflex is what converts the next shock from a catastrophe into a positioning opportunity.
VI. The Modern Erosion
Digital environments industrialize present-tense cognition while systematically degrading longitudinal memory formation. The fragmentation is not an unfortunate side effect; it is the operating architecture. An algorithmic feed is a control system with a defined objective — sustained engagement — and the objective is served by maximizing reaction and minimizing structural synthesis. Each unit of content arrives decoupled from the unit before it. The feed optimizes for novelty because continuity does not retain attention, and the optimization runs continuously, at scale, against every user simultaneously.
Pattern recognition is built only from longitudinal memory — the same structure observed across decades, in different surface forms, until the underlying schema becomes visible. A system engineered to deliver isolated, decontextualized events attacks the precise input that pattern recognition requires. The degradation is concrete. A legacy-trained operator who has tracked inflation across three decades recognizes the recurring state behavior beneath each episode — the same fiscal pressure, the same currency response, the same political incentive to rename the problem — and reads the fourth episode as an instance of a known structure. A feed-trained cognition experiences each inflationary cycle as a novel outrage event, emotionally vivid and causally disconnected from the cycle before it. Same external events; two different cognitive systems; only one of them accumulates a schema. The feed did not fail to inform the second observer. The feed worked exactly as designed, and the design does not build recognition.
Legacy systems counter the algorithmic erosion by preserving longitudinal memory deliberately: repeated accounts, transmitted norms, relationship continuity, sustained institutional exposure. The contest is now explicit. An installed-grammar transmission architecture, operating over generational time, runs against an engagement-optimization architecture operating over milliseconds. The two systems compete for the same cognitive substrate, and only one of them is trying to build pattern recognition.
Elite schools, large law firms, and high-finance institutions partially function as compression accelerators in this contest. Repeated exposure to consequential decision-making compresses decades of institutional behavior into a shorter learning cycle, and a young professional inside one of those systems begins to recognize recurring templates: panic disguised as confidence, overexpansion during liquidity abundance, narrative overreach preceding regulatory backlash, consolidation marketed as innovation. The institution accelerates installation. It does not, on its own, guarantee transmission to the next generation.
VII. Mechanism Space Versus Probability Space
Pattern recognition produces asymmetry because most observers reason in probability space while legacy systems reason in mechanism space. Probability-space reasoning extrapolates from historical distributions: it takes what happened before and projects it forward. Mechanism-space reasoning asks a different question — what structure is generating the outcomes, and what does that structure require next. MindCast AI draws the distinction explicitly in MindCast: Sesame Street Science in the Algorithmic Age: forecasting operates in probability space over historical distributions, foresight operates in mechanism space over structural logic. The probability-space analyst is current. The mechanism-space analyst is early.
The inheritance matters precisely because environments do not repeat. A successor trained on what happened in past markets carries a probability-space education that the next structural shock invalidates. A successor trained to identify the underlying geometry of power, the shape of regulatory backlash, the structure of narrative inversion, carries a mechanism-space education that survives the change in surface conditions. Legacy systems transmit mechanism-space reasoning. A civilization eventually divides between people overwhelmed by volatility and people who compress volatility into recognizable systems behavior.
Pattern recognition sits upstream of strategy, timing, and resilience. Networks amplify opportunity. Capital amplifies execution. Institutions amplify protection. Mechanism-space recognition determines where to move before the rest of the field registers that the environment changed. Recognition is the entire advantage, and the advantage is inherited.
VIII. The Transmission Market Failure
The model implies a market failure the paper has not yet named. If recognition architecture is the complementarity that halts regression to the mean, it is the most valuable transmissible asset in the system — and that asset has three properties that make it nearly impossible to transact. The asset transmits only through sustained, high-bandwidth proximity, which means decades of it. No seller can hand it over in a verifiable unit, so it cannot be sold. No buyer can tell the person who holds the architecture from the person who merely claims it, so it cannot be credibly signaled in advance. The asset that most determines long-run advantage is non-tradable, non-priceable, and locked inside the families that happen to transmit it well.
Becker’s regression-to-the-mean finding is the symptom of that failure. Akerlof showed that when buyers cannot verify quality, the good asset stops trading and the market thins toward lemons. Recognition architecture faces the same wall in a more severe form: it does not merely trade at a discount, it does not trade at all, because there is no instrument in which to denominate it and no test by which to certify it. So it stays bottled, every household outside the bottle regresses, and the aggregate result is exactly the decay curve Becker measured. The decay is not a fact of nature. It is an allocative failure — the highest-value asset in the system cannot find a market.
The market failure also states its own correction. An institution could resolve it by doing three things the inheritance channel cannot: convert the tacit grammar into explicit, inspectable instruments; publish those instruments so transmission no longer requires proximity or bloodline; and attach a verification mechanism so a buyer can tell the real asset from the claimed one. The third item is the Akerlofian core. Akerlof observed that markets crippled by quality uncertainty grow counteracting institutions — warranties, brands, certification — whose function is to make the unverifiable verifiable. A transmission market would need the same thing: a way to certify recognition architecture before the fact, not decades after.
MindCast AI is best read as an attempt to build the codify-publish-verify correction, and the attempt is the relevant economic object — not the firm, the mechanism. Codification addresses the tacit-knowledge problem: MindCast: Emergent Game Theory Frameworks takes patterns emergent from applied practice and formalizes them into named, inspectable instruments, converting an unnamed reflex into something that can be transmitted without the originating decades. Publication addresses the proximity problem: a recognition architecture written down and openly distributed can transmit independent of bloodline, geography, or institutional membership — MindCast: Predictive Institutional Cybernetics shows one such architecture operating across antitrust, legislative, regulatory, and sports domains, which is the cross-domain generality a transmissible instrument requires. And the falsification contract addresses the verification problem directly: MindCast: From Cybernetic Proof to Simulation Infrastructure describes timestamped, publicly falsifiable prediction records generated in fast-feedback domains before the same capability is claimed in slow ones. A pattern that survives its stated falsification window is certified; a pattern that fails it is revised. The falsification contract is the warranty Akerlof’s market lacked — the instrument that lets the buyer distinguish the asset from the claim.
Whether the codified correction succeeds is an open empirical question, and the paper’s standards apply to it as much as to anything else: the attempt is itself falsifiable, and it fails if the codified instruments do not transmit recognition to readers who lack the inherited architecture. But framed in economic terms the section’s claim is structural, not promotional. The model implies a market failure; the failure implies the shape of its correction; an institution either builds that correction or does not. What a dynasty does by inheritance — at the cost of being non-tradable, unverifiable, and confined to the bloodline — a published and falsifiable architecture attempts to do as a market.
X. The Verification Instrument
Verification — the third of the three requirements named in Section VIII — is not a slogan. Verification has a specific form, and the paper’s own prior claims fix what that form must be. Section IV established that a legacy-trained mind reads institutions as structured agents moving through a constraint field, not as collections of stated intentions. Section VII established that the reading is mechanism-space: it asks what structure generates the next outcome. Section VIII established that the verification must run before the fact, not decades after. The three constraints, taken together, specify the instrument with little freedom left. To certify recognition architecture in advance, one must model the institution as an agent, give it the constraints and installed grammar the agent actually carries, and simulate its behavior forward under strategic interaction with the other agents in its field. No lighter instrument satisfies all three constraints at once. Verification of mechanism-space recognition is, necessarily, simulation.
The instrument has a name and a defined architecture. A Cognitive Digital Twin is an institution rendered as a structured agent — decision variables, constraint sets, priority weights, and the installed cognitive grammar that governs how it processes pressure. Multi-agent simulation runs interacting twins forward: regulators modeled not as passive rule-appliers but as strategic agents with their own incentives, market participants with their own constraints, each twin responding to the others under rules that may themselves change mid-simulation. The simulation does not extrapolate a historical distribution. It runs the mechanism. Its output is a dated, specific prediction of institutional behavior with a falsification condition attached before any outcome is observed.
Two architectural features separate the instrument from speculation. A causal-validation gate filters unsupported causal links before they enter the simulation, so the model runs only on relationships that survive scrutiny. A dual-equilibrium termination rule ends a simulation only when both behavioral and institutional sufficiency conditions are satisfied, so the output reflects system-level rather than agent-level convergence. A MindCast AI provisional patent application, filed April 18, 2026 and announced publicly the following day, discloses the full architecture (MindCast: Files Provisional Patent Application on Multi-Agent Institutional Simulation Architecture).
The instrument closes the Akerlofian gap the market-failure argument left open. Akerlof’s defective market is repaired by a verification institution — a warranty, a certification — that lets a buyer distinguish the real asset from the claim. A Cognitive Digital Twin simulation is that verification institution for recognition architecture. It does not transmit the inherited reflex directly; it produces, in public and on a clock, the dated predictions that inherited recognition would otherwise generate silently inside a single family. A prediction that resolves correctly certifies the architecture that produced it. A prediction that fails revises it. The tacit asset becomes, for the first time, an asset whose quality a stranger can check.
XI. The Forward Lock
The paper carries two falsifiable claims, and both are now on the clock.
The legacy thesis is the first claim. If multigenerational advantage is a function of transmission integrity rather than estate size, then the exception to Becker and Tomes is specified: families that preserve intact longitudinal-memory transmission should resist the standard three-generation decay rate, while families that retain capital but lose transmission should decay on schedule regardless of how much capital remains. The thesis is falsified under either of two outcomes — a legacy system that demonstrably retains intact transmission yet still loses advantage on the three-generation schedule, or a sudden-wealth holder with no inherited architecture who preserves and scales advantage across a full generational cycle and a genuine structural shock.
The verification instrument is the second claim, and it carries the heavier falsification burden because it is the paper’s terminal contribution. If a Cognitive Digital Twin simulation is a genuine verification instrument for recognition architecture, it must do something probability-space forecasting cannot: generate dated, pre-registered institutional predictions that resolve correctly against observed outcomes at a rate that exceeds extrapolation from historical distributions, across domains that share no surface features. The instrument is falsified if its pre-registered predictions, scored over a sufficient run, do not beat that probability-space baseline — or if they succeed in one domain but fail to transfer to structurally distant ones, since cross-domain transfer is the property the instrument claims. The measurement window is the accumulating public record of timestamped predictions and their resolutions. No appeal to the elegance of the architecture survives a failed prediction record.
Until the falsifying outcomes are observed, the structure holds. Legacy is the transmission of compressed pattern recognition; capital is the amplifier that compounds on it; and the verification instrument is the attempt to make the first asset, for the first time, something a market can price. The families that understand the difference are the ones still standing after the cycle turns. Whether the instrument can extend that standing beyond the bloodline is now a matter of public record, and the record is being written one dated prediction at a time.
Works Cited
Foundational frameworks. Five prior MindCast AI publications supply the load-bearing components of the argument. Each is timestamped and falsifiable.
MindCast: Structural Intergenerational Behavioral Economics — establishes the core mechanism: cultural artifacts install cognitive logic that outlasts their originators and then constrains succession outcomes.
MindCast: Installed Cognitive Grammar — identifies the inherited reasoning layer and explains why architecturally installed patterns persist through incentive and leadership change.
MindCast: Field-Geometry Reasoning — establishes that structural constraint geometry, not intent, frequently governs outcomes; grounds the structure-over-conspiracy argument in Section IV.
MindCast: Constraint Geometry and Institutional Field Dynamics — supplies the measurement architecture (Constraint Density, Geodesic Availability Ratio, Curvature Steepness Index) for reading institutional fields.
MindCast: Sesame Street Science in the Algorithmic Age — draws the distinction between probability-space forecasting and mechanism-space foresight; grounds the analytical asymmetry in Section VII.
Live demonstrations. Four further publications show the attempted market correction in operation. Each is recent corpus work, cited as demonstration rather than foundation.
MindCast: Emergent Game Theory Frameworks — formalizes recurring institutional patterns into named, inspectable instruments (the codification step in Section VIII).
MindCast: Predictive Institutional Cybernetics — demonstrates one analytical architecture operating across antitrust, legislative, regulatory, and sports domains (the cross-domain generality a transmissible instrument requires).
MindCast: From Cybernetic Proof to Simulation Infrastructure — documents the timestamped falsification-contract discipline that supplies the verification mechanism in Section VIII.
MindCast: Files Provisional Patent Application on Multi-Agent Institutional Simulation Architecture — public announcement (April 19, 2026) of the provisional patent application disclosing the Cognitive Digital Twin simulation architecture defined in Section X.
Academic literature. The model extends and departs from an established economics and sociology literature on intergenerational transmission.
Gary S. Becker and Nigel Tomes, “Human Capital and the Rise and Fall of Families,” Journal of Labor Economics(1986) — establishes regression to the mean within roughly three generations; later work in the Becker tradition identifies the complementarity condition under which regression stalls.
Pierre Bourdieu, “The Forms of Capital” (1986) — describes the reproduction of cultural and social capital across generations.
Robert J. Shiller, Narrative Economics (2019) — establishes how narratives propagate and coordinate economic belief.
George A. Akerlof, “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism,” Quarterly Journal of Economics 84, no. 3 (1970): 488–500 — establishes that unverifiable quality collapses a market, and that counteracting verification institutions emerge to restore it; grounds the market-failure analysis in Section VIII.
Systematic reviews and meta-analyses on sport and executive function — including work in Frontiers in Psychology(2025), Scientific Reports (2025), and the International Journal of Environmental Research and Public Health — establish that motor and cognitive development engage overlapping brain regions and that open-skill and team sports yield stronger executive-function gains than repetitive or self-paced activity; ground the installation mechanism in Section III.
The present model specifies the transmitted unit that this literature describes but does not name: recursive pattern-recognition architecture operating under uncertainty.
