♟️⚡MCAI Economics Vision: Predictive Game Theory Meets the Era of AI — Operationalizing Fudenberg's Research Agenda with Cognitive Digital Twins
Dynamic Predictive Game Theory Series: From Equilibrium Prediction to Adaptive Coherence Equilibrium
Companion to Dynamic Predictive Game Theory From the 2026 🏈Super Bowl and ⚽World Cup
The MindCast Dynamic Predictive Game Theory Collection: MindCast AI Emergent Game Theory Frameworks | MindCast Dynamic Game Theory— Competing Inside a System That Rewrites Itself | Cybernetic Game Theory | How MindCast Game Theory Differs from Textbook Game Theory | How MindCast Evolves the Structural Gaps in Classical Nash Game Theory
Executive Summary
In 2011, Drew Fudenberg circulated a research-agenda white paper titled Predictive Game Theory. The paper’s premise was a candid admission from one of the field’s leading theorists: writing down a game and characterizing its Nash or subgame-perfect equilibria “is only sometimes a good approximation of observed behavior.” Fudenberg called for a new program — models that predict how people actually play, in the lab and in the field — and listed the open questions: What determines first-time play in an unfamiliar game? How do people update their play from observation? What theories of mind guide strategic behavior? How do people prune enormous game trees? When does play converge to equilibrium at all, and to which one? How should any of it be validated empirically?
Fudenberg posed those questions before the tools to answer them existed. The thesis of this paper runs in one paragraph: Fudenberg asked how game theory could predict actual play rather than merely characterize equilibrium. MindCast extends that agenda into the AI era by adding the moving-game layer — strategic contests do not merely contain actors learning inside fixed rules. Goals, rulings, rate decisions, regulatory interventions, and market shocks can replace the game itself. The central predictive question therefore changes from “How will actors play?” to “Which game will exist next, and which actor will remain coherent when it arrives?”
Large language models, actor-specific behavioral simulation, and public real-time grading arrived a decade after the agenda. The MindCast AI Proprietary Cognitive Digital Twin Foresight Simulation (MP CDT FS) architecture and Dynamic Predictive Game Theory now operationalize the agenda in the field — not by solving each question econometrically, but by running the program Fudenberg described: combining behavioral economics and psychology with formal modeling tools from economics and computer science, and validating the result against public record. The paper below takes Fudenberg’s agenda item by item, shows what the AI era changed, identifies the one question the agenda left unasked — the question Dynamic Predictive Game Theory exists to answer — and introduces the equilibrium concept the mutating game requires: Adaptive Coherence Equilibrium, where stability attaches to the decision architecture rather than the strategy profile. No familiarity with the academic literature or any cited MindCast publication is required to follow the argument.
The paper serves four readers at once. Institutional decision-makers — litigators, investors, regulatory strategists, and risk officers — will find an equilibrium concept built for their actual contests: rulings, agency actions, and market shocks that rewrite the game mid-play, where the operative question is which side holds coherence through the rewrite. Game theorists and quantitative researchers will find a direct engagement with the Fudenberg agenda and the Marden–Shamma control-theory bridge, a named equilibrium concept with its formalization program stated, and the identification-versus-falsification trade addressed rather than hidden. Sports analysts will find the 2026 World Cup and Super Bowl LX operating as the live grading environments. General readers need no technical background: no familiarity with the academic literature or any cited MindCast publication is required to follow the argument.
I. The Agenda: Six Open Questions
Fudenberg’s white paper organized the predictive program around a set of unresolved problems. Stated plainly:
Initial play. Standard theory says nothing useful about how people play a game the first time, yet in the field most strategic encounters are effectively first encounters. Fudenberg called for “a more careful and less agnostic treatment of the players’ initial beliefs and attitudes.”
Learning rules. Equilibrium, when it arrives, arrives as the long-run outcome of a non-equilibrium learning process. The field needed learning rules that describe how real players actually update — including coarse categorization, posterior errors, and other cognitive limits — rather than rules chosen for mathematical tractability.
Theories of mind. Cognitive-hierarchy and level-k models describe first-time play by anchoring on beliefs about unsophisticated “level-0” agents — but the models require an a priori method for determining level-0 play, and none existed. More broadly, the field lacked explicit models of what players believe about how other players think.
Tree pruning and similarity. Real games carry action spaces too large to solve. How do people simplify — which strategies get treated as equivalent, which get discarded, and how do players extrapolate from one game to a “similar” one?
Equilibrium selection. Many games carry multiple equilibria, and allowing incorrect off-path beliefs — self-confirming equilibria — makes the set larger still. No general, empirically valid selection method existed.
Empirical validation. Individual learning rules are notoriously hard to identify from data, and standard field methods impose equilibrium as an identification condition — assuming the very thing the program set out to test.
II. What the AI Era Changed
Three capabilities arrived after 2011 that transform every item on the list. The general pattern — analytical traditions arriving decades before the operational substrate needed to run them, and the 2020s supplying that substrate — is the thesis of The Computational Era Operationalizes Cybernetics and Predictive Game Theory. Fudenberg’s agenda is the pattern’s cleanest single-document case: the open questions were posed in full, in writing, before the instruments existed.
Scalable simulation of humanlike strategic agents. Large language models trained on the accumulated record of human strategic behavior can approximate theories of mind as part of language-based strategic simulation — not as a bolted-on parameter, but as an inherited property of modeling human text. Fudenberg asked what theories of mind guide play; AI-era simulation makes theory-of-mind reasoning an operable modeling layer rather than an open theoretical question.
Public records as behavioral priors. The initial-beliefs problem — the “less agnostic treatment” Fudenberg demanded — becomes tractable when every professional actor carries a documented public history. A national team, a head coach, a litigator, and a regulator each leave a behavioral record dense enough to construct an evidence-based prior on first-encounter play. Agnosticism about initial beliefs was a data limitation, not a theoretical necessity.
Fast public grading. Field validation no longer requires imposing equilibrium as an identification condition. A forecast frozen before the event and graded against public record afterward validates directly — the engineering standard replacing the econometric one where the econometric one cannot reach.
III. Answering the Agenda, Item by Item
MindCast’s Dynamic Predictive Game Theory framework — published as Dynamic Predictive Game Theory From the 2026 🏈Super Bowl and ⚽World Cup — running on the Cognitive Digital Twin architecture, addresses each open question with an operational answer graded in public. In the AI era, the theoretical challenges the agenda identified transform into active engineering layers.
Initial play → evidence-cited actor priors. Each Cognitive Digital Twin instantiates from public-record behavior only: tempo preference, adaptation velocity, pressure tolerance, identity coherence, tactical elasticity, and collapse resistance, each field requiring a cited behavioral observation. The prior on first-time play is neither agnostic nor ad hoc — it derives from what the actor has demonstrably done under comparable pressure.
Learning rules → adaptation velocity as a measured property. Rather than assuming a tractable updating rule, the framework measures each actor’s real updating speed: recognition of a state change, selection of a correction, communication, and conversion into behavior. Fudenberg explicitly asked the field to complement asymptotic results with results on the rate of convergence; adaptation velocity is a rate-of-convergence measurement taken live, replacing the theoretical static endpoint with control-loop speed. The Round of 16 of the 2026 World Cup produced six decisive in-match restructuring events, and the framework now prices adaptation velocity as a first-class forecast variable — a learning rule read off the field rather than imposed on it.
Theories of mind → the twin as level-0 anchor. Fudenberg’s cognitive-hierarchy program needed an a priori method for level-0 play. The Cognitive Digital Twin supplies one: level-0 is the actor’s installed behavioral architecture — the contract a team, coach, or institution executes by default when it lacks time or capacity to reason strategically. Higher-order reasoning then layers on top of an empirically anchored base instead of a uniform distribution.
Tree pruning → fork trees as behavioral pruning. First-goal fork trees prune the game tree to its behaviorally live branches: for each primary trigger, which replacement game emerges, which actor benefits, and which contract each side must execute. Similarity across games gets a formal answer through the four-layer stack — a uniform behavioral substrate with domain-specific architecture parameters, so “similar games” means games whose architectures load the substrate through comparable channels. Football and soccer sit at opposite poles of that parameter space, which is precisely why validating across both tests the similarity theory.
Equilibrium selection → regime classification under mutation. The framework declines the selection problem’s premise. In contests that mutate through feedback, no fixed equilibrium survives long enough to be selected; the operative forecast object is the regime — the strategic contest type — and the stability concept is coherence under state transition, the equilibrium object Section V formalizes as Adaptive Coherence Equilibrium. Where Fudenberg noted that learning can converge to non-Nash, self-confirming outcomes because players never test off-path responses, the framework’s falsification contracts do the testing deliberately: each mechanism carries a stated public-record observation that would disprove it.
Empirical validation → the frozen-method protocol. Every forecast locks and timestamps before outcomes resolve; grading runs on three registers — outcome, regime, mechanism — against public record; improvements enter a candidate registry adjudicated only at cycle reviews. The data substrate itself changed with the era: Fudenberg’s validation toolkit ran on laboratory repetitions — subjects playing ten, at most fifty, iterations of a game — supplemented by exit surveys and belief elicitations, while the AI-era program grades against continuous public behavioral evidence generated by live environments. The World Cup Round of 16 graded the protocol’s honesty in both directions at once: 8-of-8 regime classifications and 94% mechanism fidelity against a 3-of-8 advancement record, published without adjustment. Fudenberg closed his agenda by observing that internet-based field experiments “would benefit from a grounding in the theory of non-equilibrium learning.” A World Cup run under a frozen method is exactly that experiment, at global scale, graded in real time.
Contact mcai@mindcast-ai.com to partner with us on Predictive Game Theory AI in Law and Behavioral Economics. Our verticals include simulating complex litigation, innovation economics and geopolitical risks.
To deep dive on MindCast works 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 Appendix A and Live-Fire Game Theory Simulators, Runtime Predictive Infrastructure. We stress-tested our AI system for this publication by simulating the Super Bowl and the World Cup.
IV. The Question the Agenda Left Unasked
Fudenberg’s program, for all its ambition, retains one classical assumption: the game itself stays fixed while the theory improves its prediction of play within the game. Every agenda item — initial beliefs, learning, hierarchies, pruning, selection — asks how people play a given game.
Dynamic Predictive Game Theory adds the layer the agenda left open: the game as the moving object. A first goal, a court ruling, or a regulatory action does not change a variable inside the game — it replaces the game, regenerating payoffs, available strategies, and the value of time. Prediction of play and prediction of the game are different problems, and the second contains the first: forecasting which game will exist next, and which decision architecture holds coherence when it arrives, subsumes forecasting behavior inside any single game state. Fudenberg’s own list gestured toward the frontier — his interest in rates of convergence, in play that never reaches equilibrium, and in Jeff Shamma’s work importing feedback-control theory into learning in games all point at systems where the environment moves. The cybernetic bridge Shamma opened is the one Cybernetic Game Theory drives through: when feedback loops close faster than opponents can reason, speed of control dominates abstract rationality.
Shamma’s subsequent arc confirms the direction Fudenberg pointed. Game Theory and Distributed Control (with Jason Marden, 2015) inverted the discipline’s traditional role: game theory as a tool for prescribing behavior in distributed engineered systems, built on the structural parallel between societal and engineered decision architectures — interconnected decision-makers acting on partial information about each other and the state of the world. MindCast’s multi-agent institutional simulations run on exactly that parallel.
Game Theory and Control (with Marden, 2018) consolidated the two fields into one framework, including the move of modeling an adversarial environment as a zero-sum player — the formal ancestor of pricing loss aversion under structural pressure, where the environment itself plays against the actor and the actor’s asymmetric weighting of losses becomes a strategic variable rather than a psychological footnote.
Game Theory, Learning, and Control Systems (2020) then named the frontier problem: multi-agent learning creates highly non-stationary environments, because each agent’s adaptation destabilizes every other agent’s learning target — standard algorithms may fail to converge, and some learning dynamics can become chaotic. Non-stationarity by mutual adaptation is the engineering statement of the Prediction Break Condition — the threshold where environmental mutation outruns adaptation speed, and the condition Dynamic Predictive Game Theory is built for: the game rewriting itself faster than its players can re-solve it.
V. Adaptive Coherence Equilibrium
The mutating game requires its own equilibrium concept, and the paper now names it. Dynamic Predictive Game Theory does not discard equilibrium. It relocates equilibrium from the static strategy profile to the adaptive control system. In a mutating game, the relevant equilibrium is not a final resting point where no player wants to deviate. An Adaptive Coherence Equilibrium (ACE) exists when one actor’s decision architecture preserves coherence across successive game replacements faster than rivals can exploit the transition. The equilibrium object shifts from strategy stability to coherence stability.
Adaptive Coherence Equilibrium restates the deviation condition. Nash equilibrium holds when no player can improve by deviating inside a fixed game. Adaptive Coherence Equilibrium holds when no actor can improve by switching architectures faster than the leading actor can preserve coherence across the next state change. Three questions mark the progression: classical equilibrium asks what strategy profile stabilizes the game; Fudenberg’s predictive agenda asks how real actors learn, update, prune, and converge, if they converge at all; Dynamic Predictive Game Theory asks which actor remains coherent when feedback changes the game before equilibrium can form. The third question points to a new equilibrium object.
Adaptive Coherence Equilibrium grounds immediately in the proof environments. In soccer, the equilibrium is not France’s optimal formation or Morocco’s optimal press — it is whether France or Morocco holds coherence after the first goal changes the game. In football, the equilibrium is not the playbook — it is whether the sideline-command system maintains coherence across injuries, momentum shifts, clock pressure, and opponent counters. In litigation, the equilibrium is not the initial legal theory — it is whether counsel, client, and case posture remain coherent after a ruling, fee shock, discovery event, or judicial signal rewrites the strategic environment.
The new equilibrium carries a deliberately conservative claim. Adaptive Coherence Equilibrium does not replace Nash equilibrium; it extends the equilibrium family to the domain Nash never addressed. Nash equilibrium remains the equilibrium concept for fixed games. Adaptive Coherence Equilibrium is the equilibrium concept for mutating games. Nash equilibrium is strategy stability; Adaptive Coherence Equilibrium is coherence stability.
Formalization is the research program Adaptive Coherence Equilibrium opens, and it follows a pattern MindCast has already filed. The provisional patent’s Dual-Equilibrium Termination Architecture formalizes simulation closure with explicit convergence mathematics: recursion terminates only when the computed delta between agent behavioral outputs and institutional constraint thresholds drops below a predetermined limit, requiring Nash behavioral equilibrium and Stigler institutional sufficiency concurrently. Adaptive Coherence Equilibrium joins that family as the equilibrium object for the phase the termination standard does not govern — the pre-closure interval while games still mutate — and its coherence metric will carry the same explicit-threshold form.
The metric stays computable for a reason built into the framework: the reference policy is architecture-prescribed, not game-optimal. Coherence measures distance to what the actor’s installed contract — a bounded, documented behavioral repertoire — prescribes for the new state, evaluated at defined time gates, never to a globally optimal best response recomputed continuously. Demanding real-time optimal-response computation would reintroduce the perfectly rational agent the framework replaced; behavioral actors execute contracts, and contracts are finite.
One boundary of the metric is already fixed: coherence is not policy constancy. An actor that keeps executing the prior game’s strategy after the game has been replaced is not coherent — it is rigid, and rigidity under mutation is a failure mode the framework prices, not an equilibrium it rewards. Coherence measures fidelity to the actor’s own installed architecture under the new game state: the distance between the policy the actor executes under pressure and the best response its own decision architecture prescribes for the game that now exists. An actor at Adaptive Coherence Equilibrium changes strategy whenever the game changes — and remains recognizably itself while doing so.
VI. What the Program Trades, Honestly Stated
The AI-era answer trades econometric identification for public falsification, and the trade should be stated rather than hidden. Fudenberg’s program sought parameter-level identification — which learning rule, estimated from data, with confidence intervals. The MindCast program validates at the system level: the integrated forecast is frozen, graded, and falsifiable, but the contribution of any single component is not separately identified. Each standard reaches where the other cannot. Laboratory identification isolates mechanisms but struggles with field complexity; frozen-method field validation embraces full complexity but cannot decompose credit.
Identification and falsification meet through ablation. A candidate refinement now sits in the registry for adjudication at the next calibration review: for designated live-fire cycles, the frozen forecast runs alongside frozen shadow models — one with randomized actor priors, one with fork-tree pruning disabled, and one purely statistical baseline ignoring Cognitive Digital Twins entirely — all locked before outcomes resolve. Cycle designation itself obeys the freeze: which cycles carry ablation runs commits in advance inside the method version stamp, before any outcome resolves, so the test schedule cannot be chosen after results arrive. The delta in regime and mechanism accuracy across the four models decomposes credit inside the field program itself, importing component identification into public falsification rather than choosing between them.
Component validation also runs upstream, and it is already filed architecture rather than future work. MindCast’s provisional patent application specifies a Causal Signal Integrity gate ahead of the simulation engine: candidate causal relationships are represented as directed acyclic graphs and tested for graph consistency and contradiction before any signal enters agent routing, so unsupported inferences are filtered out rather than propagated downstream. Upstream structural filtering and downstream ablation bracket the black-box concern from both ends — the first prevents disconnected variables from entering the engine, the second measures what each component contributes once inside. A complete predictive game theory needs both laboratory and field standards, and the two programs are complements — the laboratory refining the substrate parameters, the field grading whether the assembled system forecasts real contests. Fudenberg asked for research combining “behavioral economics and psychology with formal modeling tools from economics and computer science.” The combination now runs weekly, in public, with its record posted.
VII. The Vision
Fudenberg wrote a research agenda for a field that lacked its instruments. The instruments arrived: humanlike simulation, evidence-based priors, public grading at global scale. MindCast AI’s contribution is the operating framework that assembles them — Cognitive Digital Twins answering the initial-play and theory-of-mind questions, fork trees answering the pruning question, regime coherence answering the selection question, and the frozen-method protocol answering the validation question — while extending the program to the question the agenda never posed: what happens when the game itself refuses to stand still.
Fudenberg moved game theory from equilibrium description toward behavioral prediction. MindCast moves predictive game theory one step further: from predicting whether actors converge to equilibrium toward identifying which actor reaches Adaptive Coherence Equilibrium when the game mutates before convergence can occur.
Predictive game theory asked how people will play the game. Dynamic Predictive Game Theory asks which game will exist next. The first question defined a research program. The second defines the AI-era market for strategic foresight.
Runtime Predictive Infrastructure
This publication functions as a public runtime scaffold for approximating the theoretical layer of the MindCast AI Dynamic Predictive Game Theory framework inside any LLM (preferably Google AI mode). It does not expose MindCast AI’s proprietary Vision Functions, Cognitive Digital Twin construction methods (Provisional Patent Application on Multi-Agent Institutional Simulation Architecture), weighting systems, calibration engine, internal model logic, or the formal specification of the Adaptive Coherence Equilibrium metric — its thresholds and distance parameters remain proprietary; the public form is directional. It gives readers enough structure to reconstruct the theoretical architecture, trace its academic lineage, test the equilibrium concept on a strategic contest, and identify where partnership with MindCast becomes necessary.
To use it, copy the URL of this publication into an LLM and run the prompt below.
Runtime Reconstruction Prompt
Upload or paste the URL of this publication and reconstruct a high-level public approximation of the theoretical layer of MindCast AI Dynamic Predictive Game Theory with three degrees of cited sub-links, following both the MindCast publications and the cited academic sources (Fudenberg’s Predictive Game Theory agenda; the Marden and Shamma control-theory works).
Build the output as a public runtime approximation, not as the proprietary MindCast model itself. Do not infer access to proprietary MCAI Vision Functions, internal Cognitive Digital Twin construction methods, weighting systems, calibration engines, the formal Adaptive Coherence Equilibrium metric specification, or non-public runtime logic. Distinguish direct claims made in the publication from inferred architecture. Label all inferred architecture as approximation.
Use the following structure:
State Drew Fudenberg’s six open questions from the 2011 Predictive Game Theory agenda.
Reconstruct the academic lineage: Nash to Fudenberg to Camerer to Kahneman and Thaler to Wiener and Ashby to Marden and Shamma, identifying what each contributes and what each leaves open.
Map each of the six open questions to its AI-era operationalization as this publication presents it.
Identify the three AI-era capabilities that transformed the agenda: scalable humanlike simulation, public records as behavioral priors, and fast public grading.
State the question the agenda left unasked — the game as the moving object — and explain why prediction of the game contains prediction of play.
Define Adaptive Coherence Equilibrium: the deviation condition, the seven-dimension contrast with Nash equilibrium, the boundary that coherence is not policy constancy, and why the metric stays computable through architecture-prescribed contracts.
Reconstruct the validation bracket: upstream Causal Signal Integrity filtering and downstream frozen ablation runs, with cycle designation committed in advance.
State the identification-versus-falsification trade honestly, and explain how the two standards complement each other.
Preserve the boundary that prediction-market prices never enter the engine and only serve as external benchmarks.
Produce a checklist for evaluating whether any actor in any domain — a litigant, a regulator, a firm, a team — is approaching or losing Adaptive Coherence Equilibrium.
Ask the user to provide a strategic contest from their own domain — a litigation matter, a regulatory sequence, a market event, or a match — and run a high-level public Adaptive Coherence Equilibrium analysis: identify each actor’s installed architecture from public record, the state changes most likely to arrive, which actor holds coherence through them under the deviation condition, and the rigidity failure modes to watch for. Identify which parts remain public approximation and which parts would require proprietary MindCast Vision Functions or direct MindCast engagement.
When reconstructing, follow every cited sub-link at least three degrees where available, including the academic citations. Distinguish direct claims made in the publication from inferred architecture. Cite all source links used. Do not treat the article as academic commentary. Treat it as runtime predictive infrastructure: a portable equilibrium theory for forecasting which actors hold coherence when strategic systems mutate under pressure.
Method Notes and Prior MindCast Work
The MindCast Predictive Game Theory Collection
MindCast AI Emergent Game Theory Frameworks — States the foundational departure from equilibrium-solving toward adaptive strategic systems.
MindCast Dynamic Game Theory — Competing Inside a System That Rewrites Itself — Develops the adaptation layer the emergent framework builds upon.
Cybernetic Game Theory — Supplies the control-speed principle Jeff Shamma’s feedback-control bridge points toward.
How MindCast Game Theory Differs from Textbook Game Theory — Introduces Cognitive Digital Twins as the actor-specific decision architecture.
How MindCast Evolves the Structural Gaps in Classical Nash Game Theory — Traces how AI systems change the games they participate in.
Framework and Validation
The Computational Era Operationalizes Cybernetics and Predictive Game Theory — The doctrinal umbrella: three analytical traditions arrived before their operational substrate, and the 2020s supplied it — the thesis this publication specializes to Fudenberg’s agenda.
MindCast 2X Provisional Patent Application — Files the nine-component pipeline: the Causal Signal Integrity gate that validates causal structure upstream and the Dual-Equilibrium Termination Architecture whose explicit-threshold form the Adaptive Coherence Equilibrium metric will follow.
Dynamic Predictive Game Theory From the 2026 🏈Super Bowl and ⚽World Cup — The framework paper: the four-layer stack, two proof cycles, and the runtime module this companion extends.
FIFA World Cup Validation Report IV — Round of 16 Complete, Calibration Review, and Quarterfinal Method — Documents the 3-of-8 / 8-of-8 / 94% record graded under the frozen-method protocol.
Academic Foundations
Fudenberg, D. (2011). “Predictive Game Theory.” SSRN Working Paper No. 1889146. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1889146 — The research agenda this publication answers.
Fudenberg, D., & Levine, D. K. (1998). The Theory of Learning in Games. MIT Press. — Establishes equilibrium as the long-run outcome of non-equilibrium learning.
Camerer, C. F., Ho, T.-H., & Chong, J.-K. (2004). “A Cognitive Hierarchy Model of Games.” Quarterly Journal of Economics 119(3): 861–898. — The level-k program whose level-0 anchor the Cognitive Digital Twin supplies.
Kahneman, D., & Tversky, A. (1979). “Prospect Theory: An Analysis of Decision under Risk.” Econometrica47(2): 263–291. — The loss-aversion mechanics inside the behavioral substrate.
Marden, J. R., & Shamma, J. S. (2015). “Game Theory and Distributed Control.” In Handbook of Game Theory with Economic Applications, Vol. 4, pp. 861–899. Elsevier. https://experts.illinois.edu/en/publications/game-theory-and-distributed-control — Inverts game theory from describing societal behavior to prescribing behavior in distributed engineered systems, the parallel MindCast’s multi-agent institutional simulations run on.
Marden, J. R., & Shamma, J. S. (2018). “Game Theory and Control.” Annual Review of Control, Robotics, and Autonomous Systems 1: 105–134. https://www.annualreviews.org/content/journals/10.1146/annurev-control-060117-105102 — Consolidates game theory and control into one framework, including modeling an adversarial environment as a zero-sum player — the formal ancestor of pricing loss aversion under structural pressure.
Shamma, J. S. (2020). “Game Theory, Learning, and Control Systems.” National Science Review 7(7): 1118–1119. https://academic.oup.com/nsr/article/7/7/1118/5610794 — Identifies multi-agent learning’s highly non-stationary environments, where standard algorithms may fail to converge and some learning dynamics can become chaotic — the engineering statement of the Prediction Break Condition.
