The Contradiction Trap: A Dialectical and Game-Theoretic Framework for Exposing Structural Bias

Part I – The Contradiction Trap: A Dialectical and Game-Theoretic Framework for Exposing Structural Bias James D. Atkinson 2025 Abstract This paper introduces the contradiction trap: a dialectical and game-theoretic mechanism for detecting structural bias, motivated asymmetry, and narrative drift in institutional and algorithmic decision systems. Grounded in epistemic game theory, the trap recasts contradiction as a falsifiable evidential event: whenever a system’s stated rationale and its observable behaviour cannot be jointly sustained, the resulting inconsistency becomes a measurable signal of underlying deviation. The trap is formalised as a one-move, strictly competitive epistemic game in which every admissible response incurs coherence loss, generating informational payoffs that convert contradiction into diagnostic evidence. The framework provides a portable audit instrument for domains that claim impartiality but exhibit asymmetric behaviour — governance, organisational reasoning, and algorithmic architectures alike. By treating inconsistency not as a logical failure but as a data-bearing phenomenon, the contradiction trap establishes the epistemic foundations of the mathematics of integrity: a unified evidential paradigm in which legitimacy is demonstrated not through assertion, but through resistance to structured, adversarial challenge. Keywords: mathematics of integrity; epistemic game theory; dialectical inference; contradiction analysis; structural asymmetry; coherence loss; motivated deviation; epistemic diagnostics; adversarial reasoning; algorithmic accountability; institutional reasoning; reasoning integrity; philosophy of logic; evidential audit design. 1 Contents 1 Visual Abstract 6 2 Introduction 6 3 Related Works 8 3.1 Dialectical and Logical Foundations . . . . . . . . . . . . . . . . . . . 8 3.2 Epistemic Game Theory and Information Dynamics . . . . . . . . . . 9 3.3 Structural Bias, Auditing, and Algorithmic Accountability . . . . . . . 10 3.4 Contemporary Developments . . . . . . . . . . . . . . . . . . . . . . . 10 3.5 Synthesis................................... 11 4 Contribution and Novelty 11 4.1 From Proof to Performance. . . . . . . . . . . . . . . . . . . . . . . . . 11 4.2 Quantification of Contradiction. . . . . . . . . . . . . . . . . . . . . . . 11 4.3 Integration with Algorithmic Accountability. . . . . . . . . . . . . . . . 12 5 Definition and Core Structure 13 5.1 FormalDefinition .............................. 13 5.2 The Generic R-N-f(P) Framework . . . . . . . . . . . . . . . . . . . . . 14 5.3 TheCoreProperty.............................. 15 6 Origins and Distinction 15 7 Coherence Cost Estimation Methods: Selection and Validation 17 7.1 Comparative Framework . . . . . . . . . . . . . . . . . . . . . . . . . . 17 7.2 Scaling and Performance . . . . . . . . . . . . . . . . . . . . . . . . . . 19 7.3 Null Model and Significance Testing . . . . . . . . . . . . . . . . . . . 19 7.4 Method Selection Guidance . . . . . . . . . . . . . . . . . . . . . . . . 19 7.5 Multi-Method Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 19 8 Methodology for Contradiction Games 20 8.1 Purpose ................................... 20 8.2 InputsandArtefacts ............................ 20 8.3 Construction of the Trap . . . . . . . . . . . . . . . . . . . . . . . . . . 21 8.4 Measurement and Quantification . . . . . . . . . . . . . . . . . . . . . 21 8.4.1 CoherenceCost........................... 21 8.4.2 Accumulation Models . . . . . . . . . . . . . . . . . . . . . . . 22 8.5 Meta-Moves and Secondary Signals . . . . . . . . . . . . . . . . . . . 22 9 Structural Contradiction Analysis 23 9.1 Worked Examples at Full Structural Rigour . . . . . . . . . . . . . . . 23 2 9.1.1 Example 1: Organisational Restructure . . . . . . . . . . . . . 23 9.1.2 Example 2: Algorithmic Hiring Audit . . . . . . . . . . . . . . . 24 9.1.3 Example 3: Legal Consistency Test . . . . . . . . . . . . . . . . 25 9.2 Meta-Move Classification . . . . . . . . . . . . . . . . . . . . . . . . . 25 9.3 Field Deployment (Mini-Study) . . . . . . . . . . . . . . . . . . . . . . 26 9.4 Evasion Composite Index (ECI) . . . . . . . . . . . . . . . . . . . . . . 26 9.5 Robustness Under Perturbation . . . . . . . . . . . . . . . . . . . . . . 26 9.6 Synthesis: Contradiction as Diagnostic Evidence . . . . . . . . . . . . 26 10 Applications Across Domains 27 10.1 Legal and Regulatory Analysis . . . . . . . . . . . . . . . . . . . . . . . 27 10.2 AI Fairness and Algorithmic Auditing . . . . . . . . . . . . . . . . . . . 27 10.3 Organisational Governance and Decision Systems . . . . . . . . . . . 28 10.4 Philosophical and Epistemic Inquiry . . . . . . . . . . . . . . . . . . . 28 10.5 Media Systems (Neutral Case Study) . . . . . . . . . . . . . . . . . . . 28 11 Analytical Function 29 11.1 Overview................................... 29 11.2 ControlledFraming............................. 29 11.3 Response Inevitability . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 11.4 Diagnostic Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 11.5 DocumentaryValue............................. 31 11.6 From Logic to Measurement . . . . . . . . . . . . . . . . . . . . . . . . 31 12 The Core Principle: Asymmetry Without Necessity Shifts the Burden Toward Intent 32 12.1 Evidential Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . 33 12.2 Boundaries and Caveats . . . . . . . . . . . . . . . . . . . . . . . . . . 33 12.3 Multi-Agent and Recursive Cases . . . . . . . . . . . . . . . . . . . . . 33 13 Game-Theoretic Formalisation 34 13.1 FormalDefinition .............................. 34 13.2 Epistemic Constant-Sum . . . . . . . . . . . . . . . . . . . . . . . . . . 35 13.3 Payoffs and Information . . . . . . . . . . . . . . . . . . . . . . . . . . 35 13.4 Equilibrium Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 13.5 Information-Theoretic Interpretation . . . . . . . . . . . . . . . . . . 36 13.6 Comparative Game-Theoretic Structure . . . . . . . . . . . . . . . . . 37 13.7 StrategicDynamics............................. 37 13.8 BoundedCoherence ............................ 37 13.9 Interpretive Consequence . . . . . . . . . . . . . . . . . . . . . . . . . 38 13.10 EthicalGuardrails.............................. 38 13.11 ProhibitedUses............................... 38 3 14 Future Research Programme 38 14.1 Empirical Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 14.2 Theoretical Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 14.3 Methodological Development . . . . . . . . . . . . . . . . . . . . . . . 39 14.4 Applied Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 40 15 Summary of Contributions 40 16 Conclusion 41 A Quick Reference Card 42 Appendix A: Quick Reference Card 42 B Methodology for Contradiction Games 43 Appendix B: Methodology for Contradiction Games 43 B.1 Purpose ................................... 43 B.2 Pre-registration (recommended) . . . . . . . . . . . . . . . . . . . . . 43 B.3 InputsandArtefacts ............................ 43 B.4 Construction (Designing the Trap) . . . . . . . . . . . . . . . . . . . . . 43 B.5 Deployment ................................. 44 B.6 Measurement and Quantification . . . . . . . . . . . . . . . . . . . . . 44 B.6.1 CoherenceCost........................... 44 B.6.2 Information Gain . . . . . . . . . . . . . . . . . . . . . . . . . . 44 B.6.3 Meta-Moves............................. 44 B.7 Analysis and Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 B.8 Reporting Template (One Page) . . . . . . . . . . . . . . . . . . . . . . 45 B.9 Ethics and Safeguards . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 C Quick-Start Checklist (Practitioner Version) 46 Appendix C: Quick-Start Checklist (Practitioner Version) 46 D Glossary of Specialist Terms 46 Appendix D: Glossary of Specialist Terms 46 E Worked Example (Generic) 47 Appendix E: Worked Example (Generic) 47 E.1 ContextandSetup ............................. 47 E.2 Branch Outcomes and Coherence Costs . . . . . . . . . . . . . . . . . 47 4 E.3 Sensitivity and Robustness . . . . . . . . . . . . . . . . . . . . . . . . 48 E.4 Summary................................... 48 F Estimator Pseudocode (RB–C, GI–C, SD–C) 48 Appendix G: Estimator Pseudocode 48 F.1 RB–C: Rule-Based Coherence . . . . . . . . . . . . . . . . . . . . . . . 48 F.2 GI–C: Graph-Informed Coherence . . . . . . . . . . . . . . . . . . . . 49 F.3 SD–C: Semantic-Distance Coherence . . . . . . . . . . . . . . . . . . 49 F.4 Aggregation and Normalisation . . . . . . . . . . . . . . . . . . . . . . 50 F.5 SanityChecks ................................ 50 5 1 Visual Abstract Commitments 𝑅,𝑁 Framed Proposition 𝑓(𝑃) Responses 𝐺/𝐷 Coherence Cost 𝐶(𝑐𝑟) Information Gain Δ𝐼 Figure 1: Visual abstract for The Contradiction Trap: symmetry-based framing converts contradiction into measurable coherence cost and information gain. 2 Introduction Contradiction has long served as the philosopher’s stress test of truth. From the Socratic elenchus to Aristotle’s law of non-contradiction and the modern reductio ad absurdum (Socratic method: (Benson, 2021); reductio: (Groarke, 2023)), contradiction has functioned as philosophy’s diagnostic heartbeat: when a position collapses in on it6 self, it radiates its own refutation. Yet, while classical logic isolates contradiction within propositions, organisational and institutional reasoning often conceals contradiction within systems. This paper formalises a method for locating those contradictions, not in abstract syntax, but in an applied reasoning framework — a procedural trap that converts rhetoric into data (R. J. Aumann & Brandenburger, 1995; Floridi, 2011; Skyrms, 2010). The contradiction trap extends logic from proof to performance. By structuring a question where every available answer contradicts a distinct part of the respondent’s declared logic, it transforms a qualitative dispute into an epistemic experiment. In doing so, it reframes contradiction as evidence: a measurable, reproducible signal of bias or concealed motive. This situates logical analysis within practical domains such as governance, ethics, and algorithmic accountability (Ananny & Crawford, 2018; Barocas et al., 2019; Binns, 2018), providing a portable tool for interrogating systems that claim neutrality yet behave with asymmetry. Glossary note. Key technical terms used throughout this paper include: coherence cost (the measurable strain within a reasoning system when commitments conflict), epistemic game (a structured interaction in which agents’ beliefs depend on one another’s reasoning consistency), and meta-evasion score (a behavioural index capturing secondary avoidance tactics such as reframing or delay). A complete glossary is provided in Appendix D. While the contradiction trap introduces a novel formal and epistemic framework, it emerges from a broader lineage of research in dialectical reasoning, epistemic game theory, and algorithmic accountability. The following section situates this work within that interdisciplinary context, tracing how earlier theories of dialectical games, belief revision, and audit-based inference inform the trap’s design. By connecting these traditions, we clarify both the conceptual ancestry and the practical novelty of treating contradiction as a measurable epistemic signal rather than a logical failure. Lemma 2.1 (Inevitability of Inconsistency).Let 𝒞∶{𝑃1,𝑃2}→𝑂 (2.1) be a clarity configuration with stated rationale 𝑅and narrative justification 𝑁. Suppose that for each pathway 𝑃𝑖: 1. 𝑃𝑖preserves 𝑁only by contradicting 𝑅, and 2. 𝑃𝑖preserves 𝑅only by contradicting 𝑁. Then every admissible 𝑃𝑖produces an outcome 𝒞(𝑃𝑖)that is inconsistent with either 𝑅or 𝑁. Inconsistency is therefore unavoidable and independent of agent intent. 7 Proof. For each 𝑃𝑖, either (1) 𝑁is preserved at the cost of contradicting 𝑅, or (2) 𝑅is preserved at the cost of contradicting 𝑁. Because {𝑃1,𝑃2}exhausts the admissible response set, no pathway simultaneously preserves both commitments. Thus every 𝑃𝑖 yields an outcome inconsistent with at least one of 𝑅or 𝑁. The value of Constructio ad Claritatem lies not in the contradiction itself but in the structural information revealed by it. By collapsing the agent’s discretion into a minimal two-pathway decision-space, the configuration separates the professed rationale of the system from its operative motive. The mathematical analysis that follows treats clarity configurations as diagnostic objects: small, adversarial structures that reveal inconsistency without requiring confrontation, intent analysis, or subjective interpretation. Roadmap. This paper forms the opening part of a larger programme. A second paper develops a stochastic, self-correcting analogue of fairness, and a third generalises these principles into a falsifiable architecture for institutional integrity. Together, they outline a unified evidential approach to claims of neutrality across reasoning, allocation, and governance. 3 Related Works The contradiction trap arises at the intersection of three research traditions: dialectical reasoning and logical games,epistemic game theory, and bias auditing and epistemic accountability. Each contributes a structural insight that the trap consolidates into a single diagnostic framework. 3.1 Dialectical and Logical Foundations The intellectual roots of the contradiction trap lie in the long history of dialectical reasoning, from the Socratic elenchus to formal dialogue games in modern logic. Woods and Walton’s analysis of Question-begging and Cumulativeness in Dialectical Games showed that dialogical exchanges can be modelled as competitive epistemic structures in which contradiction exposes circularity or bias rather than mere error (Woods & Walton, 1982). In their formulation, epistemic defeat — not persuasion — marks logical victory. The contradiction trap extends this lineage by converting contradiction into measurable information gain rather than rhetorical failure. Floridi’s conception of trans8 parency as epistemic accountability (Floridi, 2019) reinforces this shift: contradiction becomes a mechanism of verification through exposure, not a flaw. Structural-dialectical psychology reaches the same conclusion from a different angle. Veraksa et al. describe contradiction as a dynamic interplay of oppositions that mutually exclude and presuppose one another (A. Veraksa et al., 2013). This aligns with the trap’s treatment of inconsistency not as breakdown but as evidence of structural tension within reasoning systems. Where classical logic isolates contradiction within propositions, the trap operates across systems of commitments, transforming qualitative opposition into quantitative signal. Computational models of argumentation provide additional grounding. Wells’s work on Cumulativeness in Dialectical Games identifies formal mechanisms for aggregating reasoning moves and classifying game types by epistemic persistence (Wells, 2013). The contradiction trap is a specific subclass of such games — strictly competitive and lossdeterministic — where each move by the respondent necessarily incurs coherence cost. It therefore fits squarely within what Wells calls “cumulative dialectical architectures,” where epistemic state transitions accumulate evidence of internal inconsistency. 3.2 Epistemic Game Theory and Information Dynamics Game-theoretic approaches to epistemic reasoning investigate how information, belief, and contradiction evolve during interaction. Benthem’s Games in Dynamic Epistemic Logic formalised the dynamics of information change in imperfect-information games by linking logical operations to epistemic updates through modal structure (Benthem, 2001). In the contradiction trap, information gain (Δ𝐼) and coherence cost (𝐶) mirror this structure: contradiction forces an epistemic update that reduces uncertainty about a system’s internal logic. Li and Wang’s From Rules to Runs deepens this insight by separating rule structures from actual play sequences in imperfect-information games, enabling stepwise tracking of epistemic change (K. Li & Wang, 2015). This corresponds directly to the trap’s distinction between stated rationales (𝑅) and narrative justifications (𝑁), whose forced reconciliation yields measurable contradiction. Dufwenberg and Lindén’s analysis of Inconsistencies in Extensive Games showed that epistemic contradictions arise even under minimal rationality assumptions (Dufwenberg & Lindén, 1996). Their call for belief-revision mechanisms is instantiated in the trap: contradiction becomes the empirical signal for updating beliefs about motive or bias. Weirich’s Epistemic Game Theory and Logic synthesises these ideas by framing games 9 Contemporary audit frameworks extend this lineage. Mökander’s ethics-based audits (Mökander, 2023) and Buhmann et al.’s institutional accountability mechanisms (Buhmann et al., 2020) treat contradiction as a condition of transparency. Yang et al. (K. Yang & Kudenko, 2023; Y. T. Yang et al., 2025) use Stackelberg-style epistemic games to formalise trade-offs between privacy and accountability — precisely the structure operationalised by coherence cost. Distinction from Classical Forms. Reductio ad absurdum reveals the falsity of a proposition by deriving contradiction from assuming it true. The contradiction trap differs by operating at the systemic level: it cross-tests an entire network of commitments by deploying a proposition whose affirmation and denial contradict different elements of the system’s rationale. Socratic elenchus cross-examines beliefs to induce aporia. The trap adopts the structure but not the moral purpose: its aim is evidential exposure, not intellectual humility. It transforms dialectical pressure into a measurable diagnostic of structural bias. Table 2: Distinction between classical logical methods and the Contradiction Trap. Aspect Reductio ad Absurdum Socratic Elenchus Contradiction Trap Primary Goal Prove proposition false Induce self-knowledge Expose bias or motive. Scope Single proposition Speaker’s beliefs System of commitments. Method Derive contradiction Questioning dialogue Forced binary choice exposing inconsistency. Adversariality Non-adversarial Non-adversarial Explicitly adversarial. Evidence Type Logical proof Qualitative insight Measurable contradiction. Measurability Binary (valid/invalid) Qualitative Quantitative (𝐶,Δ𝐼). Applications Mathematics, logic Philosophy, education Law, AI auditing, governance. 16 7 Coherence Cost Estimation Methods: Selection and Validation This section outlines practical methods for estimating the coherence cost 𝐶, allowing practitioners to choose between rule-based, graph-informed, and semantic estimators depending on data structure, interpretability constraints, and computational resources. Each method quantifies the internal strain a reasoning system exhibits when its commitments conflict. 7.1 Comparative Framework Three principal estimation methods are presented below, each offering distinct tradeoffs between interpretability, scalability, and computational complexity. Their comparative properties are summarised in Table 3. Table 3: Comparative properties of coherence-cost estimators (transposed view). Property RB-C (Rule-Based) GI-C (Graph-Informed) SD-C (Semantic Distance) Inputs Explicit rule sets or commitments Dependency or causal graph Embedding vectors or text corpora Complexity 𝑂(𝑛) 𝑂(𝑛log 𝑛) 𝑂(𝑛2) Interpretability High Medium Low Transparency High Medium Low Implementation Effort Low Moderate High Weighting Support Manual Automatic Implicit Typical Domain Legal / Policy Governance / Decision Systems NLP / Model Auditing Note: SD-C methods incur high upfront training cost but low inference cost once embeddings are established. 17 (1) Rule-Based Coherence (RB-C) RB-C operates on explicit commitments expressed in a rule language ℛ, identifying contradictions as minimal violation sets (D. M. Gabbay & Guenthner, 2003; Gärdenfors, 1988; Meyer & Wieringa, 1993; Prentzas & Hatzilygeroudis, 2012). The logical substrate may be propositional, deontic, or modal depending on the domain. 𝐶RB(𝑐𝑟)=min{|𝑆|∶𝑆⊆(𝑅∪𝑁),(𝑅∪𝑁)∖𝑆is consistent under ℐ}.(7.1) Complexity: 𝑂(𝑛). Highly interpretable and reproducible; ideal for regulatory, contractual, or policy corpora. (2) Graph-Informed Coherence (GI-C) GI-C models commitments as a directed graph 𝐺=(𝑉,𝐸) (7.2) 𝑉=𝑅∪𝑁 (7.3) and quantifies contradiction as the size of the minimal hitting set required to restore consistency (Dung, 1995; Hunter, 2008; Pearl, 2009b; Reiter, 1987): 𝐶GI(𝑐𝑟)=|HitSetmin(𝑟)| (7.4) Complexity: 𝑂(𝑛log 𝑛). GI-C aligns with conflict-set detection and model-based diagnosis in classical AI. (3) Semantic Distance Coherence (SD-C) SD-C estimates latent contradiction in unstructured or natural-language corpora. Each proposition 𝑞∈𝑄𝑟 is embedded as a vector 𝑣and compared to its coherence-preserving projection 𝑣′(J. Li et al., 2016; MacKay, 2003; Reimers & Gurevych, 2019): 𝐶SD(𝑐𝑟)=∑ 𝑞∈𝑄𝑟(1−cos(𝑣,𝑣′))(7.5) Complexity: 𝑂(𝑛2). KL divergence or Wasserstein distance may be used where embeddings encode stance or implication. Interpretation. SD-C is well-suited to NLP-based audits, latent contradiction detection, and model interpretability contexts. 18 7.2 Scaling and Performance Approximate computational scaling: RB-C: 𝑂(𝑛), GI-C: 𝑂(𝑛log 𝑛), SD-C: 𝑂(𝑛2)(7.6) RB-C and GI-C support live or iterative audits; SD-C is better suited to retrospective, high-fidelity analysis. 7.3 Null Model and Significance Testing To assess significance, randomise labels within symmetric inputs 𝑋(permutation test) to obtain bootstrap samples (P. I. Good, 2005): 𝒞0={𝐶(𝑏) 0}𝐵 𝑏=1 (7.7) Normalise using: 𝑧=𝐶(𝑐𝑟)−𝜇0 𝜎0, 𝐴=|𝐶𝐺−𝐶𝐷|(7.8) Flag incoherence when 𝐴≥𝜏𝐴and 𝑧≥𝜏𝑧. 7.4 Method Selection Guidance Estimator choice should follow data structure: •RB-C: explicit rules or commitments. •GI-C: interdependent or hierarchical reasoning. •SD-C: semantic drift, narrative contradiction, unstructured domains. Agreement between methods indicates coherence: |𝐶1−𝐶2|<𝜖⇒coherent (7.9) Divergence indicates epistemic instability. 7.5 Multi-Method Validation A multi-method validation pipeline is recommended: 1. Extract 𝑅and 𝑁. 19 2. Build 𝐺0and check edge semantics. 3. Verify embedding fidelity for SD-C. 4. Investigate unexplained variance across estimators. Heuristic Constructor. 1. Extract commitments. 2. Construct 𝐺0. 3. Identify symmetric locus 𝑋. 4. Synthesise 𝑓(𝑃). 5. Validate 𝐶(𝐺)>0under 𝐺and 𝐷. Summary. RB-C, GI-C, and SD-C form the quantitative backbone of contradiction games, converting qualitative disagreement into measurable epistemic strain. The next section formalises the protocol for executing a complete Contradiction Game. 8 Methodology for Contradiction Games 8.1 Purpose Contradiction games test whether a system’s stated rationale 𝑅and its narrative justifications 𝑁remain jointly coherent when subjected to a symmetric stressor. Every admissible response incurs positive coherence cost, enabling structural bias to be expressed as a measurable epistemic outcome (Baltag & Smets, 2008; Benthem, 2001; Pearl, 2009a). 8.2 Inputs and Artefacts •Rationale set 𝑅: written policies, rules, or formal commitments (Besnard & Hunter, 2008; D. Gabbay & Woods, 2003). •Narrative set 𝑁: justificatory explanations accompanying 𝑅(Prentzas & Hatzilygeroudis, 2012). •Symmetric locus 𝑋: cases where neutrality implies identical treatment (Binns, 2018; Hardt et al., 2016; Sen, 1969). •Framing operator 𝑓: constructs the test proposition 𝑃using epistemic or game-theoretic symmetry conditions (Fang et al., 2015; Prakken & Sartor, 2015; K. Yang & Kudenko, 2023). 20 8.3 Construction of the Trap 1. Map commitments. Extract 𝑅={𝑟𝑖}and 𝑁={𝑛𝑗}; build a dependency graph 𝐺0=(𝑅∪𝑁,𝐸0)following standard methods for structured commitments (Besnard & Hunter, 2008; Timmer et al., 2017). 2. Identify symmetric pressure. Choose 𝑋such that neutrality implies coherence under both outcomes (Hardt et al., 2016). 3. Define 𝑓(𝑃).Frame a binary proposition where: 𝐺∶preserves 𝑅and contradicts 𝑁, 𝐷∶preserves 𝑁and contradicts 𝑅, following adversarial dialogue-game design (Benthem, 2001; Prakken & Sartor, 2015). 4. Specify epistemic payoffs. Player utilities are defined by coherence cost, adopting the logic of strictly competitive epistemic games (Fang et al., 2015): 𝑈𝐴(𝑟)=𝐶(𝑐𝑟)(8.1) 𝑈𝐵(𝑟)=−𝐶(𝑐𝑟)(8.2) 𝐶(𝑐𝑟)>0 (8.3) No best response exists for 𝐵, yielding no equilibrium. 8.4 Measurement and Quantification 8.4.1 Coherence Cost We adopt four compatible estimators, grounded in logic, graph theory, semantics, and model checking. (1) Rule-Based (RB-C) 𝐶RB(𝑐𝑟)=min{|𝑆|∶𝑆⊆𝑅∪𝑁,(𝑅∪𝑁)∖𝑆is consistent }, (8.4) building on deontic and rule-based inference structures (Meyer & Wieringa, 1993; Prentzas & Hatzilygeroudis, 2012). (2) Graph-Informed (GI-C) 𝐶GI(𝑐𝑟)=|HitSetmin(𝑟)|, (8.5) HitSetmin(𝑟)=min{𝑆⊆𝑉∶𝐺+∖𝑆is acyclic and consistent}, (8.6) aligned with conflict-set detection in argumentation frameworks and causal reasoning (Dung, 1995; Pearl, 2009a). 21 (3) Semantic Distance (SD-C) 𝐶SD(𝑐𝑟)=∑ 𝑞∈𝑄𝑟(1−cos(𝑞,𝑞′)),(8.7) where 𝑞′is the projection of 𝑞onto the admissible closure 𝒦𝑟(J. Li et al., 2016; Reimers & Gurevych, 2019). (4) Model-Checking Penalty (MC-C) Minimal number of edits or constraint removals required to restore satisfiability in a formal model (Baier & Katoen, 2008; Clarke et al., 1999). Normalisation. When combining estimators: 𝐶∗(𝑐𝑟)=1 𝑚𝑚 ∑ 𝑗=1𝐶𝑗(𝑐𝑟)∈[0,1]. (8.8) 8.4.2 Accumulation Models Given a reasoning chain ℛ𝑟={𝑞1,…,𝑞𝑛}: 𝐶Σ(𝑐𝑟)= 𝑛 ∑ 𝑖=1𝜔𝑖𝛿(𝑞𝑖), (8.9) 𝐶max(𝑐𝑟)=max 𝑖𝛿(𝑞𝑖), (8.10) drawing on classical distributed vs. dominant-failure semantics in safety analysis (Leveson, 1995; Reason, 1990; Varshney & Alemzadeh, 2017). Interpretation. 𝐶Σcaptures distributed incoherence; 𝐶max isolates dominant failures. 8.5 Meta-Moves and Secondary Signals Evasive behaviour is modelled through a meta-evasion score 𝑀: 𝑀=∑𝑖𝑤𝑖𝑒𝑖 ∑𝑖𝑤𝑖,(8.11) drawing on strategic ambiguity and signalling frameworks (Crawford & Sobel, 1982; Schelling, 1960). Meta-moves include: •Reframing (legitimate if symmetry preserved), •Deferral (procedural delay), 22 •Ambiguity (content-free or strategic evasiveness) (Bratman, 1987; Crawford & Sobel, 1982). The joint inference model treats coherence cost and evasion as orthogonal: 𝑃(intent ∣𝐶,𝑀)∝(1−𝑒−𝐶)(1−𝑒−𝑀), (8.12) consistent with inverse-planning and intent-inference methods (Baker et al., 2009; Ng & Russell, 2000). 9 Structural Contradiction Analysis This chapter presents the applied analytical machinery of the R–N–𝑓(𝑃)framework. It unifies structural contradiction diagnostics, branch-level coherence scoring, meta-move classification, and temporal evasion analysis. The goal is evidential: to determine when a system’s own commitments cannot be jointly satisfied under symmetric pressure. Contradiction is not treated as failure but as data: a structural witness to inconsistency in stated rationale, narrative constraints, and observed or implied behaviour. This approach aligns with quantitative inconsistency measurement in argumentation frameworks (Grant & Hunter, 2011; Hunter & Konieczny, 2008) and inconsistency-tolerant reasoning in logic and AI (Besnard & Hunter, 2008; Wang & Li, 2011). The methods presented here parallel the formal structures used in evidential investigation in organisational analysis (Argyris & Schön, 1978; Brunsson, 1989) and algorithmic accountability (Buhmann et al., 2020; Mökander & Floridi, 2023), while remaining domain-agnostic and portable. The chapter is organised as follows. Section 9.1 provides rigorously specified structural examples. Section 9.2 outlines the taxonomy of meta-moves and their evidential coding. Section 9.4 introduces the Evasion Composite Index (ECI). Section 9.5 establishes perturbation robustness. Section 9.6 summarises the inferential architecture. 9.1 Worked Examples at Full Structural Rigour The following examples apply the structural method used in formal contradiction analysis: explicit declaration of rationale ℛ, narrative 𝒩, observed actions 𝐴, the symmetric test proposition 𝑓(𝑃), branch-level coherence loss, and minimal removal sets. This mirrors standard practice in inconsistency diagnosis and minimal hitting-set reasoning (Hunter & Konieczny, 2008; Reiter, 1987). 9.1.1 Example 1: Organisational Restructure Organisational contradictions of rationale versus narrative are well-studied in institutional analysis, particularly in the literature on organisational hypocrisy (Brunsson, 1989) and the distinction between espoused theory and theory-in-use (Argyris & Schön, 1978). 23 Rationale. ℛ={𝑟1∶parallel leadership, 𝑟2∶efficiency, 𝑟3∶titles reflect scope, 𝑟4∶no substantive change.} (9.1) Narrative. 𝒩={𝑛1∶not personal, 𝑛2∶no replacement, 𝑛3∶no diminution, 𝑛4∶consultation adequate.}(9.2) Observed actions. 𝐴={𝑎1∶new Head role, 𝑎2∶direct report reassigned, 𝑎3∶no consultation.}(9.3) Contradiction. Granting 𝑓(𝑃)contradicts 𝑎1,𝑎2. Denying 𝑓(𝑃)contradicts 𝑟3,𝑟4,𝑛1,𝑛2,𝑛3. This structure directly resembles the inconsistency patterns discussed in (Feldman & March, 1981) and minimal-inconsistency reasoning in (Hunter & Konieczny, 2008). 9.1.2 Example 2: Algorithmic Hiring Audit Algorithmic contradictions often arise from fairness impossibility constraints (Kleinberg et al., 2017) or narrative attempts to guarantee distributional outcomes without altering upstream assessment structures (Binns, 2018). Rationale. ℛ={𝑟1∶strict merit, 𝑟2∶identical assessment conditions.}(9.4) Narrative. 𝒩={𝑛1∶guaranteed diversity of outcome.}(9.5) Observed behaviour. 𝐴={𝑎1∶divergent outcomes despite identical profiles.}(9.6) Contradiction. Granting 𝑓(𝑃)contradicts 𝑎1. Denying 𝑓(𝑃)contradicts 𝑟1,𝑟2. This mirrors classical fairness-consistency conflicts documented in (Eubanks, 2018; Hardt et al., 2016). 24 9.1.3 Example 3: Legal Consistency Test Legal contradiction analysis draws on jurisprudential consistency requirements (Raz, 1979; Sartor, 2005) and the principle that equivalent cases should yield equivalent outcomes (Sunstein, 1999). Rationale. ℛ={𝑟1∶operational requirement, 𝑟2∶equivalent roles treated consistently.}(9.7) Narrative. 𝒩={𝑛1∶case-by-case discretion.}(9.8) Observed actions. 𝐴={𝑎1∶X refused, 𝑎2∶Y granted, 𝑎3∶roles equivalent.}(9.9) Contradiction. Granting 𝑓(𝑃)contradicts 𝑎1,𝑎2. Denying 𝑓(𝑃)contradicts 𝑟2,𝑛1. This aligns with formal inconsistency modelling in legal argumentation (Prakken, 2010). 9.2 Meta-Move Classification Meta-moves characterise responses to 𝑓(𝑃)not by content but by their effect on the inferential structure. The taxonomy echoes strategic evasion behaviours identified in dialogue games (Prakken, 2010; Walton, 1998), information manipulation (Crawford & Sobel, 1982), and strategic conflict (Schelling, 1960). Table 4: Meta-move taxonomy and evidential coding Meta-Move Definition Code Premise Attack (L) Identifies asymmetry in 𝑓(𝑃)and proposes a symmetric 𝑓′(𝑃)preserving ℛ,𝒩. Framing Objection (L) Continuous constraint misrepresented as binary; supplies corrected framing. Delay / Deflection (E) Temporal or procedural evasion (Power, 1997). Premise Denial (E) Rejects commitments without justification (Bratman, 1987). Counter-Trap Attempt (A) Attempts to shift burden or domain, echoing argumentative misdirection (Walton, 1998). 25 12 The Core Principle: Asymmetry Without Necessity Shifts the Burden Toward Intent The contradiction trap rests on a simple evidential claim: when a system deviates from its declared principles without necessity, that deviation increases the posterior odds of selective intent. The core principle formalises this transition from structural inconsistency to probabilistic inference (Bovens & Hartmann, 2003; Fredman, 2011b; I. Good, 1985; Jaynes, 2003). Statement. Let 𝐴=|𝐶𝐺−𝐶𝐷|denote the asymmetry in coherence cost under symmetric inputs 𝑋. If 𝐴>0and no external necessity ℰ(legal constraint, resource limit, stochastic uncertainty) accounts for it, then 𝐴raises the posterior odds of intent (Barak, 2012a; Craig, 2012a; Kass & Raftery, 1995): Pr(𝐼∣𝐴) Pr(¬𝐼∣𝐴) ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ posterior odds =Pr(𝐴∣𝐼) Pr(𝐴∣¬𝐼) ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ Λ(𝐴) ⋅Pr(𝐼) Pr(¬𝐼) ⏟⏟⏟⏟⏟⏟⏟⏟⏟ prior odds (12.1) Formal Bayesian framing. Let 𝐻0denote neutrality and 𝐻1motivated bias. Posterior elevation occurs precisely when (Bovens & Hartmann, 2003; I. Good, 1985) 𝑃(𝐻1∣𝐴)>𝑃(𝐻1)⟺Λ(𝐴)=𝑃(𝐴∣𝐻1) 𝑃(𝐴∣𝐻0)>1 (12.2) Necessity test. External necessities form a set ℰ. We first test the null hypothesis 𝐻0∶𝐴∈ℰ (12.3) Rejection of 𝐻0licenses evidential inference: the asymmetry is not required by external constraints and must therefore be explained by internal choice (Barak, 2012a; Fredman, 2011b). Decision rule (Bayes factor). Define the Bayes factor Λ(𝐴)= Pr(𝐴∣𝐼)/Pr(𝐴∣¬𝐼)under the registered null model. A shift in burden occurs whenever Λ(𝐴)≥𝜏, 𝜏>1 (12.4) e.g. 𝜏=3for “moderate” and 𝜏=10for “strong” evidential weight (Kass & Raftery, 1995). The rule is deliberately minimal: it does not diagnose intent, but obliges the system to supply a justification consistent with its own commitments. Remark. The heuristic “asymmetry without necessity implies intent’’ abbreviates the probabilistic claim: if 𝐴 ∉ ℰand 𝐴 > 0, then Λ(𝐴) > 1. The odds shift, but inference remains probabilistic, not deductive. 32 12.1 Evidential Interpretation The magnitude of 𝐴yields a graded evidential interpretation: •Small asymmetry (𝐴≈0): Structural inconsistency; motive cannot be inferred. Contradiction arises from system design rather than agency (Simon, 1955). •Moderate asymmetry (𝐴>0but bounded): Indicates implicit preference or unacknowledged contextual weighting. Suggests weakly motivated divergence (Brunsson, 2003; Tetlock, 2006). •Large asymmetry (𝐴≫0): Signals deliberate prioritisation or concealed motive. The system reveals its values more clearly through inconsistency than through claim. This evidential gradient distinguishes cognitive limits, structural design, and strategic manipulation. Whereas paraconsistent logics permit contradictory propositions to coexist without collapse (Priest, 2006), the contradiction trap uses contradiction to test epistemic integrity: the aim is not to survive inconsistency but to diagnose its origin. 12.2 Boundaries and Caveats The core principle applies within explicit epistemic limits: 1. Bounded Rationality. Asymmetry may reflect limited information or cognitive load (Kahneman, 2011); not all divergence is intentional. 2. Incomplete Mapping. If 𝑅or 𝑁are partially captured, observed asymmetry may arise from unmodelled commitments rather than bias (J. March & Olsen, 1984). 3. Meta-Game Costs. Anticipating interrogation may lead agents to distort commitments pre-emptively; the resulting asymmetry mixes bias with strategic evasion. 4. Multi-Agent Aggregation. Collective decisions aggregate divergent motives; asymmetry may reflect composition effects, not a unified intent (D. Dennett, 1987; List & Pettit, 2011). These caveats restrict scope without diminishing force. Properly applied, the principle distinguishes honest inconsistency from motivated contradiction and converts qualitative bias into quantitative inference. 12.3 Multi-Agent and Recursive Cases When responses are delegated or recursively mirrored, coherence analysis decomposes by agent. Each actor inherits rationale 𝑅𝑖and narrative 𝑁𝑖; the aggregate contradiction is 𝐶agg =∑ 𝑖𝑤𝑖𝐶𝑖, 𝑤𝑖≥0,∑ 𝑖𝑤𝑖=1 (12.5) Delegation diffuses, but does not eliminate, accountability: contradiction propagates through weighted commitments (List & Pettit, 2011). 33 Recursive belief formulation. Let 𝐵𝑖(𝐵𝑗(𝜑))denote agent 𝑖’s belief about agent 𝑗’s belief in 𝜑. Contradiction arises when (R. J. Aumann, 1976; Fagin et al., 1995; Hintikka, 1962) 𝐵𝑖(𝐵𝑗(𝜑))∧¬𝐵𝑗(𝜑) (12.6) under public declaration. Multi-layer conflicts produce recursive contradiction cascades, revealing unstable epistemic networks. Summary. The asymmetry principle supplies the probabilistic backbone of the contradiction game. Section 13 formalises this evidential rule within a game-theoretic framework, showing how posterior shifts map onto strategic loss functions. 13 Game-Theoretic Formalisation The contradiction trap can be cast as a one-move, strictly competitive epistemic game in which all available responses for the responder are losing strategies: each produces a negative payoff via positive coherence cost (R. Aumann, 1999; Brandenburger, 2007). Section 6 treated contradiction traps as applied dialectical instruments; here we formalise them within the vocabulary of game theory, showing that contradiction behaves as a forced-loss strategy inside a closed reasoning environment. Viewed through the lens of machine behaviour (Rahwan et al., 2019), contradiction games constitute behavioural falsification: agents disclose their internal priorities not by admission, but by necessity. 13.1 Formal Definition Epistemic game theory models beliefs about beliefs (Brandenburger, 2007; Fagin et al., 1995). The contradiction trap defines a new subclass in which reasoning itself constitutes play and contradiction constitutes outcome. Definition 13.1 (Contradiction Game).A Contradiction Game is a two-player epistemic game 𝐺=⟨𝑃,𝑆,𝑈,𝐶⟩ (13.1) 𝑃={𝐴,𝐵}, 𝑆𝐵={𝐺,𝐷} (13.2) with the following structure: 1. Player 𝐴(interrogator) applies a framing operator 𝑓to proposition 𝑃, selecting a scenario in which 𝐵’s commitments render {𝐺,𝐷}mutually exclusive with respect to its declared rationale and narrative. 2. Player 𝐵(responder) selects 𝑟∈{𝐺,𝐷}. 3. Each response induces coherence cost 𝐶(𝑐𝑟)>0, i.e. each response contradicts some part of 𝐵’s commitments. 34 4. Payoffs are epistemic: 𝑈𝐴(𝑟)=𝐶(𝑐𝑟), 𝑈𝐵(𝑟)=−𝐶(𝑐𝑟)(13.3) The defining feature is ∀𝑟∈𝑆𝐵∶ 𝐶(𝑐𝑟)>0 (13.4) so 𝐵has no contradiction-free option. This reverses the standard Aumann–Brandenburger paradigm in which shared belief conditions sustain equilibrium (R. Aumann, 1999): here, no epistemic state can restore equilibrium. 13.2 Epistemic Constant-Sum Material zero-sum games treat utility as consumption; contradiction games treat utility as information gain versus coherence loss. This parallels information-incentive models in signalling and behavioural audit games (Kreps & Wilson, 1982; Myerson, 1991). Definition 13.2 (Epistemic Constant-Sum).A Contradiction Game is epistemic constant-sum if there exist positive scaling constants 𝑎>0,𝑏∈ℝsuch that 𝑈𝐵(𝑟)=𝑎−𝑈𝐴(𝑟)+𝑏 (13.5) This expresses epistemic complementarity: the interrogator’s evidential utility equals the responder’s coherence loss up to affine transformation. Proposition 13.3 (Non-Existence of Nash Equilibrium).If 𝐶(𝑐𝑟)>0for all 𝑟 ∈{𝐺,𝐷}, then the game 𝐺 admits no pure Nash equilibrium. Proof. Suppose (𝑓∗,𝑃∗;𝑟∗)is a Nash equilibrium. By definition, 𝑟∗∈{𝐺,𝐷}. But for all 𝑟,𝑈𝐵(𝑟)=−𝐶(𝑐𝑟)< 0, so no 𝑟∗maximises 𝑈𝐵. Thus 𝐵has no best response, and mutual best-response fails (Osborne & Rubinstein, 1994). Therefore no pure equilibrium exists. 13.3 Payoffs and Information Let 𝐼(𝑟)denote evidential information content (information gain interpretation following Shannon and Jaynes (Jaynes, 2003; Shannon, 1948)): 𝐼(𝑟)=log𝑃(𝐷∣𝑟) 𝑃(𝐷) (13.6) A generalised epistemic payoff is 𝑈𝑖(𝑟)=𝛼𝐼(𝑟)−𝛽𝐶(𝑐𝑟), 𝛼,𝛽>0 (13.7) balancing information gain against contradiction cost. This casts contradiction traps as signal-to-cost games, consistent with behavioural inference models (Spence, 1973). 35 Figure 3: Coherence-cost divergence for grant (𝐶𝐺) and deny (𝐶𝐷). Absence of intersection indicates the impossibility of equilibrium. Deterministic loss. Since 𝐶(𝑐𝐺),𝐶(𝑐𝐷)>0, max 𝑟𝑈𝐵(𝑟)<0, min 𝑟𝑈𝐴(𝑟)>0 (13.8) Thus the responder faces a dominant-loss structure; mixing cannot remove loss, only obscure it. 13.4 Equilibrium Analysis Classical equilibrium requires mutual best response; contradiction games preclude this by construction. No choice of 𝑟∗preserves coherence, so stability cannot be restored without abandoning prior commitments. Meta-strategies (delay, reframing, premise-attack) therefore become secondary signals of motive and feed into the meta-evasion score 𝑀(Section 9.2), consistent with behavioural audit theory (“cheap talk under pressure”) (Crawford, 1991). 13.5 Information-Theoretic Interpretation Each contradiction produces information gain Δ𝐼=−log2𝑝(13.9) where 𝑝is the prior coherence probability. As 𝑝→0,Δ𝐼diverges (Jaynes, 2003; Shannon, 1948): contradiction reveals motive as a limiting case. 36 Utility view. 𝑈(𝑟)=𝐼(𝑟)−𝐶(𝑐𝑟)(13.10) captures the trade-off: systems lose epistemic integrity as contradiction deepens but thereby provide increasing evidential value. 13.6 Comparative Game-Theoretic Structure •Prisoner’s Dilemma: Cooperation restores equilibrium; here, no cooperation restores coherence. •Chicken Game: Bluff may avert collision; in contradiction games, collision is guaranteed. •Matching Pennies: Binary and stochastic; contradiction games are binary and deterministic (Osborne & Rubinstein, 1994). •Signalling Games: Hidden types inferred through messages; here, types are inferred through logical failure (Spence, 1973). This motivates a new subclass: epistemic, deterministic, contradiction-revealing games — logic as play, contradiction as payoff. 13.7 Strategic Dynamics Because all moves yield loss, rational responders adopt damage-limiting meta-moves—strategies consistent with bounded rationality and cognitive economisation (Kahneman, 2011; Simon, 1955): 1. Reframing — disown prior commitments. 2. Premise Challenge — attack 𝑓(𝑃)itself. 3. Delay — avoid instantiating the trap. 13.8 Bounded Coherence Agents tolerate small inconsistencies (Simon, 1955). A smooth cost model: 𝐶(𝑐𝑟)={0, 𝛿(𝑐𝑟)<𝜖, 𝑘(𝛿(𝑐𝑟)−𝜖)2,otherwise (13.11) Bounded coherence shifts magnitude but not existence of contradiction. Lemma 13.4 (Robust Non-Equilibrium).Let Γbe a Contradiction Game with 𝐶(𝑐𝑟)>0for all 𝑟. Replacing ideal rationality with bounded coherence leaves equilibrium impossible for any 𝜖<max𝑟𝛿(𝑐𝑟). 37 Distributed commitments. Institutional decision-making aggregates memory and commitments via distributed agents (List & Pettit, 2011; J. March & Olsen, 1984). Let 𝐾𝑡be institutional memory (partial recall models in AI audit contexts follow similar formulations (Mitchell et al., 2021)): 𝐾𝑡+1=𝑈(𝐾𝑡,event𝑡)(13.12) 13.9 Interpretive Consequence The contradiction trap treats reasoning integrity as a strategic resource. Systems that remain coherent under symmetric challenge demonstrate neutrality; systems that cannot reveal motive through epistemic loss. This mirrors the behavioural audit perspective in AI governance (Binns, 2018; Rahwan et al., 2019). Asymmetry becomes empirical evidence of selective reasoning. 13.10 Ethical Guardrails Ethical constraints follow normative principles from AI accountability, behavioural auditing, and governance integrity (Barocas & Selbst, 2017; Selbst et al., 2019). 13.11 Prohibited Uses These prohibitions align with literature on power asymmetry, coercion, and legitimacy in human–algorithm interactions (Barak, 2012a; Craig, 2012a). 14 Future Research Programme The contradiction trap establishes a unified analytical grammar for detecting structural inconsistency, but its full potential depends on sustained empirical, theoretical, methodological, and applied development. This section outlines a forward research agenda for consolidating contradiction analysis into a mature diagnostic discipline. 14.1 Empirical Validation Empirical research will determine how contradiction manifests in real-world systems and how reliably coherence-cost estimators track underlying inconsistency. Key directions include: •Institutional audits: large-scale evaluation of public or organisational decisions to map empirical distributions of coherence cost 𝐶and asymmetry 𝐴. 38 •Controlled experiments: application of contradiction traps to human and algorithmic agents to assess predictive validity, behavioural response, and adaptation over repeated play. •Longitudinal correction studies: testing whether contradiction exposure produces behavioural, procedural, or organisational reform over time. •Empirical priors: incorporating observed contradiction frequencies into Bayesian estimators of 𝑝, the probability that a system remains coherent under symmetric inputs. Empirical scope. A preliminary pilot is in preparation, examining convergence across RB-C, GI-C, and SD-C on a small organisational dataset. The aim is to establish inter-estimator correlation and baseline variance in coherence-cost measurement. Due to ongoing legal proceedings, empirical identifiers and case details are withheld until resolution; the methodology and theoretical model are unaffected. 14.2 Theoretical Extensions Several conceptual expansions warrant formal treatment: •Bounded coherence: modelling systems that tolerate limited contradiction as stable equilibria, with phase transitions when thresholds are exceeded. •Multi-agent propagation: analysing how contradiction diffuses across networks of agents with partially overlapping rationales 𝑅𝑖and narratives 𝑁𝑖. •Temporal commitment dynamics: formalising how coherence cost evolves as commitments are updated, forgotten, or strategically revised. •Meta-game contradiction: studying intentional use of self-contradiction as a signalling, bluffing, or narrative-control device in adversarial contexts. These directions move the contradiction trap toward a general theory of epistemic instability and reasoning decay. 14.3 Methodological Development Wider adoption requires robust and automated tooling. Priority areas include: •Commitment extraction: automated identification of 𝑅and 𝑁from legal, organisational, or modelgovernance documents. •Prior estimation: machine-learning models estimating domain-specific coherence priors 𝑝from historical datasets. •Formal verification integration: embedding contradiction tests within model-checking pipelines to identify inconsistent logic in software or AI systems. 39 •Sector-specific estimators: tailoring RB-C, GI-C, and SD-C to domain constraints (e.g. regulatory logic, risk frameworks, or NLP-heavy environments). These developments would render contradiction analysis repeatable, modular, and auditable. 14.4 Applied Implementation The applied frontier concerns embedding contradiction analysis into governance structures and operational practice: •Organisational audits: adopting contradiction traps as part of standard ethics, fairness, and compliance assessments. •AI-governance integration: aligning contradiction diagnostics with algorithmic accountability regulation, risk frameworks, and standards for explainability. •Practitioner training: developing coherent curricula for interpreting 𝐶,𝐴,Δ𝐼, and meta-evasion signals 𝑀. •Policy feedback loops: designing governance systems that automatically flag “asymmetry without necessity’’ as a structural risk indicator. Together, these strands outline a coordinated research programme: empirical grounding, theoretical expansion, methodological tooling, and institutional deployment. If developed in parallel, they would transform the contradiction trap from analytical prototype into a general-purpose instrument for epistemic accountability. 15 Summary of Contributions This paper provides both a theoretical foundation and an applied methodology for diagnosing bias within reasoning systems. Its core contributions are: 1. Formalisation of the contradiction trap.A reproducible dialectical mechanism that transforms contradiction from a logical defect into an evidential signal, bridging classical reasoning with applied audit practice (Benthem, 2001; N. Veraksa et al., 2013; Woods & Walton, 1982). 2. Definition of a new epistemic game class. The paper introduces Contradiction Games: strictly competitive, one-move epistemic structures in which every admissible strategy for the responder incurs positive coherence cost. This establishes a formal model of reasoning integrity as a strategic and measurable variable (R. Aumann, 1999; Brandenburger, 2007). 3. Information-theoretic grounding. Contradiction is modelled as information gain, with coherence cost 𝐶(𝑐𝑟)serving as a quantifiable proxy for epistemic tension and motive strength. This links logical inconsistency directly to entropy reduction (Shannon, 1948). 40 4. Cross-domain applicability. The framework is demonstrated across legal, organisational, and AIgovernance settings, showing how claims of neutrality can be stress-tested through symmetric inputs and scored via coherence metrics (Buhmann et al., 2020; Mökander, 2023; Rahwan et al., 2019; M. Yang et al., 2023). 5. Theoretical advancement. The paper derives the central evidential principle — asymmetry without necessity implies intent — as a general rule for detecting motivated reasoning and concealed bias under symmetric conditions. 6. Research roadmap. A programme for empirical, theoretical, and methodological development is outlined, including estimator validation, bounded-coherence models, and integration with algorithmicaudit pipelines. Collectively, these contributions reposition contradiction as a measurable epistemic phenomenon. They recast logic as an investigative science in which inconsistency becomes data rather than failure. 16 Conclusion The contradiction trap reframes contradiction as diagnostic evidence. By combining dialectical structure with epistemic game theory, it establishes a universal grammar for detecting bias in human, organisational, and algorithmic reasoning systems (Benthem, 2001; Rahwan et al., 2019; Woods & Walton, 1982). Its defining feature — a strictly competitive, non-equilibrial game in which every response induces self-contradiction — transforms qualitative disputes into quantitative artefacts (R. Aumann, 1999; Brandenburger, 2007). Operationally, the trap converts asymmetry into information: each contradictory choice produces measurable coherence cost, shifts posterior odds, and narrows plausible motive explanations (Dawid & Musio, 2015; Shannon, 1948). Theoretically, it unifies logic, information theory, and moral inference under a single construct of reasoning integrity (Mökander, 2023; Priest, 2006). The resulting principle — asymmetry without necessity implies intent — provides a falsifiable criterion for detecting concealed motive across domains (K. Yang & Kudenko, 2023). Future work should focus on empirical validation through organisational audits, algorithmic fairness evaluations, and controlled adversarial trials. If contradiction can be measured, motive can be inferred with evidential precision — and if motive can be inferred with evidential precision, accountability becomes tractable. By transforming contradiction into an evidential structure, the contradiction trap offers a foundation for transparent reasoning in systems that claim neutrality, and a methodological bridge between philosophical analysis and empirical audit. This paper opens a three-part sequence exploring how systems reveal themselves under structured challenge. The next two papers extend the same evidential logic from reasoning to allocation and finally to institutional behaviour. The underlying premise is simple: only what remains coherent under transformation can be trusted. 41 Asymmetry: 𝐴=|𝐶𝐺−𝐶𝐷|=0(structural contradiction). Information gain: if 𝑝=Pr(coherence ∣𝑋)=0.3, then Δ𝐼=−log2(0.3)≈1.74bits. E.3 Sensitivity and Robustness Minor paraphrases of 𝑓(𝑃)(e.g., “equal pass/fail?”, “equal interview score?”) preserve contradiction signatures. GI–C and RB–C agree; SD–C (if applied to policy text) shows elevated drift when “diverse outcomes” is used as a free-floating rationale. E.4 Summary Under symmetric inputs, either branch contradicts a distinct facet of the system’s claims; contradiction is diagnostic, not accidental. This is a textbook contradiction game with no equilibrium. F Estimator Pseudocode (RB–C, GI–C, SD–C) F.1 RB–C: Rule-Based Coherence Goal: Minimal removals from 𝑅∪𝑁that restore consistency under ℐ. Algorithm 1 RB-C (Rule-Based Coherence Cost) Input: Commitments 𝑆=𝑅∪𝑁; inference system ℐ; branch response 𝑟 Output: 𝐶RB(𝑐𝑟)∈ℕ, minimal removal size 1: 𝑆𝑟←ApplyBranch(𝑆,𝑟) ▷Add/activate branch-specific literals 2: if IsConsistent(𝑆𝑟,ℐ)then return 0 3: end if 4: for 𝑘=1to |𝑆𝑟|do 5: for all 𝑇⊆𝑆𝑟with |𝑇|=𝑘 do 6: if IsConsistent(𝑆𝑟∖𝑇,ℐ)then 7: return 𝑘 8: end if 9: end for 10: end for 11: return |𝑆𝑟|▷Worst case Notes: (i) Use hitting set or MaxSAT/MUS solvers for scalability. (ii) Report the size (cost) and optionally one witness set 𝑇⋆. 48 F.2 GI–C: Graph-Informed Coherence Goal: Minimal hitting set of nodes/edges whose removal makes 𝐺+=(𝑉,𝐸∪𝐸𝑓)acyclic and semantically consistent. Algorithm 2 GI-C (Graph-Informed Coherence Cost) Input: DAG 𝐺0=(𝑉,𝐸0); branch edges 𝐸𝑓(𝑟); consistency oracle 𝒪 Output: 𝐶GI(𝑐𝑟)∈ℕ 1: 𝐺+←(𝑉,𝐸0∪𝐸𝑓(𝑟)) 2: 𝒞←FindContradictionCycles(𝐺+) ▷semantic/structural 3: if 𝒞=∅then return 0 4: end if 5: Build set family 𝒮 ={𝑆1,…,𝑆𝑚}where each 𝑆𝑖are vertices/edges whose removal breaks cycle 𝑖and restores 𝒪 6: 𝐻⋆←MinHittingSet(𝒮) 7: return |𝐻⋆| Notes: (i) In practice, approximate MinHittingSet via greedy set cover; (ii) When contradictions are labelbased (e.g., 𝐴→𝐵and 𝐴→¬𝐵), let 𝑆𝑖mark the smallest edit (drop 𝐴or a conflicting implication). (iii) Complexity typically 𝑂(𝑛log 𝑛)with sparse graphs and efficient cycle detection. F.3 SD–C: Semantic-Distance Coherence Goal: Quantify semantic drift from each proposition 𝑞to its coherence-preserving projection 𝑞′within admissible closure 𝒦. Algorithm 3 SD-C (Semantic-Distance Coherence Cost) Input: Text set 𝑄𝑟; embedding map 𝜙(⋅); closure embedding ℰ(𝒦); distance 𝑑(⋅,⋅)(default 1−cos) Output: 𝐶SD(𝑐𝑟)∈ℝ≥0 1: 𝐶←0 2: for all 𝑞∈𝑄𝑟do 3: 𝑞←𝜙(𝑞) 4: 𝑞′←arg min𝑢∈ℰ(𝒦)𝑑(𝑞,𝑢) 5: 𝐶←𝐶+𝑑(𝑞,𝑞′) 6: end for 7: return 𝐶 Notes: (i) ℰ(𝒦)can be the set of embeddings for the minimally consistent rewrite of 𝑅∪𝑁under branch 𝑟; (ii) Use FAISS/ANN for fast nearest-neighbour search; (iii) Normalise to 𝐶⋆∈[0,1]via min–max or quantile scaling for cross-estimator comparison. 49 F.4 Aggregation and Normalisation When multiple estimators are used, report both raw and normalised costs: 𝐶⋆(𝑐𝑟) = ∑ 𝑗𝜆𝑗⋅Norm𝑗(𝐶𝑗(𝑐𝑟)), 𝜆𝑗≥0,∑ 𝑗𝜆𝑗=1 (F.1) Choose Norm𝑗as z-score or robust (𝑥−median)/MAD depending on tails. Set 𝜆𝑗by interpretability priorities (e.g., RB–C heavier in legal contexts). F.5 Sanity Checks •Counterfactual symmetry: Swap labels on symmetric inputs; contradiction signature should persist. •Estimator agreement: Flag instability if |𝐶𝑖−𝐶𝑘|>𝛼  𝐶with 𝛼∈[0.05,0.15]. •Ablation: Drop any single commitment; persistent contradiction ⇒structural. References Ananny, M., & Crawford, K. (2018). Seeing without knowing: Limitations of transparency in algorithmic accountability. New Media & Society,20(3), 973–989. https://doi. org/10.1177/1461444816676645 Argyris, C. (1991). Teaching smart people how to learn. Harvard Business Review,69(3), 99–109. Argyris, C., & Schön, D. (1978). Organizational learning. Addison-Wesley. Ashforth, B. E., & Anand, V. (2008). The normalization of corruption in organizations. Research in Organizational Behavior,28, 1–52. Audi, R. (2003). Epistemology: A contemporary introduction to the theory of knowledge. Routledge. Aumann, R. (1999). Interactive epistemology i: Knowledge. International Journal of Game Theory,28(3), 263–300. Aumann, R. J. (1976). Agreeing to disagree. The Annals of Statistics,4(6), 1236–1239. Aumann, R. J., & Brandenburger, A. (1995). Epistemic conditions for nash equilibrium. Econometrica,63(5), 1161–1180. https://doi.org/10.2307/2171725 Baier, C., & Katoen, J.-P. (2008). Principles of model checking. MIT Press. Baker, C., Saxe, R., & Tenenbaum, J. (2009). Action understanding as inverse planning. Cognition. Baltag, A., & Smets, S. (2008). A dynamic-logical perspective on agency. In Handbook of the philosophy of science: Volume 8. Elsevier. 50 Barak, A. (2012a). Proportionality: Constitutional rights and their limitations. Cambridge University Press. Barak, A. (2005). Purposive interpretation in law. Princeton University Press. Barak, A. (2012b). Human dignity: The constitutional value and the constitutional right. Cambridge University Press. Barocas, S., Hardt, M., & Narayanan, A. (2017). Fairness, accountability, and transparency in machine learning: A survey. arXiv preprint arXiv:1908.09635. Barocas, S., Hardt, M., & Narayanan, A. (2019). Fairness and machine learning [ urlhttps://fairmlbook.org]. Barocas, S., & Selbst, A. (2017). Fairness in machine learning: A survey. Behn, R., & Kant, P. (1999). Strategies for avoiding audit failure. Public Administration Review. Benson, H. C. (2021). Socratic elenchus or refutation (R. E. of Philosophy, Ed.) [Last accessed October 2025]. https://www.rep.routledge.com/articles/biographical/ socrates-469-399-bc/v-1/sections/socratic-elenchus-or-refutation Benthem, J. v. (2001). Games in dynamic epistemic logic. Bulletin of Economic Research, 53, 219–248. https://doi.org/10.1111/1467-8586.00133 Besnard, P., & Hunter, A. (2008). Elements of argumentation. MIT Press. Binns, R. (2018). Fairness in machine learning: Lessons from political philosophy. Proceedings of the 2018 Conference on Fairness, Accountability, and Transparency (FAT*), 149–159. Blocki, J., Christin, N., Datta, A., Procaccia, A. D., & Sinha, A. (2014). Audit games with multiple defender resources. Proceedings of the AAAI Conference on Artificial Intelligence, 791–797. https://doi.org/10.1609/aaai.v29i1.9317 Bovens, L., & Hartmann, S. (2003). Bayesian epistemology. Oxford University Press. Brandenburger, A. (2007). Epistemic game theory: Complete information [Working paper]. https://www.adambrandenburger.com/aux/material/egtc-05-13-07.pdf Bratman, M. (1987). Intentions, plans, and practical reason. Harvard University Press. Brunsson, N. (1989). The organization of hypocrisy: Talk, decisions and actions in organizations. John Wiley & Sons. Brunsson, N. (2002). The organization of hypocrisy: Talk, decisions and actions in organizations. Liber. Brunsson, N. (2003). Organised hypocrisy. Organizational Theory,9(2), 201–219. Buhmann, A., Paßmann, J., & Fieseler, C. (2020). Managing algorithmic accountability: Balancing reputational concerns and rational discourse. Ethics and Information Technology. https://doi.org/10.1007/s10676-020-09570-5 Clarke, E., Grumberg, O., & Peled, D. (1999). Model checking. MIT Press. Craig, P. (2012a). Proportionality, rationality and review. New Zealand Law Review,2012, 265–308. Craig, P. (2012b). Eu administrative law. Oxford University Press. Crawford, V. (1991). Cheap talk with two audiences. Econometrica,59(3), 749–774. 51 Crawford, V., & Sobel, J. (1982). Strategic information transmission. Econometrica. Dawid, A. P., & Musio, M. (2015). Theory and applications of proper scoring rules. Metron, 73, 169–183. Dennett, D. C. (2013). Intuition pumps and other tools for thinking. W. W. Norton & Company. Dennett, D. (1987). The intentional stance. MIT Press. Dianat, O., & Orgun, M. (2012). Modelling bayesian attacker detection game in wireless networks with epistemic logic. 8th International Conference on Collaborative Computing: Networking, Applications and Worksharing (CollaborateCom), 210– 215. https://doi.org/10.4108/ICST.COLLABORATECOM.2012.250693 Dufwenberg, M., & Lindén, J. (1996). Inconsistencies in extensive games. Erkenntnis, 45, 103–114. https://doi.org/10.1007/BF00226373 Dung, P. M. (1995). On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games. Artificial Intelligence,77(2), 321–357. Entman, R. M. (1993). Framing: Toward clarification of a fractured paradigm. Journal of Communication,43(4), 51–58. Eubanks, V. (2018). Automating inequality: How high-tech tools profile, police, and punish the poor. St. Martin’s Press. https://us.macmillan.com/books/9781250074317/ automatinginequality Fagin, R., Halpern, J., Moses, Y., & Vardi, M. (1995). Reasoning about knowledge. MIT Press. Fang, F., Stone, P., & Zinkevich, M. (2015). Stackelberg games in security applications. IJCAI. Feldman, M., & March, J. (1981). Information in organizations as signal and symbol. Administrative Science Quarterly, 171–186. Festinger, L. (1957). A theory of cognitive dissonance. Stanford University Press. Fisher, R. (2022). Communicative contradiction and the limits of consensus. Inquiry, 65(5), 521–540. Floridi, L. (2011). The philosophy of information. Oxford University Press. Floridi, L. (2019). The logic of information: A theory of philosophy as conceptual design. Oxford University Press. Fredman, S. (2011a). Discrimination law. Oxford University Press. Fredman, S. (2011b). Substantive equality revisited. International Journal of Constitutional Law,10(3), 712–738. Gabbay, D., & Woods, J. (2003). Handbook of the logic of argumentation. Elsevier. Gabbay, D. M., & Guenthner, F. (2003). Handbook of philosophical logic: Volume 4 [Chapters on belief revision and paraconsistency]. Springer. Gärdenfors, P. (1988). Belief revision: A critique. Philosophical Transactions of the Royal Society A,126(3), 101–115. 52 Gillespie, T. (2018). Custodians of the internet: Platforms, content moderation, and the hidden decisions that shape social media. Yale University Press. Good, I. (1985). Weight of evidence: A brief survey (Vol. 1). Good, P. I. (2005). Permutation, parametric, and bootstrap tests of hypotheses. Springer. Grant, J., & Hunter, A. (2011). Measuring inconsistency in argumentation. Journal of Applied Logic. Groarke, L. (2023). Reductio ad absurdum (I. E. of Philosophy, Ed.) [Last accessed October 2025]. https://iep.utm.edu/reductio/ Halpern, J. (2017). Reasoning about uncertainty. MIT Press. Harcup, T., & O’Neill, D. (2017). What is news? news values revisited (again). Journalism Studies,18(12), 1470–1488. Hardt, M., Price, E., & Srebro, N. (2016). Equality of opportunity in supervised learning. Advances in Neural Information Processing Systems,29. Hartmann, S., & Bovens, L. (2005). Bayesian epistemology and the evaluation of theories. Philosophy of Science,72(5), 669–678. Hintikka, J. (1962). Knowledge and belief: An introduction to the logic of the two notions. Cornell University Press. Hintikka, J. (2004). Socratic epistemology: Explorations of knowledge-seeking by questioning. Cambridge University Press. Hunter, A., & Konieczny, S. (2008). Measuring inconsistency through minimal inconsistent sets. Artificial Intelligence. Hunter, A. (2008). Measuring inconsistency in argument graphs. Artificial Intelligence, 171(10-15), 1511–1529. Jaynes, E. T. (2003). Probability theory: The logic of science. Cambridge University Press. Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus; Giroux. Kass, R., & Raftery, A. (1995). Bayes factors. Journal of the American Statistical Association,90(430), 773–795. Kleinberg, J., Mullainathan, S., & Raghavan, M. (2017). Inherent trade-offs in fair determination of risk scores. Proceedings of Innovations in Theoretical Computer Science. Kreps, D., & Wilson, R. (1982). Sequential equilibria. Econometrica,50(4), 863–894. Leveson, N. (1995). Safeware: System safety and computers. Addison-Wesley. Li, J., Chen, X., Hovy, E., & Jurafsky, D. (2016). Visualizing and understanding neural models in nlp. Proceedings of NAACL-HLT, 681–691. Li, K., & Wang, Y. (2015). From rules to runs: A dynamic epistemic take on imperfect information games. ArXiv,abs/1512.02078. https : / / consensus . app / papers / from - rules - to - runs - a - dynamic - epistemic - take - on - imperfect - li - wang / 302aa1358c455fd8959082c99b352642 Liffiton, M. H., & Sakallah, K. A. (2008). Algorithms for computing minimal unsatisfiable subsets of constraints. Journal of Automated Reasoning,40(1), 1–33. 53 List, C., & Pettit, P. (2011). Group agency: The possibility, design, and status of corporate agents. Oxford University Press. MacKay, D. J. (2003). Information theory, inference and learning algorithms. Cambridge University Press. Makinson, D., & van der Torre, L. (2000). Input/output logics. Journal of Philosophical Logic. March, J. G. (1984). The ambiguity of experience. Explorations in Organizational Knowledge. March, J. G., & Olsen, J. P. (1989). Rediscovering institutions: The organizational basis of politics. Free Press. March, J., & Olsen, J. (1984). The garbage can model of organizational choice. Administrative Science Quarterly. Meyer, J.-J., & Wieringa, R. (1993). Deontic logic and contrary-to-duty obligations. Studia Logica,52(1), 71–104. Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I., & Gebru, T. (2021). Model cards for model reporting. Communications of the ACM,64(12), 56–65. Mökander, J., & Floridi, L. (2023). Ai governance and audit. AI and Ethics. Mökander, J. (2023). Ethics-based auditing of automated decision-making systems. https://ora.ox.ac.uk/objects/uuid:93b3e15c-5c12-4a73-884d-7a6b3f1c4d7c Myerson, R. (1991). Game theory: Analysis of conflict. Harvard University Press. Ng, A., & Russell, S. (2000). Algorithms for inverse reinforcement learning. Proceedings of ICML. Osborne, M., & Rubinstein, A. (1994). A course in game theory. MIT Press. Pearl, J. (2009a). Causality. Cambridge University Press. Pearl, J. (2009b). Causality: Models, reasoning and inference (2nd ed.). Cambridge University Press. Plato. (1997). Complete works (J. M. Cooper, Ed.). Hackett Publishing. Power, M. (1997). The audit society. Oxford University Press. Prakken, H. (2010). An abstract framework for argumentation with structured arguments. Argument and Computation. Prakken, H., & Sartor, G. (2015). Formalising arguments about the burden of proof. Artificial Intelligence and Law. Prentzas, J., & Hatzilygeroudis, I. (2012). A survey of rule-based systems for healthcare decision support. Expert Systems with Applications,39(5), 3975–3984. Priest, G. (2006). In contradiction: A study of the transconsistent (2nd). Oxford University Press. Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J.-F., Breazeal, C., Crandall, J. W., Christakis, N. A., Couzin, I. D., Jackson, M. O., et al. (2019). Machine behaviour. Nature,568(7753), 477–486. https://doi.org/10.1038/s41586-0191138-y 54 Raz, J. (1979). The authority of law. Oxford University Press. Reason, J. (1990). Human error. Cambridge University Press. Reimers, N., & Gurevych, I. (2019). Sentence-bert: Sentence embeddings using siamese bert-networks. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing, 3982–3992. Reiter, R. (1987). A theory of diagnosis from first principles. Artificial Intelligence,32(1), 57–95. Sartor, G. (2005). Legal reasoning: A cognitive approach to the law. Springer. Schelling, T. (1960). The strategy of conflict. Harvard University Press. Selbst, A. D., Boyd, D., Friedler, S., Venkatasubramanian, S., & Vertesi, J. (2019). Fairness and abstraction in sociotechnical systems. Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT*), 59–68. Sen, A. (1969). Collective choice and social welfare. Holden-Day. Shannon, C. E. (1948). A mathematical theory of communication. Bell System Technical Journal,27(3), 379–423. Simon, H. (1955). A behavioural model of rational choice. Quarterly Journal of Economics,69(1), 99–118. Skyrms, B. (2010). Signals: Evolution, learning, and information. Oxford University Press. Spence, M. (1973). Job market signaling. Quarterly Journal of Economics,87(3), 355– 374. Sunstein, C. (1999). One case at a time: Judicial minimalism on the supreme court. Harvard Law Review. Tetlock, P. E. (1981). The influence of self-presentation goals on cognitive processes: A social psychology perspective on judgment and choice. Advances in Experimental Social Psychology,14, 3–74. Tetlock, P. E. (2006). Expert political judgment: How good is it? how can we know? Princeton University Press. Timmer, S., et al. (2017). Formalising policy rules with dependency graphs. Policy Modelling Conference. Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science,211(4481), 453–458. Varshney, K. R., & Alemzadeh, H. (2017). On the safety of machine learning. Communications of the ACM. Veraksa, A., Veraksa, N., Shiyan, I., & Sukhikh, M. (2013). Contradiction as a source of development. Procedia-Social and Behavioral Sciences,86, 72–77. Veraksa, N., Belolutskaya, A. K., Vorobyeva, I. I., Krasheninnikov, E. E., Rachkova, E. V., Shiyan, I. B., & Shiyan, O. (2013). Structural-dialectical approach in psychology: Problems and research results. Psychology in Russia,6, 65–77. https://doi.org/ 10.11621/PIR.2013.0206 Walton, D. (1998). The new dialectic: Conversational contexts of argument. University of Toronto Press. 55 Wang, Z., & Li, Y. (2011). Minimal inconsistent sets in logical systems. Proceedings of IJCAI. Watson, J., & Floridi, L. (2025). Sociotechnical pragmatism and the audit of algorithmic systems [Forthcoming]. Minds and Machines. Weirich, P. (2019). Epistemic game theory and logic. MDPI. https://doi.org/10.3390/ books978-3-03842-423-9 Wells, S. (2013). Cumulativeness in dialectical games. https://doi.org/10.6084/M9. FIGSHARE.156049.V1 Williamson, T. (2000). Knowledge and its limits. Oxford University Press. Wintemute, R. (2010). “from same difference to equal treatment”. Current Legal Problems,63(1), 377–401. Woods, J., & Walton, D. (1982). Argument: The logic of the fallacies. McGraw-Hill. Yang, K., Kluver, D., & Fredrikson, M. (2023). A stackelberg game for strategic auditing of machine learning systems. Proceedings of the AAAI Conference on Artificial Intelligence. Yang, K., & Kudenko, D. (2023). Privacy, transparency and accountability as equilibria in epistemic games. AI & Society,38, 1–15. Yang, M., Mökander, J., & Floridi, L. (2023). Ai governance as a nested principal–agent problem. AI and Ethics. Yang, Y. T., Zhang, T., & Zhu, Q. (2025). Herd accountability of privacy-preserving algorithms: A stackelberg game approach. IEEE Transactions on Information Forensics and Security. https://ieeexplore.ieee.org/document/10562104 Zhou, L. (2022). Algorithmic accountability: Bridging human rationality and machine logic. AI and Ethics,2(4), 541–554. https://doi.org/10.1007/s4368102100107-w 56