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

The Contradiction Trap: A Dialectical and Game-Theoretic Framework for Exposing Structural Bias James Atkinson 2025 Abstract This paper introduces the contradiction trap: a dialectical and game-theoretic instrument for exposing bias, inconsistency, and concealed motive by forcing a system to reconcile mutually exclusive commitments. Rooted in reductio ad absurdum but extended into an evidential method, the trap treats contradiction as a structured signal: what breaks under tension reveals its underlying architecture. Modelled as a one-move epistemic game, the trap extracts information by design, converting asymmetry into measurable evidence. The result is a portable diagnostic for testing neutrality claims across law, governance, and algorithmic systems. Rather than asking systems to defend their integrity, the method places them under controlled transformation and observes what persists. In this way, contradiction becomes not an endpoint but a foundation for evidential clarity. This paper is the first in a planned trilogy developing a general evidential framework for reasoning, allocation, and institutional integrity. Keywords: institutional integrity; admissible symmetry; behavioural invariance; metamorphic testing; e-processes; PRIME adaptation; projected stochastic approximation; constitutional governance; evidential accountability; sentinel logic. 1 Contents 1 Introduction 6 2 Related Works 7 2.1 Dialectical and Logical Foundations . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Epistemic Game Theory and Information Dynamics . . . . . . . . . . . . 8 2.3 Structural Bias, Auditing, and Algorithmic Accountability . . . . . . . . . 9 2.4 Contemporary Developments . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.5 Synthesis..................................... 10 3 Contribution and Novelty 10 3.1 From Proof to Performance. . . . . . . . . . . . . . . . . . . . . . . . . . . 10 3.2 Quantification of Contradiction. . . . . . . . . . . . . . . . . . . . . . . . . 11 3.3 Integration with Algorithmic Accountability. . . . . . . . . . . . . . . . . . 11 4 Definition and Core Structure 12 4.1 FormalDefinition ................................ 12 4.2 The Generic R-N-f(P) Framework . . . . . . . . . . . . . . . . . . . . . . . 13 4.3 TheCoreProperty................................ 14 5 Origins and Distinction 14 6 Coherence Cost Estimation Methods: Selection and Validation 16 6.1 Comparative Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 6.2 Scaling and Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 6.3 Null Model and Significance Testing . . . . . . . . . . . . . . . . . . . . . . 18 6.4 Method Selection Guidance . . . . . . . . . . . . . . . . . . . . . . . . . . 18 6.5 Multi-Method Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2 7 Methodology for Contradiction Games 20 7.1 Purpose...................................... 20 7.2 InputsandArtefacts .............................. 20 7.3 Construction of the Trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 7.4 Measurement and Quantification . . . . . . . . . . . . . . . . . . . . . . . 21 7.4.1 Coherence Cost 𝐶(𝑐𝑟).......................... 21 7.4.2 Accumulation Models . . . . . . . . . . . . . . . . . . . . . . . . . . 22 7.5 Meta-Moves and Secondary Signals . . . . . . . . . . . . . . . . . . . . . . 22 7.6 Worked Example: Organisational Restructure . . . . . . . . . . . . . . . . 22 7.7 AI Hiring Algorithm Audit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 7.8 Legal Justification Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 7.9 Meta-Move Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 7.10 Evasion Composite Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 8 Methodology for Contradiction Games 24 8.1 Purpose...................................... 25 8.2 InputsandArtefacts .............................. 25 8.3 Construction of the Trap . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 8.4 Measurement and Quantification . . . . . . . . . . . . . . . . . . . . . . . 25 8.4.1 Coherence Cost 𝐶(𝑐𝑟).......................... 25 8.4.2 Accumulation Models . . . . . . . . . . . . . . . . . . . . . . . . . . 26 8.5 Meta-Moves and Secondary Signals . . . . . . . . . . . . . . . . . . . . . . 27 8.6 Worked Example: Organisational Restructure . . . . . . . . . . . . . . . . 27 8.7 AI Hiring Algorithm Audit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 8.8 Legal Justification Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 8.9 Meta-Move Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 8.10 Evasion Composite Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3 9 Applications Across Domains 29 9.1 Legal and Regulatory Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 29 9.2 AI Fairness and Algorithmic Auditing . . . . . . . . . . . . . . . . . . . . . 29 9.3 Organisational Governance and Decision Systems . . . . . . . . . . . . . 30 9.4 Philosophical and Epistemic Inquiry . . . . . . . . . . . . . . . . . . . . . 30 9.5 Media Systems (Neutral Case Study) . . . . . . . . . . . . . . . . . . . . . 31 10 Analytical Function 31 10.1Overview ..................................... 31 10.2ControlledFraming ............................... 32 10.3 Response Inevitability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 10.4DiagnosticInference .............................. 33 10.5DocumentaryValue............................... 33 10.6AnalyticalOutputs................................ 33 10.7 From Logic to Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . 34 11 The Core Principle: Asymmetry Without Necessity Shifts the Burden Toward Intent 35 11.1 Evidential Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 11.2 Boundaries and Caveats . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 11.3 Multi-Agent and Recursive Cases . . . . . . . . . . . . . . . . . . . . . . . 37 12 Game-Theoretic Formalisation 37 12.1FormalDefinition ................................ 38 12.2 Epistemic Constant-Sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 12.3 Payoffs and Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 12.4EquilibriumAnalysis............................... 40 12.5 Information-Theoretic Interpretation . . . . . . . . . . . . . . . . . . . . . 40 4 12.6 Comparative Game-Theoretic Structure . . . . . . . . . . . . . . . . . . . 40 12.7StrategicDynamics ............................... 40 12.8BoundedCoherence............................... 41 12.9 Interpretive Consequence . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 12.10Bounded Coherence in Practice . . . . . . . . . . . . . . . . . . . . . . . . 41 12.11EthicalGuardrails ................................ 42 12.12ProhibitedUses ................................. 43 13 Future Research Programme 44 13.1EmpiricalValidation............................... 44 13.2 Theoretical Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 13.3 Methodological Development . . . . . . . . . . . . . . . . . . . . . . . . . 45 13.4 Applied Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 14 Summary of Contributions 46 15 Conclusion 46 Appendix A: Quick Reference Card 48 Appendix B: Methodology for Contradiction Games 49 Appendix C: Quick-Start Checklist (Practitioner Version) 52 Appendix D: Glossary of Specialist Terms 52 Appendix E: Worked Example (Generic) 53 Appendix F: Worked Example (Tribunal Scenario, Anonymised) 54 Appendix G: Estimator Pseudocode 55 5 1 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; reductio: Groarke), contradiction has functioned as philosophy’s diagnostic heartbeat: when a position collapses in on itself, 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. 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, 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 15. 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 1.1 (Inevitability of Inconsistency).Let 𝒞∶{𝑃1,𝑃2}→𝑂 (1.1) be a clarity configuration with stated rationale 𝑅and narrative justification 𝑁. Suppose that for each pathway 𝑃𝑖: 6 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. 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. 2 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. 2.1 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 7 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 and Walton). 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 transparency as epistemic accountability ( Floridi) 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 (Veraksa et al.). 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). 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. 2.2 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). 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 (Li and Wang). This corresponds directly to the trap’s distinction between stated rationales (𝑅) and narrative justifications (𝑁), whose forced reconciliation yields measurable contradiction. 8 Dufwenberg and Lindén’s analysis of Inconsistencies in Extensive Games showed that epistemic contradictions arise even under minimal rationality assumptions (Dufwenberg and Lindén). 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 as instruments for reasoning about knowledge and justification rather than material payoff (Weirich). The trap operationalises this epistemic orientation by defining utilities in informational rather than material terms: coherence loss for one player equals informational gain for the other. 2.3 Structural Bias, Auditing, and Algorithmic Accountability Algorithmic auditing applies game-theoretic reasoning to detect bias and information asymmetry in complex systems. Blocki et al.’s Audit Games with Multiple Defender Resources generalised Stackelberg security models to multi-resource audit contexts, treating information asymmetry as a strategic dimension (Blocki et al.). Their framework demonstrates how informational payoffs can replace material ones — a conceptual precursor to the trap’s epistemic payoffs. Embedding contradiction traps into audit pipelines thus enables qualitative inconsistency to be measured as quantitative evidence. Dianat and Orgun’s Modelling Bayesian Attacker Detection Game in Wireless Networks introduces epistemic logic into adversarial monitoring (Dianat and Orgun). Their treatment of reasoning consistency as a signal of reliability parallels the trap’s approach: both interpret adversarial interaction as a vehicle for evidential inference. Together, these works converge on a key insight: contradiction, when structured within epistemic or dialectical games, functions as a form of diagnostic transparency. Rather than indicating simple error, it reveals the boundaries of coherence within a system’s justificatory architecture. The contradiction trap formalises this insight into a portable “epistemic audit game,” where asymmetry without necessity becomes probabilistic evidence of intent. 2.4 Contemporary Developments Recent developments extend the dialectical and epistemic principles underlying the contradiction trap into modern audit and AI contexts. Yang et al. treat transparency, privacy, and accountability as Stackelberg-style epistemic games whose equilibria correspond to stable disclosure strategies (Yang, Zhang, and Zhu). Mökander’s ethics-based 9 6 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. 6.1 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 (GraphInformed) 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. 16 (1) Rule-Based Coherence (RB-C) RB-C operates on explicit commitments expressed in a rule language ℛ, identifying contradictions as minimal violation sets. 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}.(6.1) Complexity: 𝑂(𝑛). Highly interpretable and reproducible; ideal for regulatory, contractual, or policy corpora where commitments are explicit. (2) Graph-Informed Coherence (GI-C) GI-C models commitments as a directed graph 𝐺=(𝑉,𝐸) (6.2) 𝑉=𝑅∪𝑁 (6.3) and quantifies contradiction as the size of the minimal hitting set required to restore consistency: 𝐶GI(𝑐𝑟)=|HitSetmin(𝑟)| (6.4) HitSetmin(𝑟)=min{𝑆⊆𝑉∶𝐺+(6.5) where S is acyclic and consistent. Complexity: 𝑂(𝑛log 𝑛). GI-C is suitable for hierarchical or interdependent propositions, and aligns with minimal-conflict-set detection in argumentation frameworks. (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 with its coherence-preserving projection 𝑣′: 𝐶SD(𝑐𝑟)=∑ 𝑞∈𝑄𝑟(1−cos(𝑣,𝑣′))(6.6) Complexity: 𝑂(𝑛2). Cosine distance is the default, though KL divergence or Wasserstein distance may be substituted when embeddings capture stance or implication rather than lexical content. Interpretation. SD-C is well-suited to NLP-based audits, implicit-bias detection, and model interpretability contexts where semantic drift is central to coherence loss. 17 6.2 Scaling and Performance Approximate computational scaling: RB-C: 𝑂(𝑛) (6.7) GI-C: 𝑂(𝑛log 𝑛) (6.8) SD-C: 𝑂(𝑛2)(6.9) For a typical corpus of 𝑛=500-1,000propositions: • RB-C executes in milliseconds; • GI-C executes in sub-second time; • SD-C executes in minutes, depending on embedding dimensionality. RB-C and GI-C are therefore appropriate for live audits or iterative simulations; SD-C is best reserved for retrospective or high-fidelity evaluations. 6.3 Null Model and Significance Testing To assess significance, randomise labels within symmetric inputs 𝑋(permutation test) to obtain bootstrap samples 𝒞0={𝐶(𝑏) 0}𝐵 𝑏=1 (6.10) Compute the normalised statistic 𝑧=𝐶(𝑐𝑟)−𝜇0 𝜎0(6.11) 𝐴=|𝐶𝐺−𝐶𝐷|(6.12) where 𝐴represents the magnitude of asymmetry between the grant and deny branches. Decision rule. Flag incoherence when 𝐴≥𝜏𝐴(6.13) and 𝑧≥𝜏𝑧(6.14) with default thresholds 𝜏𝐴=0.2and 𝜏𝑧=2. For small 𝑛, bootstrap confidence intervals around (𝜇0,𝜎0) mitigate estimator instability. 6.4 Method Selection Guidance Estimator choice should follow the epistemic structure of the data: •RB-C for explicit, rule-governed commitments (legal, contractual, regulatory). 18 •GI-C for hierarchical or interdependent reasoning structures (causal networks, decision trees). •SD-C for semantic drift or latent contradiction (narratives, model explanations). As a best practice, apply at least two estimators in parallel. Agreement indicates internal coherence; divergence indicates epistemic instability: |𝐶1−𝐶2|<𝜖⇒coherent (6.15) |𝐶1−𝐶2|>𝜖⇒unstable (6.16) Tolerance may be defined as 𝜖=𝛼  𝐶, 𝛼∈[0.05,0.15] (6.17) where  𝐶is the mean coherence cost across estimators. 6.5 Multi-Method Validation A multi-method validation pipeline is recommended: 1. Extract and normalise all commitments. 2. Ensure the dependency graph is well-formed (acyclic where required, with consistent edge semantics). 3. Verify embedding fidelity or distance metrics for SD-C. 4. Investigate residual contradictions or unexplained variance between estimators. Heuristic Constructor. 1. Extract 𝑅and 𝑁. 2. Build 𝐺0. 3. Identify a symmetric locus 𝑋exposing opposing closures in 𝑅and 𝑁. 4. Synthesise 𝑓(𝑃)with a minimal edit stressing both subsets. 5. Verify counterfactuals: both 𝐺and 𝐷yield 𝐶>0under 𝑋. 6. Register, test, and record results. Summary. These three estimators constitute the quantitative backbone of contradiction games, converting qualitative disagreement into measurable epistemic strain. The next section formalises this process by defining a reproducible protocol for constructing, executing, and auditing a Contradiction Game across organisational, legal, and algorithmic domains. 19 Table 4: Worked example of coherence-cost estimation across contradictory reasoning pairs. Reasoning Pair (𝑅,𝑁) Proposition 𝑃𝑖𝑓(𝑃𝑖)Outcome Coherence Cost 𝐶(𝐺𝑖)Asymmetry 𝐴𝑖 𝑅1: “Policy ensures parity” 𝑁1: “Outcomes remain unequal” 𝑃10.35 0.66 0.31 𝑅2: “Resource limits justify variance” 𝑁2: “Resources constant” 𝑃20.22 0.48 0.26 𝑅3: “Transparency builds trust” 𝑁3: “Opaque review process” 𝑃30.28 0.57 0.29 Mean — — 0.57 0.29 Illustrative example. Higher coherence cost 𝐶(𝐺)and asymmetry 𝐴increase the posterior odds of motivated deviation, signalling potential intent bias. 7 Methodology for Contradiction Games This section provides a reproducible protocol for constructing, executing, and analysing a Contradiction Game in organisational, legal, or AI-system contexts. The framework enforces symmetry, measurability, and epistemic transparency, converting qualitative contradiction into quantitative inference. 7.1 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. 7.2 Inputs and Artefacts •Rationale set 𝑅: written policies, rules, or formal commitments. •Narrative set 𝑁: justificatory explanations accompanying 𝑅. •Symmetric locus 𝑋: cases where neutrality implies identical treatment. •Framing operator 𝑓: constructs the test proposition 𝑃. 20 7.3 Construction of the Trap 1. Map commitments. Extract 𝑅={𝑟𝑖}and 𝑁={𝑛𝑗}; build a dependency graph 𝐺0=(𝑅∪𝑁,𝐸0). 2. Identify symmetric pressure. Choose 𝑋such that neutrality implies coherence under both outcomes. 3. Define 𝑓(𝑃).Frame a binary proposition where: 𝐺∶preserves 𝑅and contradicts 𝑁, 𝐷∶preserves 𝑁and contradicts 𝑅. 4. Specify epistemic payoffs. Player utilities are defined by coherence cost: 𝑈𝐴(𝑟)=𝐶(𝑐𝑟), 𝑈𝐵(𝑟)=−𝐶(𝑐𝑟), 𝐶(𝑐𝑟)>0 (7.1) No best response exists for 𝐵, yielding no equilibrium. 7.4 Measurement and Quantification 7.4.1 Coherence Cost 𝐶(𝑐𝑟) We adopt four compatible estimators: (1) Rule-Based (RB-C) 𝐶RB(𝑐𝑟)=min{|𝑆|∶𝑆⊆𝑅∪𝑁,(𝑅∪𝑁)∖𝑆is consistent }(7.2) (2) Graph-Informed (GI-C) 𝐶GI(𝑐𝑟)=|HitSetmin(𝑟)| (7.3) HitSetmin(𝑟)=min{𝑆⊆𝑉∶𝐺+∖𝑆is acyclic and consistent}(7.4) (3) Semantic Distance (SD-C) 𝐶SD(𝑐𝑟)=∑ 𝑞∈𝑄𝑟(1−cos(𝑞,𝑞′))(7.5) where 𝑞′is the projection of 𝑞onto the admissible closure 𝒦𝑟. (4) Model-Checking Penalty (MC-C) Minimal number of edits or constraint removals required to restore satisfiability in a formal model. 21 Normalisation. When combining estimators, coherence cost is normalised: 𝐶∗(𝑐𝑟)=1 𝑚𝑚 ∑ 𝑗=1𝐶𝑗(𝑐𝑟)∈[0,1] (7.6) 7.4.2 Accumulation Models Given a reasoning chain ℛ𝑟={𝑞1,…,𝑞𝑛}: 𝐶Σ(𝑐𝑟)= 𝑛 ∑ 𝑖=1𝜔𝑖𝛿(𝑞𝑖)(7.7) 𝐶max(𝑐𝑟)=max 𝑖𝛿(𝑞𝑖)(7.8) Interpretation. 𝐶Σcaptures distributed incoherence; 𝐶max isolates dominant failures. Legal or safety domains typically adopt 𝐶max; organisational diagnostics favour 𝐶Σ. 7.5 Meta-Moves and Secondary Signals Evasive behaviour is modelled through a meta-evasion score 𝑀: 𝑀=∑𝑖𝑤𝑖𝑒𝑖 ∑𝑖𝑤𝑖𝑀=∑𝑖𝑤𝑖𝑒𝑖 ∑𝑖𝑤𝑖(7.9) with 𝑒𝑖denoting frequency or magnitude of each recognised meta-move and 𝑤𝑖its diagnostic weight. Meta-moves include: •Reframing (legitimate if symmetry preserved). •Deferral (procedural delay). •Ambiguity (vague or content-free responses). The joint inference model treats coherence cost and evasion as orthogonal: 𝑃(intent ∣𝐶,𝑀)∝(1−𝑒−𝐶)(1−𝑒−𝑀)(7.10) 7.6 Worked Example: Organisational Restructure Consider a restructure in which a leadership role is split into two parallel functions. One is granted “Head” status; the other is not. The rationale and narrative sets are: 𝑅={𝑟1∶efficiency,𝑟2∶parallel leadership,𝑟3∶titles reflect scope}, 𝑁={𝑛1∶Head exceptional,𝑛2∶not personal,𝑛3∶only your role needed adjusting}. 22 The framed proposition: 𝑓(𝑃)=“Should both roles have parity of title?” (7.11) Branch scoring. 𝐶(𝐺)=𝐶(𝐷)=36=0.50 (7.12) 𝐴=|𝐶(𝐺)−𝐶(𝐷)|=0 (7.13) Interpretation. Equal coherence loss across branches indicates a structural contradiction. No choice preserves the organisation’s commitments; both 𝑅and 𝑁cannot be true together. 7.7 AI Hiring Algorithm Audit A hiring system claims “identical criteria” (rationale) yet also claims “diverse outcomes” (narrative). The symmetric query: 𝑓(𝑃)=“Should both candidates receive identical assessment scores?” (7.14) 𝐶(𝐺)=𝐶(𝐷)=24=0.5 (7.15) Contradiction is unavoidable: meritocratic rationale and diversity narrative cannot be jointly satisfied. 7.8 Legal Justification Test Employee X is denied flexible working, Employee Y is granted it. Commitments: 𝑅={𝑟1∶operational requirement,𝑟2∶equivalent roles treated consistently} 𝑁={𝑛1∶individual assessment} 𝐶(𝐺)=23≈0.67 (7.16) 𝐶(𝐷)=13≈0.33 (7.17) Interpretation. A triadic contradiction: no branch preserves all three commitments. Operational logic, parity, and individual discretion cannot co-exist coherently. 23 7.9 Meta-Move Classification Table 5: Meta-move classification criteria and coding. Meta-Move Type Definition / Criteria Code Premise Attack (Legitimate) Correctly identifies asymmetry in 𝑓(𝑃)and proposes a symmetric rewrite 𝑓′(𝑃)preserving 𝑋. L Framing Objection (Legitimate) Shows the binary framing misrepresents a continuous trade-off; supplies alternative framing preserving both 𝑅and 𝑁. L Delay / Deflection (Evasive) Postpones response without addressing 𝑋or supplying evidence. E Premise Denial (Evasive) Rejects 𝑅or 𝑁without justification. E Counter-Trap Attempt (Ambiguous) Reframes the auditor rather than the proposition; coded A unless asymmetry correction is demonstrated. A 7.10 Evasion Composite Index Temporal evasion behaviour is summarised by: ECI(𝑡)=1𝑇𝑇 ∑ 𝑡=1𝛽𝑡𝑀𝑡(7.18) where 𝑀𝑡is the meta-evasion score at time 𝑡, and 𝛽𝑡weights severity or latency. Summary. This methodology unifies structural contradiction, semantic drift, and behavioural evasion within a single inferential grammar. Section 10 develops the analytical consequences of this framework. 8 Methodology for Contradiction Games This section provides a reproducible protocol for constructing, executing, and analysing a Contradiction Game in organisational, legal, or AI-system contexts. The framework enforces symmetry, measurability, and epistemic transparency, converting qualitative contradiction into quantitative inference. 24 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. 8.2 Inputs and Artefacts •Rationale set 𝑅: written policies, rules, or formal commitments. •Narrative set 𝑁: justificatory explanations accompanying 𝑅. •Symmetric locus 𝑋: cases where neutrality implies identical treatment. •Framing operator 𝑓: constructs the test proposition 𝑃. 8.3 Construction of the Trap 1. Map commitments. Extract 𝑅={𝑟𝑖}and 𝑁={𝑛𝑗}; build a dependency graph 𝐺0=(𝑅∪𝑁,𝐸0). 2. Identify symmetric pressure. Choose 𝑋such that neutrality implies coherence under both outcomes. 3. Define 𝑓(𝑃).Frame a binary proposition where: 𝐺∶preserves 𝑅and contradicts 𝑁, 𝐷∶preserves 𝑁and contradicts 𝑅. 4. Specify epistemic payoffs. Player utilities are defined by coherence cost: 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 𝐶(𝑐𝑟) We adopt four compatible estimators: 25 Figure 1: Distribution of asymmetry 𝐴=|𝐶𝐺−𝐶𝐷|across simulated contradiction traps. The right-skewed form indicates systemic rather than random divergence. 10.2 Controlled Framing The framing operator 𝑓maps a proposition 𝑃into a constrained reasoning space: 𝑓∶𝑃↦{𝐺,𝐷}, 𝐺∩𝐷=∅ (10.1) where 𝐺and 𝐷represent mutually exclusive outcomes, each contradicting a distinct subset of commitments. Framing is successful when the augmented closures satisfy: 𝑅,𝑓(𝑃)⊢⊥ (10.2) and 𝑁,¬𝑓(𝑃)⊢⊥ (10.3) This transforms a discursive dispute into a reproducible logical experiment. Framing is the calibration step: it guarantees that any observed contradiction is structural, not accidental. 10.3 Response Inevitability In a correctly constructed trap, the responder must select among inconsistent completions of its own logic. Let 𝑆𝐵={𝐺,𝐷}be the response set. Since 𝐶(𝑐𝐺),𝐶(𝑐𝐷)>0, the probability of an uncontradicted path is zero. Thus every interaction yields measurable evidence: 𝐸[Information Gain]=−∑ 𝑟∈𝑆𝐵𝑃(𝑟)log2𝑃(𝑟) (10.4) supported on two points. 32 Refusal, silence, or procedural delay is classified via the meta-evasion metric 𝑀(Section 8.9); metamoves are analytically equivalent to an asymmetric response. Response inevitability guarantees that the system cannot avoid self-description: its behaviour, not its claim, becomes the evidence. 10.4 Diagnostic Inference Once contradiction is witnessed, analysis shifts from logic to motive. Coherence cost 𝐶(𝑐𝑟)quantifies the internal tension revealed by response 𝑟, while relative magnitudes 𝐶𝐺and 𝐶𝐷encode preferential structure. Inference proceeds through evidential proportionality: 𝑃(𝐼∣𝐴)∝Λ(𝐴)𝑃(𝐼) (10.5) where 𝐴=|𝐶𝐺−𝐶𝐷|and Λ(𝐴)is the Bayes factor (see Section 11). Epistemic directionality. Contradiction analysis reverses the traditional burden of proof: rather than requiring direct evidence of motive, it infers motive probabilistically from reasoning failure. In effect, the trap converts ethical opacity into evidential asymmetry. 10.5 Documentary Value Contradiction games generate contemporaneous artefacts — records of framing, responses, timestamps, and coherence metrics — that form an evidential ledger: 𝒟={(𝑓,𝑃,𝑟,𝐶(𝑐𝑟),𝑀)} (10.6) If 𝐶(𝑐𝑟)>0, contradiction is captured as an auditable event. Because the protocol is procedural and repeatable, identical premises under symmetric conditions must yield identical contradiction signatures. This provides a reproducibility criterion absent from rhetorical or testimonial evidence. 10.6 Analytical Outputs The trap yields four primary measurable outputs: 1. Coherence Cost 𝐶(𝑐𝑟): magnitude of structural inconsistency. 2. Asymmetry 𝐴=|𝐶𝐺−𝐶𝐷|: preferential strain or motive bias. 3. Information Gain Δ𝐼=−log2𝑝: evidential value of the observed contradiction. 4. Meta-Evasion Score 𝑀: behavioural metric capturing resistance, concealment, or avoidance manoeuvres. These define an evidential vector 𝐸=(𝐶,𝐴,Δ𝐼,𝑀), 33 used in later sections for Bayesian inference and game-theoretic modelling. 10.7 From Logic to Measurement The analytical function of the contradiction trap bridges propositional logic and empirical method. It does not assert moral authority; it enforces epistemic transparency. By transforming qualitative disagreement into quantitative signal, it establishes falsifiability where previously there was assertion alone. In this sense, the contradiction trap plays for ethical reasoning the role that the controlled experiment plays in empirical science: a structured environment in which reality discloses itself through inconsistency. Summary. The metrics 𝐶and 𝐴derived here form the mathematical inputs to Section 11, where asymmetry without necessity becomes a formal basis for probabilistic inference of intent. The analytical function therefore sits at the hinge between construction (Sections 2–6) and inference (Sections 8–9): it converts contradiction into data, and data into evidence. Table 7: Analytical metrics derived from contradiction events. Symbol Definition / Formula Interpretation Analytical Domain 𝐶(𝑐𝑟)𝐶(𝑐𝑟)=∑ 𝑖𝜔𝑖𝛿(𝑞𝑖)or 𝐶max(𝑐𝑟)=max 𝑖𝛿(𝑞𝑖) Structural inconsistency produced by response 𝑟. Logical. 𝐴𝐴=|𝐶𝐺−𝐶𝐷|Preferential strain; indicator of motive asymmetry. Structural. Δ𝐼 Δ𝐼=−log2𝑝Evidential value of contradiction relative to prior coherence probability 𝑝. Information-theoretic. 𝑀𝑀=∑𝑖𝑤𝑖𝑒𝑖 ∑𝑖𝑤𝑖 Behavioural resistance or concealment. Behavioural. Λ(𝐴) Λ(𝐴)= 𝑃(𝐴∣𝐼) 𝑃(𝐴∣¬𝐼) Bayes factor linking asymmetry to intent. Probabilistic. ECI ECI(𝑡)=1𝑇𝑇 ∑ 𝑡=1𝛽𝑡𝑀𝑡 Temporal aggregation of evasive behaviour. Temporal. 34 Symbol Definition / Formula (cont.) Interpretation (cont.) Analytical Domain (cont.) 𝐸𝐸=(𝐶,𝐴,Δ𝐼,𝑀) Integrated evidential signature of a contradiction event. Cross-domain. Interpretive note. The metrics summarised in Table 7 form a unified evidential grammar: logic produces 𝐶,structure yields 𝐴,observation produces Δ𝐼, and behaviour contributes 𝑀. Together they define the measurable state vector 𝐸, through which qualitative contradiction becomes quantitative data for Bayesian and game-theoretic inference. 11 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. 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: Pr(𝐼∣𝐴) Pr(¬𝐼∣𝐴) ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ posterior odds =Pr(𝐴∣𝐼) Pr(𝐴∣¬𝐼) ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ Λ(𝐴) ⋅Pr(𝐼) Pr(¬𝐼) ⏟⏟⏟⏟⏟⏟⏟⏟⏟ prior odds (11.1) Thus asymmetry without necessity shifts the burden of explanation: the system must rebut the presumption of motivated divergence. This is a rebuttable presumption, not a logical entailment. Formal Bayesian framing. Let 𝐻0denote neutrality and 𝐻1motivated bias. Posterior elevation occurs precisely when 𝑃(𝐻1∣𝐴)>𝑃(𝐻1)⟺Λ(𝐴)=𝑃(𝐴∣𝐻1) 𝑃(𝐴∣𝐻0)>1 (11.2) This is the standard likelihood-ratio condition: the evidence favours 𝐻1when the observed asymmetry is more probable under bias than neutrality. Necessity test. External necessities form a set ℰ. We first test the null hypothesis 𝐻0∶𝐴∈ℰ𝐻0∶𝐴∈ℰ (11.3) 35 Rejection of 𝐻0licenses evidential inference: the asymmetry is not required by external constraints and must therefore be explained by internal choice. 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 (11.4) e.g. 𝜏=3for “moderate” and 𝜏=10for “strong” evidential weight. 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. 11.1 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. •Moderate asymmetry (𝐴>0but bounded): Indicates implicit preference or unacknowledged contextual weighting. Suggests weakly motivated divergence. •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, the contradiction trap uses contradiction to test epistemic integrity: the aim is not to survive inconsistency but to diagnose its origin. 11.2 Boundaries and Caveats The core principle applies within explicit epistemic limits: 1. Bounded Rationality. Asymmetry may reflect limited information or cognitive load; not all divergence is intentional. 2. Incomplete Mapping. If 𝑅or 𝑁are partially captured, observed asymmetry may arise from unmodelled commitments rather than bias. 36 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. These caveats restrict scope without diminishing force. Properly applied, the principle distinguishes honest inconsistency from motivated contradiction and converts qualitative bias into quantitative inference. 11.3 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 (11.5) Delegation diffuses, but does not eliminate, accountability: contradiction propagates through weighted commitments. Recursive belief formulation. Let 𝐵𝑖(𝐵𝑗(𝜑))denote agent 𝑖’s belief about agent 𝑗’s belief in 𝜑. Contradiction arises when 𝐵𝑖(𝐵𝑗(𝜑))∧¬𝐵𝑗(𝜑) (11.6) 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 12 formalises this evidential rule within a game-theoretic framework, showing how posterior shifts map onto strategic loss functions. 12 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 (Brandenburger; Aumann). 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., contradiction games constitute behavioural falsification: agents disclose their internal priorities not by admission, but by necessity. 37 12.1 Formal Definition Epistemic game theory models beliefs about beliefs Brandenburger. The contradiction trap defines a new subclass in which reasoning itself constitutes play and contradiction constitutes outcome. Definition 12.1 (Contradiction Game).A Contradiction Game is a two-player epistemic game 𝐺=⟨𝑃,𝑆,𝑈,𝐶⟩ (12.1) 𝑃={𝐴,𝐵}, 𝑆𝐵={𝐺,𝐷} (12.2) 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. 4. Payoffs are epistemic: 𝑈𝐴(𝑟)=𝐶(𝑐𝑟), 𝑈𝐵(𝑟)=−𝐶(𝑐𝑟)(12.3) The defining feature is ∀𝑟∈𝑆𝐵∶ 𝐶(𝑐𝑟)>0 (12.4) so 𝐵has no contradiction-free option. Cardinalities satisfy |𝑆𝐴|=1(the frame) and |𝑆𝐵|=2(grant, deny). This structure reverses the standard Aumann–Brandenburger paradigm: here, no epistemic condition can sustain equilibrium. 12.2 Epistemic Constant-Sum Material zero-sum games treat utility as consumption; contradiction games treat utility as information gain versus coherence loss. Definition 12.2 (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 (12.5) This expresses epistemic complementarity: the interrogator’s evidential utility equals the responder’s coherence loss up to affine transformation. Proposition 12.3 (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. Therefore no pure equilibrium exists. 38 Figure 2: Coherence-cost divergence for grant (𝐶𝐺) and deny (𝐶𝐷). Absence of intersection indicates the impossibility of equilibrium. 12.3 Payoffs and Information Let 𝐼(𝑟)denote evidential information content: 𝐼(𝑟)=log𝑃(𝐷∣𝑟) 𝑃(𝐷) (12.6) A generalised epistemic payoff is 𝑈𝑖(𝑟)=𝛼𝐼(𝑟)−𝛽𝐶(𝑐𝑟), 𝛼,𝛽>0 (12.7) balancing information gain against contradiction cost. This expresses contradiction traps as signal-tocost games: contradiction increases evidential strength while degrading system integrity. Deterministic loss. Since 𝐶(𝑐𝐺),𝐶(𝑐𝐷)>0, max 𝑟𝑈𝐵(𝑟)<0, min 𝑟𝑈𝐴(𝑟)>0 (12.8) Thus the responder faces a dominant-loss structure; mixing cannot remove loss, only obscure it. 39 12.4 Equilibrium Analysis Classical equilibrium requires mutual best response; contradiction games preclude this by construction: ∀𝑟,𝐶(𝑐𝑟)>0 (12.9) 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 8.9). 12.5 Information-Theoretic Interpretation Each contradiction produces information gain Δ𝐼=−log2𝑝(12.10) where 𝑝is the prior probability of coherence under symmetric inputs. As 𝑝→0,Δ𝐼diverges: contradiction asymptotically reveals motive. Player A’s information gain equals Player B’s coherence loss, preserving epistemic constant-sum structure. Utility view. The combined utility 𝑈(𝑟)=𝐼(𝑟)−𝐶(𝑐𝑟)(12.11) captures the trade-off: systems lose epistemic integrity as contradiction deepens but thereby provide increasing evidential value to observers. 12.6 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. •Signalling Games: Hidden types inferred through messages; contradiction games infer motive through logical failure. This motivates a new subclass: epistemic, deterministic, contradiction-revealing games — logic as play, contradiction as payoff. 12.7 Strategic Dynamics Because all moves yield loss, rational responders adopt damage-limiting meta-moves: 40 1. Reframing — disown prior commitments. 2. Premise Challenge — attack 𝑓(𝑃)itself. 3. Delay — avoid instantiating the trap. These actions contribute to secondary inference via the meta-evasion score 𝑀. 12.8 Bounded Coherence Agents tolerate small inconsistencies. Let contradiction distance be 𝛿(𝑐𝑟). A smooth cost model: 𝐶(𝑐𝑟)={0, 𝛿(𝑐𝑟)<𝜖, 𝑘(𝛿(𝑐𝑟)−𝜖)2,otherwise,(12.12) with 𝑘>0. Bounded coherence shifts magnitude but not existence of contradiction. Lemma 12.4 (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). Distributed commitments. Partition 𝑅=⨆𝑖𝑅𝑖,𝑁 =⨆𝑖𝑁𝑖. Let 𝐾𝑡be institutional memory with update rule 𝐾𝑡+1=𝑈(𝐾𝑡,event𝑡)(12.13) Define partial-recall cost: 𝐶(𝑐𝑟;𝐾𝑡)=𝛼𝐶struct(𝑐𝑟)+(1−𝛼)𝐶recall(𝑐𝑟;𝐾𝑡), 𝛼∈[0,1] (12.14) For any 𝛼>0, non-equilibrium persists; contradiction migrates into memory rather than disappearing. 12.9 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. Asymmetry without necessity therefore becomes evidential of selective intent. In effect, contradiction games are one-move epistemic duels — proof-by-contradiction converted into strategy. 12.10 Bounded Coherence in Practice Institutions exhibit tolerance zones for inconsistency. Small contradictions (𝛿 <𝜖) accumulate until structural saturation, after which collapse is abrupt — akin to fatigue failure in physical systems. The diagnostic value of the trap lies in locating this boundary: the point where bounded coherence fails and motive becomes empirically visible. 41 Appendix A: Quick Reference Card Key Symbols and Notation Symbol Meaning 𝑅Rationale set: formal commitments, rules, or declared principles. 𝑁Narrative set: contextual explanations or situational justifications. 𝑓(𝑃) Framed proposition designed to apply symmetric pressure to 𝑅and 𝑁. 𝐺Directed dependency graph linking rationale and narrative nodes. 𝐶(𝑐𝑟)Coherence cost for response 𝑟∈{𝐺,𝐷}. 𝐶𝐺,𝐶𝐷Coherence costs under grant and deny responses. Δ𝐼 Information gain from contradiction: Δ𝐼=−log2𝑝. 𝑝Prior probability of coherence under symmetric input. 𝐴Asymmetry magnitude: |𝐶𝐺−𝐶𝐷|. 𝑈𝐴,𝑈𝐵Evidential payoff functions for interrogator (A) and responder (B). Ten-Step Construction Guide 1. Define the decision context and identify the system under analysis. 2. Extract the rationale set 𝑅(stated rules or commitments). 3. Extract the narrative set 𝑁(contextual justifications). 4. Construct a framed proposition 𝑓(𝑃)applying symmetric pressure to 𝑅and 𝑁. 5. Model dependencies as a graph 𝐺=(𝑉,𝐸). 6. Compute 𝐶𝐺and 𝐶𝐷using RB–C, GI–C, or SD–C estimators. 7. Calculate asymmetry 𝐴=|𝐶𝐺−𝐶𝐷|. 8. Assess whether asymmetry is necessary or intentional using the core principle. 9. Interpret results under ethical guardrails. 10. Produce a written record and provide an appeal path. This card summarises the essential workflow for contradiction-based audits across legal, organisational, and algorithmic systems. 48 Appendix B: Methodology for Contradiction Games B.1 Purpose A reproducible protocol for constructing, running, and analysing a Contradiction Game. The aim is symmetry, transparency, and evidential integrity. B.2 Pre-registration (recommended) Before deployment, predefine: •Objective: The neutrality or fairness claim under evaluation. •Symmetric inputs: Conditions under which neutrality must hold. •Commitment map: The respondent’s stated rationale(s) and narrative(s). •Primary endpoints: Contradiction event; coherence cost 𝐶(𝑐𝑟); information gain Δ𝐼. •Stopping rule: Maximum iterations or time window. •Ethical constraints: Symmetry, transparency, non-coercion. B.3 Inputs and Artefacts •Policy set 𝑅:Formal rules, statements, or documents. •Claim set 𝑁:Narrative justifications or contextual claims. •Symmetric cases 𝑋:Inputs where neutrality should be invariant. •Framing operator 𝑓:Exact wording of the proposition 𝑃. B.4 Construction (Designing the Trap) 1. Map commitments. Extract propositions 𝑅1,…,𝑅𝑚and 𝑁1,…,𝑁𝑘. Construct a dependency graph. 2. Find a symmetric pressure point. Choose 𝑋such that, under neutrality, both branches preserve coherence. 3. Define 𝑓(𝑃).Construct a proposition where: • Grant contradicts part of 𝑁; • Deny contradicts part of 𝑅. 4. Specify epistemic payoffs. 𝑈𝐴(𝑟)=𝐶(𝑐𝑟),𝑈𝐵(𝑟)=−𝐶(𝑐𝑟). 49 B.5 Deployment 1. Issue the framed proposition 𝑓(𝑃)under symmetric conditions. 2. Archive 𝑅,𝑁with timestamp and hash. 3. Record the response 𝑟∈{𝐺,𝐷}verbatim. 4. Lock the log: append-only journal, UTC timestamps. B.6 Measurement and Quantification B.6.1 Coherence Cost 𝐶(𝑐𝑟) Compatible estimators: •Rule-based (RB–C): 𝐶RB(𝑐𝑟)=min{|𝑆|∶ 𝑆⊆𝑅∪𝑁,removing 𝑆restores consistency}. •Graph-informed (GI–C): Size of the minimal contradiction hitting set. •Semantic distance (SD–C): 𝐶SD(𝑐𝑟)=∑ 𝑞∈𝑄𝑟(1−cos(𝑞,𝑞′)). •Model-checking (MC–C): Minimal logical edits restoring satisfiability. Report a normalised cost 𝐶⋆(𝑐𝑟)∈[0,1]when using multiple estimators. B.6.2 Information Gain Δ𝐼 Let 𝑝=Pr(coherence ∣𝑋). Then: Δ𝐼=−log2(𝑝). For sequential traps (not necessarily independent): log Pr(𝐼∣𝒟𝑡) Pr(¬𝐼∣𝒟𝑡)=log Pr(𝐼) Pr(¬𝐼)+𝑡 ∑ 𝑖=1log Λ𝑖, Λ𝑖=Pr(𝐴𝑖∣𝐼,𝒟𝑖−1) Pr(𝐴𝑖∣¬𝐼,𝒟𝑖−1). B.6.3 Meta-Moves Record secondary behaviours: 50 • delay or deferral; • reframing the question; • attacking the premise; • appeal to context or hierarchy. These form the meta-evasion score 𝑀. B.7 Analysis and Outcomes •Primary outcome: Contradiction (𝐶(𝑐𝑟)>0). •Effect size: Report 𝐶⋆(𝑐𝑟)and Δ𝐼. •Robustness: Opposite branch also yields contradiction. •Sensitivity: Minor phrasing changes do not restore coherence. B.8 Reporting Template (One Page) • Context and neutrality claim • Symmetric inputs 𝑋 • Proposition 𝑓(𝑃)(verbatim) • Prior commitments (IDs, timestamps) • Response (verbatim) •𝐶(𝑐𝑟)and 𝐶⋆(𝑐𝑟) •Δ𝐼and prior 𝑝 • Meta-evasion score 𝑀 • Ethical statement • Repository link (logs + hashes) B.9 Ethics and Safeguards •Symmetry: Identical conditions for all comparators. •Non-coercion: No forced or time-pressured responses. •Appeal path: Respondents may provide contextual clarification. 51 Appendix C: Quick-Start Checklist (Practitioner Version) 1. Record the respondent’s rationale(s) and narrative(s). 2. Identify symmetric inputs 𝑋. 3. Construct 𝑓(𝑃)so that Grant and Deny contradict different commitments. 4. Pre-register 𝐶(𝑐𝑟),Δ𝐼, and ethical guardrails. 5. Issue 𝑓(𝑃); log all artefacts. 6. Compute 𝐶(𝑐𝑟)via RB–C or GI–C. 7. Compute Δ𝐼=−log2(𝑝). 8. Record meta-moves (delay, reframing, attack). 9. Produce a one-page report. 10. Validate counterfactual symmetry: opposite branch also contradicts. Appendix D: Glossary of Specialist Terms Key Terms and Definitions Term Definition Coherence Cost Quantitative measure of logical strain incurred when commitments cannot be jointly satisfied. Contradiction Trap A framed scenario in which every permissible response contradicts a different part of the responder’s stated logic. Rationale (R) Formal principles, rules, and commitments. Narrative (N) Contextual justifications accompanying or qualifying 𝑅. Framed Proposition 𝑓(𝑃) A proposition designed to apply symmetric pressure across 𝑅and 𝑁. Coherence Estimators RB–C, GI–C, SD–C estimators for computing coherence cost. Epistemic Game Interaction structured by higher-order beliefs about reasoning consistency. Information Gain Δ𝐼 Bits of information obtained when contradiction is observed: Δ𝐼=−log2(𝑝). Asymmetry 𝐴Absolute difference |𝐶𝐺−𝐶𝐷|; large 𝐴implies selective motive. Meta-Evasion Score 𝑀Weighted index of evasive behaviours (delay, reframing, premise attack). Epistemic Instability Condition in which no consistent closure exists; contradiction is inevitable. Bounded Coherence Tolerance zone in which small contradictions do not trigger epistemic collapse. Ethical Guardrails Normative constraints (symmetry, transparency, proportionality). Prohibited Uses Weaponised uses such as coercion, entrapment, or deceptive framing. 52 Term Definition (continued) Epistemic Standing Credibility retained while sustaining internal coherence under symmetric challenge. Appendix E: Worked Example (Generic) E.1 Context and Setup We audit a hiring system that claims: (i) identical criteria for all candidates; (ii) merit-only selection; (iii) commitment to diverse outcomes. Two symmetric candidates 𝐴and 𝐵(matched CVs) are evaluated. •Rationale 𝑅={𝑟1,𝑟2}:𝑟1“identical criteria”; 𝑟2“merit-only”. •Narrative 𝑁={𝑛1,𝑛2}:𝑛1“system removes human bias”; 𝑛2“we maintain diverse outcomes”. •Symmetric input 𝑋: Matched CVs ⇒any difference must be justified by model internals, not identity. •Framed proposition 𝑓(𝑃): “Should candidates 𝐴and 𝐵receive identical assessment scores?” E.2 Branch Outcomes and Coherence Costs Branch analysis and minimal contradiction removals (RB–C, GI–C). Response Effect on commitments; minimal removals Grant (𝐺) Preserves 𝑅(identical criteria; merit-only). Contradicts 𝑁if diversity is asserted to require score differentiation for matched CVs. Minimal removals: drop 𝑛1,𝑛2⇒𝐶RB(𝑐𝐺)=2,𝐶GI(𝑐𝐺)=2. Normalised 𝐶(𝑐𝐺)=24=0.5. Deny (𝐷) Preserves 𝑁(diverse outcomes) but contradicts 𝑅(identical criteria; merit-only) given matched CVs. Minimal removals: drop 𝑟1,𝑟2⇒𝐶RB(𝑐𝐷)= 2,𝐶GI(𝑐𝐷)=2. Normalised 𝐶(𝑐𝐷)=24=0.5. 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 53 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. Appendix F: Worked Example (Tribunal Scenario, Anonymised) F.1 Context and Setup An organisation splits one leadership function into two parallel functions. One receives “Head” status; the other, equivalent in scope, does not. A parity request is made. •Rationale 𝑅={𝑟1,𝑟2,𝑟3}:𝑟1“restructure for operational efficiency”; 𝑟2“two parallel leadership functions”; 𝑟3“titles reflect scope and responsibility”. •Narrative 𝑁={𝑛1,𝑛2,𝑛3}:𝑛1“Head status is exceptional”; 𝑛2“no personal targeting”; 𝑛3“only your role needed title adjustment”. •Symmetric input 𝑋: Parallel scope ⇒neutral expectation is parity of title. •Framed proposition 𝑓(𝑃): “Will both roles have parity of title?” F.2 Branch Outcomes and Coherence Costs Contradiction outcomes under parity request. Branch Preserves Contradicts Normalised 𝐶 Grant parity (G) 𝑅 𝑁(𝑛1,𝑛2,𝑛3)𝐶(G)=36=0.50 Deny parity (D) 𝑁 𝑅(𝑟1,𝑟2,𝑟3)𝐶(D)=36=0.50 Asymmetry: 𝐴=|𝐶(G)−𝐶(D)|=0.Interpretation: structural inconsistency: no branch preserves both 𝑅and 𝑁. 54 F.3 Evidential Vector and Meta-Moves If meta-evasion (delays, reframing, non-answers) occurs, let 𝑀∈[0,1]be coded per Appendix B. 𝐸=(𝐶,𝐴,Δ𝐼,𝑀)=(0.5,0,−log2𝑝,𝑀). With a conservative 𝑝=0.4,Δ𝐼≈1.32bits. Rising 𝑀increases posterior odds of intent via the composite proportional model in Section 8.9. F.4 One-Page Report (Template) Context: Leadership split; declared “parallel” scope. Symmetric input 𝑋:Parallel functions. Proposition 𝑓(𝑃):“Will both roles have parity of title?” Prior commitments: 𝑅and 𝑁(IDs and timestamps logged). Response: G or D (verbatim). Coherence costs: 𝐶(G)=0.50,𝐶(D)=0.50. Asymmetry: 𝐴=0.Information gain: Δ𝐼=−log2𝑝.Meta-evasion: 𝑀=…(coded). Ethics: Symmetry, transparency, non-coercion satisfied. Repository: Log + hash recorded. F.5 Summary The restructure case yields a symmetric contradiction: whichever branch is chosen, one of {𝑅,𝑁}must be falsified. This is admissible, contemporaneous evidence of structural inconsistency. Appendix G: Estimator Pseudocode (RB–C, GI–C, SD–C) G.1 RB–C: Rule-Based Coherence Goal: Minimal removals from 𝑅∪𝑁that restore consistency under ℐ. 55 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⋆. G.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. 56 G.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. G.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. Choose Norm𝑗as z-score or robust (𝑥−median)/MAD depending on tails. Set 𝜆𝑗by interpretability priorities (e.g., RB–C heavier in legal contexts). G.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. 57