Three Sins of Contextual Mismatch: Obscurity, Ambiguity, and Confabulation (medium.com)

Researchers propose the "Three Sins of Contextual Mismatch"—obscurity, ambiguity, and confabulation—as a diagnostic framework for failures that occur when models, terms, or metrics migrate between disciplines. Using Penn & Patty (2025) as a case study, the authors show how an otherwise valuable shift—bringing behavioral feedback into algorithmic fairness—can both illuminate policy questions and obscure the formal machinery that guarantees stability. Obscurity: borrowed terms lose their referents (e.g., equilibrium asserted but unproven). Ambiguity: distinct technical meanings collapse (e.g., "noise" as both stochastic damping and moral inclusion). Confabulation: narrative fills analytic voids, creating persuasive but fragile arguments. For the AI/ML community the lesson is practical and technical: once classifiers influence the behavior of agents you create a closed feedback loop best analyzed with control-theoretic tools (controller = designer’s classifier, plant = agents, state-space = compliance-cost distribution). The critique points to missing eigenvalue/Lyapunov stability analysis and absent dynamic-response proofs in Penn & Patty’s work, and cautions against treating stochastic regularization metaphors (dither/stochastic resonance) as general moral prescriptions. The recommended standard is “translate with structure”: either carry the full analytical invariants across contexts or explicitly redefine terms and provide the formal proofs that validate cross-domain claims.