ABSTRACT
This Article introduces the Structural Defect Framework (SDF) to address a critical failure in modern liability doctrine: the law’s tendency to misclassify harms produced by invisible structural defects—architectural design choices hidden from end-users but controlled by institutional designers. Across four disparate case studies – adaptive streaming (Netflix-style bitrate attenuation), model quantization in legal AI, the systemic hollowing of Arizona’s victims’ ‘right to confer’, and unstructured discretion in Title VII employment decisions – the Article demonstrates how systems preserve outward continuity while degrading, or failing to constitute, normatively relevant signal, producing predictable, patterned harms that doctrine wrongly dismisses as mere optimization, random error, or individual malice.
SDF proceeds as one framework with two analytic arms. The Transmissive (T) arm addresses systems where a bounded artifact exists upstream and is delivered downstream under stable outward identity, while a concrete fidelity parameter (a ‘knob’ such as bitrate or bit-precision) is reduced relative to a higher-fidelity baseline. The Proxy-Mediated (P) arm addresses systems where the authoritative decision signal does not pre-exist as a master artifact but is constituted through institutional proxy capture and record practices; here the decisive attenuation parameter is the system’s capture and reviewability capacity – how much normatively relevant information the architecture requires, preserves, and makes contestable. Many legally salient systems operate as mixed T→P pipelines, in which transmitted artifacts are reconstituted into proxy records that become legally authoritative.
The framework begins with a threshold premise: fidelity is a choice – either a transmissive fidelity parameter that can be tuned down, or a capture architecture that can be left thin, while outward continuity remains stable. This premise is followed by a tripartite gateway that engineers Structural Invisibility: (1) a fixed infrastructure, interface, or measurement layer that stabilizes the user’s view; (2) nominal equivalence, where distinct configurations or processes are treated as one; and (3) opacity at the moment of reliance, where the tradeoff is non-auditable. Where these preconditions hold, a four-element diagnostic predicts where harm will occur and persist: (1) Structural Invisibility; (2) Foreseeable Harm; (3) Doctrinal Mismatch; and (4) Epistemic Asymmetry. By formalizing these patterns, the Article reframes AI safety, civil rights, consumer protection, and victims’ rights as parallel instances of designed-in, inspection-resistant harm. A companion paper extends and operationalizes the two-arm framework across mixed pipelines and institutional settings.
Chung, Jon, Structural Defect Framework (December 12, 2025).
Leave a Reply