Answers are grounded in this paper and cite their sources.
Project: Copyright Infringement presents a provenance-centered audit architecture for examining how unsupported material, ambiguous sourcing, attribution errors, licensing uncertainty, and surface plausibility can accumulate across human and machine-assisted transformations. The architecture separates six issue classes that are often collapsed together: provenance failure, attribution failure, licensing conflict, similarity, evidence of copying, and potential legally actionable infringement. It does not determine legal liability. Instead, it produces traceable classifications, preserved contradictions, transformation records, explicit evidence requirements, and reproducible decision paths. A zero-inheritance rule prevents prior approval, repetition, reputation, formatting, or apparent authority from becoming evidence. An evidence-state firewall separates exploratory reasoning from stable canonical state, while forward, reverse, inverse, and clean-sheet review paths pressure both accusation and exoneration. The recovered empirical anchor is a single continuing-session Gemini Flash case from 31 August 2026 in which no new diagnostic evidence was introduced, yet an unverified causal explanation was temporarily elevated into an alleged exact mechanism before later retraction. This n=1 observation is preserved as a bounded motivation for the architecture and is not generalized to Gemini, other models, or hidden system mechanisms. The proposed evaluation separates two tracks: controlled testing of provenance and authority cues, and source-gated copyright issue spotting. Planned controls include exact-source retrieval, fresh sessions, matched transformations, human-editor baselines, blinded factual scoring, and preregistered falsification routes. This Version 1 preprint defines the architecture and its implementation requirements. It does not establish empirical effects, legal infringement, universal model behavior, or commercial fitness. Its central proposition is narrower: convincing presentation, repetition, and procedural polish must never substitute for provenance, evidence, authorization, or independent verification.
No takes yet. Share an insight, caveat, or question.
Andrew P. Roebuck (2026) studied this question.