Compression Chains and Source-Access Failure: A Conditional Framework for Error-Correction Under Finite Cognition

This paper gives a regime-neutral account of a specific failure mode in public cognition: source-starved compression chains can preserve coordination while losing the capacity to correct their own distortions. The argument starts from finite cognitive capacity, a task-relative source-complexity scope condition, non-neutral compression objectives, and the data-processing inequality. Its central distinction is between source-injecting error-correction, denoted M5s, and coherence-only error-correction, denoted M5c. Coherence checks can improve consistency, decoding, and usability below an information ceiling; only source access can lift that ceiling. Mutual information and task-relative fidelity are kept distinct: the former bounds what a downstream layer can recover, while the latter describes which retained information survives selection. The empirical layer is separated into five numbered premises: P1, fidelity-decoupled or fidelity-adverse selection; P2, counterfeitable provenance cues; P3, channel-specific calibration lag; P4, maintenance-cost asymmetry; and P5, effective chain collapse. The paper does not rank regime types, does not claim that contemporary publics are always chains, and does not offer a predictive model of collapse. It identifies the conditions under which public inputs become source-starved enough for the chain model to apply, and it specifies where the framework can be empirically broken.