Answers are grounded in this paper and cite their sources.
Purpose This paper aims to ask what happens after AI uncertainty has been quantified but an institution commits to a consequential action. This study develops a governance account of excluded but decision-relevant alternatives using Schelling’s ground, separation and remainder. Design/methodology/approach This paper combines conceptual analysis, system design and a protocol-tier proof of concept. It separates predictive uncertainty, hallucination, thermodynamic irreversibility and qualified residuals, then specifies a Qualified Residual Ledger (QRL) with qualification, commit-before-act receipts and threat-model-dependent anchoring. Findings Residual governance concerns the history and hierarchy of consequential exclusion. A reference implementation satisfied the specified conformance tests for receipt omission, pruning, cross-action receipt substitution, keyless forgery and retroactive edits, while leaving insider resistance and domain-level signal validity open. Landauer remains a material boundary; contestability and evidential asymmetry ground the normative claim. Research limitations/implications Protocol enforceability is tested only on a reference prototype. Evaluator quality, residual-expert alignment, threshold stability, reviewer load, privacy and domain deployment remain empirical questions. Practical implications Deployers can precommit failure policies, keep sensitive evidence off-chain and match integrity anchors to the relevant threat model. Social implications Precommitted residual evidence can support appeal and oversight without public disclosure of raw data when retention, access and erasure duties are designed into the system. Originality/value This paper shifts AI governance from uncertainty estimation to auditable preservation of qualified rejected alternatives and links Schellingian Verkehrung to the distinction between evidential falsification and rule inversion.
No takes yet. Share an insight, caveat, or question.
Se Hoon Son (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: