Research Brief · Human Agency Policy Framework
A record was corrected. Does the decision it produced still deserve to stand?
Micheal Charles Preble · Independent Researcher · Policy paper (not the research paper)
The Question
When materially consequential computational state is corrected, withdrawn, superseded, or invalidated after decisions dependent on that state have already produced continuing effects, should the responsible originator or recipient determine whether those materially dependent continuing consequences still possess an adequate basis — and if so, through what process?
The Problem
Persistent computational systems can preserve the effects of earlier classifications, scores, inferences, and decisions after the state that supported them has changed. A restriction can continue after its authorizing determination is corrected. A downstream organization can retain a decision made using the former state. A derived classification can persist in another workflow. Correcting the source does not necessarily determine whether decisions made under the former state remain justified, or whether the consequences those decisions produced should continue.
The Contribution
Distinguishes correction from revalidation and develops a Consequence Revalidation Framework for persistent computational systems, proceeding through seven linked stages: Correction → Dependency → Provenance → Participation → Revalidation → Remedy → Closure. Introduces corrective continuity: the capacity of a valid correction to remain effective across the decision and consequence pathways through which superseded state would otherwise continue to affect a person over time. The central rule: correction should trigger revalidation, not automatic reversal — and not automatic continuation either.
How the Argument Works
A policy-institutional synthesis, not an empirical study, grounded in real regulatory and case authority: the EU AI Act's lifecycle obligations (Articles 20, 72), GDPR Articles 16 and 18–19, the CJEU's SCHUFA Holding and Dun & Bradstreet Austria decisions on automated scoring, U.S. federal privacy law (Lee v. Geren), military special selection boards (10 U.S.C. §628), Roth v. United States and Voyageur Outward Bound Sch. v. United States on administrative reconsideration, and international human-rights remedy doctrine (UN Human Rights Committee General Comment 31, the UN Guiding Principles on Business and Human Rights, UNESCO's Recommendation on the Ethics of AI). It classifies material dependency into four forms — necessary, contributory, incidental, and indeterminate — to prevent both indiscriminate reopening and treating uncertainty as proof that corrected state didn't matter.
Why It Matters
Correction changes state. Revalidation determines whether effects produced from the former state retain authority. The paper's central asymmetry test: does an institution's architecture preserve the effects of old state more effectively than it preserves valid correction of that state? Persistent systems must be capable of both remembering and superseding — continuity is not maximum retention.
Evidence Status
Evidence status: Policy analysis and proposed framework. It does not report new empirical results.
What This Does Not Claim
- Does not require automatic reversal when state changes, and does not require automatic continuation either — correction should trigger revalidation, not one fixed outcome.
- Does not require a full revalidation file for every material correction; the trigger is a still-operative consequence plus a plausible path of material dependency, not every conceivable causal effect.
- Is not a replacement for existing AI governance; it addresses what happens after those mechanisms have already done their work.
- Is not a universal liability rule — it does not determine damages, create jurisdiction, displace sector-specific law, eliminate finality, or establish one remedy for every domain.
- This is the policy paper (SSRN 7509878); it is distinct from, and does not import the empirical apparatus of, the companion research paper (SSRN 7496758).
Open Questions
The paper directly engages competing objections without fully resolving all of them: that revalidation duties could discourage correction by creating incentive effects (Sec. IX.L); that meaningful participation can collide with trade-secret and confidentiality interests (Sec. IX.M, citing Dun & Bradstreet Austria); and that high-volume systems create cost and scale constraints a case-by-case revalidation file cannot meet (Sec. IX.N).
Read / Cite the Research
Read the canonical paper record → · Full paper PDF
Preble, Micheal Charles. “Human Agency Under Consequential State: A Policy Framework for Correction, Revalidation, and Recoverable Consequences” Manuscript v4, September 2026. SSRN 7509878, submitted to SSRN, 2026.