Research Poster · Human Agency Policy Framework

Human Agency Under Consequential State: A Policy Framework for Correction, Revalidation, and Recoverable Consequences

Micheal Charles Preble · Independent Researcher · SSRN 7509878

Canonical paperFull PDF

Problem · Background · Research Question

Problem

Persistent computational systems can preserve the effects of earlier classifications and decisions after the state that supported them has changed.

Background

The EU AI Act, GDPR, and administrative-reconsideration doctrine each supply pieces of this problem; the paper connects them around one object — the continuing authority of consequences produced from superseded state.

Research question

Should the responsible actor determine whether materially dependent continuing consequences still possess an adequate basis after correction — and through what process?

Framework: the central rule

Correction should trigger revalidation, not automatic reversal.

Central diagram: the Consequence Revalidation lifecycle

Four forms of dependency

Necessary

The consequence could not have been sustained without the superseded state.

Contributory

The state materially affected the decision but was not its sole basis.

Incidental

The state was present but did not materially affect the outcome.

Indeterminate

Available evidence is insufficient to determine the state’s role.

Propositions / Observable Implications

  • Correction is not revalidation: correcting information does not by itself establish whether a decision made using it remains justified.
  • Output reversibility is not consequence recoverability: a score can change while a downstream record continues to encode the earlier conclusion.
  • Propagation is not control: an originator may be responsible for communicating a correction without possessing authority to reverse a recipient's independent decision.

Evidence Required

Real, cited regulatory and case authority, not new empirical data: EU AI Act Arts. 20 & 72; GDPR Arts. 16, 18–19; SCHUFA Holding (CJEU 2023) and Dun & Bradstreet Austria (CJEU 2025) on automated scoring; Lee v. Geren (D.D.C. 2007); 10 U.S.C. §628 special selection boards; Roth v. United States (Fed. Cir. 2004); Voyageur Outward Bound Sch. v. United States (D.D.C. 2020); UN Human Rights Committee General Comment 31; UNESCO's 2021 Recommendation on the Ethics of AI.

Limitations

Policy analysis and proposed framework; reports no new empirical results. Not a universal liability rule — does not determine damages, create jurisdiction, or displace sector-specific law. Explicitly bounded: revalidation should extend as far as material operative dependency extends, and no farther.

Falsifiers / Open Questions

The paper directly engages, without fully resolving, four hard objections: existing correction and automated-decision rights may already be sufficient (Sec. IX.K); revalidation duties could discourage correction (Sec. IX.L); meaningful participation can collide with confidentiality (Sec. IX.M); and high-volume systems create cost and scale constraints (Sec. IX.N).

Citation / QR

Suggested citation
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.

research.perfinitive.com/publications/human-agency-policy-framework