The Correction Stack: How criticism reaches decisions in recurring AI systems

A recurring AI workflow can repeat an error after a reviewer has identified it. The objection may lose its evidence during synthesis, arrive after a decision, or fail to change the reference material that later work reuses. This paper develops the correction stack as a framework for auditing that route. It distinguishes detection, carriage, assessment, authorization and verification, supported by eight diagnostic functions. These functions are a proposed taxonomy, not a proven minimum architecture. The unit of audit is a claim version, its objection and the destinations at which the claim will be acted upon or reused. A proposed transport study supplies the same objections to competing workflows and separates ordinary handling, explicit records and enforcement of disposition rules. A second study varies model composition and correction pathways independently. Outcome rules count false corrections, stale downstream copies, missed deadlines and justified holds separately. No experiment or implementation evaluation is reported. The proposal concerns functional correction and makes no inference about consciousness. Edition and provenance: AX-41, second edition 2.0, 14 November 2026. This is a revised standalone working paper by Michael S. Moniz. The selected source edition appears in THE AXIS: Collected Papers and Source Archive, reading edition 0.2, Volume V, pp. 77-83, DOI 10.5281/zenodo.24207813. This record is derived from that source collection; it is not a replacement or new version of the entire collection. The June 2026 date belongs to the source manuscript's stated edition record. Development drafts and the original reconciled manuscript are retained in the accompanying archive. Files: reading PDF, editable DOCX, plain-text manuscript, and a supplementary archive containing source history, editorial changes, the targeted primary-source check and the proposed evaluation design. No completed experimental data or executable evaluation system is claimed. Michael S. Moniz is author of record. ChatGPT assisted with criticism, drafting, literature comparison and editorial preparation under his direction. CC BY-NC-SA 4.0 applies to the author's original material; cited works retain their own rights.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-14
DOI
https://doi.org/10.5281/zenodo.24219908
Primary Topic
Ethics and Social Impacts of AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

The Correction Stack: How criticism reaches decisions in recurring AI systems

Michael S. Moniz
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
article

The Correction Stack: How criticism reaches decisions in recurring AI systems

Michael S. Moniz
article en

Abstract

A recurring AI workflow can repeat an error after a reviewer has identified it. The objection may lose its evidence during synthesis, arrive after a decision, or fail to change the reference material that later work reuses. This paper develops the correction stack as a framework for auditing that route. It distinguishes detection, carriage, assessment, authorization and verification, supported by eight diagnostic functions. These functions are a proposed taxonomy, not a proven minimum architecture. The unit of audit is a claim version, its objection and the destinations at which the claim will be acted upon or reused. A proposed transport study supplies the same objections to competing workflows and separates ordinary handling, explicit records and enforcement of disposition rules. A second study varies model composition and correction pathways independently. Outcome rules count false corrections, stale downstream copies, missed deadlines and justified holds separately. No experiment or implementation evaluation is reported. The proposal concerns functional correction and makes no inference about consciousness. Edition and provenance: AX-41, second edition 2.0, 14 November 2026. This is a revised standalone working paper by Michael S. Moniz. The selected source edition appears in THE AXIS: Collected Papers and Source Archive, reading edition 0.2, Volume V, pp. 77-83, DOI 10.5281/zenodo.24207813. This record is derived from that source collection; it is not a replacement or new version of the entire collection. The June 2026 date belongs to the source manuscript's stated edition record. Development drafts and the original reconciled manuscript are retained in the accompanying archive. Files: reading PDF, editable DOCX, plain-text manuscript, and a supplementary archive containing source history, editorial changes, the targeted primary-source check and the proposed evaluation design. No completed experimental data or executable evaluation system is claimed. Michael S. Moniz is author of record. ChatGPT assisted with criticism, drafting, literature comparison and editorial preparation under his direction. CC BY-NC-SA 4.0 applies to the author's original material; cited works retain their own rights.

Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon