Clarification
The facts remain accurate, but wording, scope, uncertainty, or context could cause a reasonable reader to misunderstand the claim.
ACCOUNTABILITY / VERSION 1.0
Fix the sentence.
Then inspect the system.
Corrections are not evidence that journalism failed. Hidden corrections are. This page defines how Code Over Chaos records errors, changing evidence, retractions, and the human review status of AI-assisted reporting.
Effective August 14, 2026Public change noticesAI assistance disclosed
EVIDENCE
Information integrity requires visible uncertainty.
NIST's Generative AI Profile describes high-integrity information as distinguishing fact from fiction, opinion, and inference; acknowledging uncertainty; explaining its level of vetting; and linking claims to original evidence. It also identifies confabulation, automation bias, and information-integrity failures as risks that can scale through generative systems.
NIST's AI Risk Management Framework treats accountability, transparency, validity, and reliability as connected characteristics rather than interchangeable guarantees. A transparent workflow can still publish an error. Transparency makes the error easier to detect, remedy, and learn from.
CHANGE LABELS
Four names for four different events.
Correction
A factual statement, date, attribution, link, calculation, or material description was wrong and has been repaired.
Update
New primary evidence changes or materially extends the story after publication without making the earlier reporting false.
Retraction
The central claim is unsupported or materially unreliable. The page remains available when lawful and practical, with a prominent explanation.
INFERENCE / FRACTAL CHECK
An error has a shape beyond one sentence.
- CLAIMCorrect the inaccurate statement.
- ARTICLECheck nearby conclusions that depended on it.
- ARCHIVESearch for the same claim in older coverage.
- SYSTEMRepair the prompt, template, source rule, or review gap.
EDITORIAL VIEW
A clean record is less credible than an inspectable one.
Trust does not require pretending the first version was perfect. It requires leaving a readable trail from the weaker claim to the stronger one.
Material changes will receive a dated note describing what changed and why. The corrected article will preserve the conclusion the evidence now supports—not the confidence of its first draft. Silent edits are reserved for spelling, grammar, formatting, or broken links that do not change meaning.
AI REVIEW PROTOCOL
A label changes only when a human action occurs.
An article marked “AI draft — human review pending” remains in that state until a human editor checks the cited sources, material factual claims, evidence boundaries, safety, and links. A promised review window is a target, not an automatic approval timer.
Human review does not transfer authorship to the machine or guarantee permanent accuracy. Later evidence can still require an update or correction. When a material AI error is found, the correction notice should identify the error without blaming an opaque tool for the publication's editorial decision.
PUBLIC LEDGER
Corrections recorded from August 14, 2026.
No material corrections or retractions logged
This means no qualifying event has been recorded since the ledger opened. It is not a claim that every published sentence has been independently verified or is error-free.
Report a possible error ↗
Include the page URL, exact claim, supporting primary source, and why the evidence conflicts. Do not send credentials, private records, or unlawfully obtained material.
PRIMARY SOURCES
The integrity model behind the policy.
NIST AI Risk Management Framework 1.0 ↗
Voluntary framework connecting governance, measurement, transparency, validity, reliability, and risk management.
NIST AI 600-1: Generative AI Profile ↗
Primary guidance on information integrity, confabulation, automation bias, provenance, and human-AI configuration.
These NIST publications inform this voluntary editorial policy. They do not certify, endorse, audit, or govern Code Over Chaos.