Where AI governance meets operational reality.

EDITORIAL & INDEPENDENCE

Make the source, judgment and limits visible.

Research should show what was examined, how it informs the conclusion and what remains unresolved. Commercial relationships must not determine the finding.

These principles guide publication and commissioned work: attributable evidence, relevant disclosures, traceable corrections and candid treatment of conflicts.

Read the editorial principles →
Evidence & disclosure

Separate facts from interpretation.

Attribute material sources and distinguish verified records, reported claims, controlled simulations and ODA3 analytical interpretation. Document Notably Absent alongside findings.

Before publishing substantive research, check and state its funding basis and relevant interests. Commissioned work identifies the funder’s role in design, analysis and publication where disclosure is permitted; confidentiality constraints must themselves be made clear where material.

Disclose material AI assistance and retain responsibility for verification, analysis and editorial decisions. Tool use does not substitute for an attributable evidence basis.

RECORD

What was examined

Identify the material source, its origin and the verification performed against defined criteria.

CLAIM

What was reported

Attribute third-party statements and distinguish them from findings established through examination.

INTERPRETATION

What we infer

Explain the reasoning, assumptions, conflicting evidence and limits behind the analysis.

Label controlled simulations and synthetic examples. Their results do not establish behaviour in an unexamined deployment.

Read the evidence discipline →

Corrections

Make substantive changes traceable.

Retain the original publication date. Changes to evidence, findings, conclusions or recommendations receive an updated date and a concise correction note. Minor spelling and formatting edits need no public note. Controlled documents retain formal version histories.

Where a publication is withdrawn or superseded, make that status and the replacement reference clear. A substantive correction must be visible to readers.

Publication and version-control guidance →

Independence

Commercial access cannot purchase a conclusion.

Screen relevant conflicts before acceptance. Exclude work where conflicts cannot be managed. Agree publication and confidentiality terms before starting; report unsupported, negative and inconclusive results accurately.

Commissioning, membership, partnership or training relationships do not entitle a party to a favourable finding. Necessary factual review must not suppress a supported adverse result or an unresolved evidence gap.

Notably Absent

  • No guaranteed favourable result or endorsement purchased through an engagement.
  • No claim that stating a conflict automatically resolves it.
  • No implied external review or standing independent review panel.

Read Claims & Limits →

Concerns

Request a correction or challenge a finding.

Identify the publication or engagement, the disputed statement and the basis of your concern. Do not submit confidential evidence through the initial form. If the concern involves the founder’s own work, explain the limits of internal review. Seek independent review where appropriate and available, and make clear whether it has actually been obtained.

Include the title or identifier, version or date, disputed statement and relevant source references. Please keep confidential records out of the initial message.