Our method
Make the evidence
easy to question.
A clear explanation should make a claim more inspectable—not more difficult to challenge.
Four different questions
- What happened? Identify a dated action or a precise claim.
- Who establishes that? Link the original record and identify the reporter.
- What does it mean? Separate interpretation from the underlying facts.
- What changed? Require separate evidence for reach, persuasion, policy, or real-world outcomes.
Attribution stays visible
“X reports” or “the investigators assess” is not disposable wording. A platform's access to private account data may support findings that readers cannot independently reproduce. We name that boundary.
Suspected geographic origin does not automatically establish government sponsorship. A shared message is not proof of shared operators. Multiple news stories repeating the same disclosure are not multiple independent investigations.
Evidence labels, not mystery scores
| Label | What it means here |
|---|---|
| Platform disclosure / investigation | The platform reports findings from its own investigation; the source note explains public evidence limits. |
| Public record / government announcement | A document establishes a public action, statement, or proposal. Implementation is a separate question. |
| Public advocacy record | A visible position or demand. It is not classified as covert simply because it is controversial. |
| Research result | A study reports an evaluated result, within its design and limitations. |
| Operational use | A capability is documented in service, without automatically proving downstream social benefits. |
| Scenario analysis | A modeled possibility or analytical research direction, not a claim that the outcome has occurred. |
Investigate the narrative and its amplification separately
A deceptive account can promote an accurate fact. A sincere resident can raise a concern that also appears in a foreign campaign. We assess identity, coordination, factual content, and intent as separate questions.
Keep the denominator attached
Total network size, issue-specific accounts, posts, impressions, real engagement, and changed beliefs are different measures. We do not substitute one for another or turn missing impact data into a measured zero.
Survey trends need comparable questions
The public-opinion explorer keeps fieldwork dates, question summaries, response choices, populations, modes, party definitions, sample sizes, and original instruments attached to the numbers. We compare waves within a consistent survey series; a differently worded poll is a separate snapshot, not the next point on the same line.
Party labels can include or exclude independents who lean toward a party. Neutral or uncertain answers stay in the stated denominator. Percentage-point changes are not percent changes, strong opposition is not total opposition, and a full-sample margin of error is not a subgroup margin. Rounding and panel overlap also matter. Statistical significance is attributed to a published test, not guessed from the chart.
A shift in opinion is not, by itself, proof that an influence operation caused it. That requires separate evidence about exposure and effects.
Benefits need the same discipline
Capability, deployment, and public benefit are different milestones. A prediction benchmark is not a treatment trial. A treatment trial is not proof of longer population-wide lifespan. Greater production is not automatically broadly shared abundance.
How these briefs are prepared
Research and drafting are AI-assisted. Each brief is checked against its cited material for this publication, with review dates and unresolved questions visible. “Reviewed” does not imply review by an independent human specialist. The initial collection consists of source briefs, not original undercover or forensic investigations.
Material revisions should be explained in the corrections record. The source register provides a direct route to the underlying documents.