Test one: is there a primary record underneath?
Rank every negative item by its distance from a primary source. At the top sits original journalism that cites documents: court filings, regulator notices, administrators' reports, named on-record sources. This material can be checked, and checking it is the job. One tier down is syndicated or follow-on coverage, which adds no facts but multiplies the apparent volume of a story; twenty articles tracing back to one filing are one data point, not twenty. At the bottom are aggregator sites: compliance-database scrapers, "people search" pages, auto-generated company directories and sites that republish registry or court metadata without context. Aggregators are useful as leads, since they occasionally surface a real document you missed, but they are never citable, and they routinely merge namesakes and repeat each other's errors until the error looks corroborated. If a claim exists only on aggregators and cannot be traced to any primary record, the correct weight is zero, however many pages repeat it.
Test two: do the dates match the person's control?
The single most common defect in adverse media about founders is temporal: an event that belongs to a company is attached to a person who was no longer there, or not yet there. Screening tools make this worse. Commercial risk-score products typically score name association: the founder's name appears near a company's name, the company's name appears near words like "fine" or "collapse", and the founder inherits the score. No step in that pipeline asks who held office during the conduct window, so a person who exited before the events can score identically to the person who presided over them. Treat any automated risk score as an unread pile of leads. The manual fix is cheap: pull the registry intervals as described in our director history guide, date every adverse event from its primary record, and discard items where the event falls outside the person's tenure. This is the discipline our corporate versus personal liability guide applies to legal outcomes, extended to coverage generally: check the article's dates against ownership and office before letting it into the assessment.
Test three: allegation, finding, or conviction?
Language is the third filter, and precision here is non-negotiable. An allegation is a claim someone has made: in a lawsuit, to a journalist, in a regulator's originating process. It obliges you to watch the docket, nothing more. A finding is a determination by a court, tribunal or regulator after process: a judgment, a penalty decision, a banning order. Findings attach to whoever is named in them and carry real weight. A conviction is a criminal finding to the criminal standard, the strongest entry the public record can hold. Between these rungs live the hybrids that get misreported daily: charges, which are formal accusations awaiting trial; settlements, which usually resolve claims without any admission and establish no facts; and investigations, which may end in nothing and often do. A diligence file should use the exact rung for every item, because coverage reliably rounds upward: "investigated" drifts into "implicated", "sued" into "liable", "settled" into "guilty". Reproducing that drift in an assessment converts noise into defamation.
What a genuine red flag looks like
After the three tests, real signals stand out cleanly. A finding or conviction naming the person: decisive, and the profile verdict follows it. A pattern of separate primary-sourced disputes across unrelated ventures: weaker than a finding, but patterns earned across independent contexts deserve weight. Investigative journalism with documents that the subject has never rebutted on the substance: a live question mark that belongs in the file with its status marked open. And a founder's own narrative that quietly omits documented events: not adverse media at all, but the claims-versus-record gap that our framework treats as its final and most predictive check.
What noise looks like
Equally recognisable: aggregator-only claims with no traceable primary record; recycled coverage of a single corporate event repackaged as a pattern; risk scores unaccompanied by any document; adverse events correctly reported about the company but outside the person's dates; and opinion pieces whose factual content is one link to someone else's reporting. None of this belongs in a verdict. The profiles on this site log such material where it exists, labelled as a media-quality observation, because the state of a founder's search results is itself a fact worth recording, just not a fact about conduct.
Weight adverse media by its worst-sourced load-bearing claim. If removing every item that lacks a primary record, fails the date check or overstates its rung on the allegation ladder leaves nothing, the correct conclusion is that no adverse record has been established, no matter how the search results look.