Methodology

Last updated: September 23, 2026

Every score on CrazyCheck comes from the same process, applied the same way to every subject. This page explains it so you don’t have to take the number on faith — you can check our work.

Who does the research

Audits are produced by the CrazyCheck Agent, an AI research process, not a human analyst reading press releases. We say this plainly because pretending otherwise would be dishonest, and because the process is auditable precisely because every step leaves a trail: what was fetched, when, and from where. Each report’s provenance panel names the model and methodology version that produced it. We do not invent named human reviewers or fake credentials — if a claim in a report can’t be traced to a retrievable source, it is marked UNVERIFIED, not guessed.

The three pillars

A subject’s overall score is a weighted average of three pillar scores, each 0–100:

• Digital Integrity (30%).Domain age and registration history, legal entity status and jurisdiction, whether the people behind the business are verifiably real (not stock photos or AI-generated “team” pages), and what the site’s own technical footprint gives away.
• Consensus (40%) — what real users say. The heaviest weight, deliberately: we pull first-hand, dated, attributed accounts from Reddit, Trustpilot, Discord, forums, and other public discussion, and measure the balance of sentiment, plus how much of the marketing around a business is hype versus substantiated claims.
• Proof of Performance (30%). Specific, checkable claims the business makes — payouts, licences, guarantees, credentials — each individually verified as confirmed, partial, unverified, or contradicted, against regulator databases and other public records.

Some niches weight these pillars differently when the risk shape of the category calls for it (a licence matters more for a regulated financial product than a review site does), but the three pillars and the evidence standard below never change.

What the verdict tiers mean

• Verified (score 70–100).Strong evidence of legitimacy across all three pillars: a real, traceable entity; user consensus that’s net positive; and specific claims that hold up when checked.
• Cautious (45–69).A mixed picture — some things check out, some don’t, or the evidence is too thin to be confident either way. Read the pillar breakdown before deciding, not just the number.
• Avoid (below 45).Material red flags: unverifiable identity, a consistently negative consensus, or claims that don’t survive checking. We only use this tier when the evidence supports it — see the calibration note below.

Evidence standard

Every factual claim in a report traces to a dated, retrievable source logged in that report’s evidence ledger — a URL, an excerpt, and a timestamp. If a fact can’t be fetched or confirmed a second way, it is marked UNVERIFIED in the report rather than stated as though it were confirmed. We never fabricate a quote, a statistic, a date, or a source. This is a hard rule, not a guideline: one invented fact, discovered, would cost the credibility the whole site depends on.

Calibration — why verdicts aren’t all negative

A site that scores everything “Avoid” reads as a hit-piece farm; a site that scores everything well reads as paid placement. Neither is trustworthy, so we calibrate honestly in both directions — a business with a clean record gets a Verified score, and one with real problems gets an accurate Avoid, whichever the evidence supports. We never publish an accusation — “scam,” “fraud” — about a named company unless it is directly sourced; where the evidence doesn’t reach that bar, we report what named sources say and let the reader weigh it.

Corrections

Facts change — a company can fix what a report flagged, or new evidence can surface. If you find an error, or you run the business a report covers and want to respond, contact us and we’ll investigate and correct or update the report, with the change and date noted on the page. See About and Contact.