An IP purity score is not a universal platform verdict.

What does an IP purity number actually mean?

It is a product-specific interpretation whose direction, inputs and scope must be read with it. IPJury’s separate experimental index exposes its policy and limitations; it is not a universal scale or a measured probability.

Source availability depends on the current deployment and each returned record. These review dates do not establish when a particular IP was observed. Method axis-scope-2026-09-15 · Score ipjury-score-v3.

"IP purity" sounds like a property an address either has or lacks on a standard scale. It is not. It is community shorthand for a family of vendor-specific composite numbers, each assembled differently, each answering a question its author chose.

What goes into the number

Vendors rarely publish their formulas, but the ingredient classes are well known. A composite score typically draws on some subset of these.

• Network role priors. Whether the address sits in hosting, consumer, mobile, or enterprise space. Its weight is a product-policy choice; read the specific method rather than assuming a universal weighting.
• Anonymity signals. Proxy, VPN, Tor, or relay indications, from published operator lists down to naming heuristics — often flattened to one flag regardless of which produced it.
• Abuse reports. Submitted or observed reports, with weightings that may or may not decay with age.
• Neighbourhood history. What other addresses in the same prefix or ASN did, applied to this one by association.
• Geolocation mismatch. Disagreement between registered and estimated country, treated as suspicious in itself.
• Proprietary heuristics. Undocumented pattern detection, model output, or business rules that cannot be inspected from outside.

The list matters because those inputs are not commensurable. A hosting role is a fact about infrastructure; an abuse report is a claim about an event; a country mismatch is a discrepancy between two databases. Assigning them all a share of one number means deciding how many reports a datacenter is worth, and that exchange rate is invented, not measured.

Why two vendors' numbers are not comparable

Differences start before the data. Scales run in opposite directions — some count up to danger, some up to safety — ranges differ, and so do the thresholds a vendor calls high. One product is built for payment fraud, where a datacenter address genuinely is unusual for a consumer checkout; another for bot mitigation, another for email deliverability. Each weights its inputs toward its own question, so the same figure from two products is two quantities sharing a numeral, so this tool provides no conversion between them.

There is no universal score

No standards body defines IP purity, no registry publishes one, and no vendor's scale is an industry reference. Platforms do not publish which signals they buy or how they weight them, so any claim that a vendor's number decides whether a site will let you in is unverifiable from outside. Treat confident assertions in either direction with suspicion, including reassuring ones. IPJury's evidence verdict remains independent of its experimental policy index. The index has published, uncalibrated weights and explicit exclusions; it must not be read as a success probability or as proof that an IP is clean.

How to read a score you have been given

Read it as a claim, not a measurement, and interrogate it as you would any unsourced claim. Which axis moved the number: role, anonymity, abuse, or geolocation? Does the underlying record concern the exact address or its prefix neighbours? When was the finding observed, and when did the dataset last refresh? What would be left if the role prior were removed? If the tool cannot answer, the number is not evidence and cannot be disputed, corrected, or acted on — only believed or not.

What a number is legitimately good for

Composite scores are not useless. Inside one vendor's system, on one question, with a consistent threshold, a score is a reasonable triage device: it ranks a queue, routes borderline cases to review, and saves an operator from reading every field on every request. What it cannot do is travel. Outside the system that produced it, carry the reasons, scope, method version and limitations with the number rather than presenting it as a universal measurement.

For why the underlying findings differ in the first place, see why IP scores disagree.