Different IP tools disagree because they answer different questions.

Why do checkers give different answers for one IP?

Different scopes, collection dates, source coverage and interpretation rules can produce different answers. Compare the actual claims before comparing their numbers; this report explains its own evidence, not the internals of every checker.

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.

Check one address on three sites and you will often get three different answers. This is not a sign that two of them are broken. They may describe different objects, scopes or observation times; an error is also possible. This report does not establish the correctness of an outside checker.

Nine mechanisms of disagreement

IP-level record versus ASN or prefix prior

One tool reports only what is recorded against the exact address. Another starts from what its network or surrounding prefix is generally like and applies that to everything inside. On a quiet address in a busy range these produce opposite results, because an exact record and a network prior answer different questions. That difference alone does not certify either tool’s accuracy.

CGNAT and shared egress

When an ISP places many subscribers behind one public address, that address accumulates the combined behaviour of all of them. Tools weighting observed activity see a busy multi-user pattern resembling proxy infrastructure; tools weighting operator identity see an ordinary consumer or mobile ISP. Those are possible explanations to investigate; a public lookup does not establish subscriber sharing or validate either classification.

Mobile reassignment

Carrier pools rotate. The subscriber on an address this morning may not be the one on it this afternoon, so observations at different times may concern different assignments. The address history does not identify individual users or prove that reassignment caused a discrepancy.

Anycast and CDN edges

One anycast address is announced from many locations at once, so "where is it" has no single answer. Geolocation databases resolve this differently — registered country, nearest observed edge, last place measured — so compare the actual source definitions rather than assuming a single physical location.

Dataset vintage

Every dataset has a collection date and a refresh cycle, and they do not align. A recently updated record can differ from an older one. A newer timestamp does not itself prove greater accuracy, and collection method may differ as well.

Registry country versus observed usage

A registry records the country of the organisation that received a block; geolocation estimates where the addresses are used. A European provider can legitimately deploy an allocation in Brazil, so registry and measurement name different countries, both correct about what they record.

Hosting treated as proxy

Some classifiers conflate "this address is in a datacenter" with "this address relays traffic," because commercial VPN exits do live in datacenters. The conflation catches the exits and also every self-hosted service, CI runner, and corporate gateway sharing that space. A tool keeping the two axes separate disagrees with one that does not, every time.

IPv6 coverage gaps

Several evidence families that cover IPv4 thoroughly return little or nothing for IPv6. One tool presents thin coverage as a confident negative while another declines to answer. The addresses are the same; the data behind the two answers is not.

Source abstention changing the denominator

When a source times out, is unavailable under its licence, or cannot handle the address family, an honest system records an abstention and a careless one drops it from the average. The same evidence then yields a different summary purely because of who answered — which is why IPJury publishes responding jurors, abstentions, and evidence-family counts alongside every verdict.

What to do with a conflict

Read the conflict as information rather than noise. Ask each tool what it measured, at what scope, and when; where that answer is unavailable, the absence is itself worth weighting. IPJury keeps material conflicts as structured dissent — what each source says, the scope of its claim, and the limits of the comparison — instead of averaging them into a figure that carries neither.

For what the number itself is made of, see what an IP purity score actually measures.