Paste a proxy exit into any IP lookup tool and you will usually get a number back: a fraud score, a risk score, an IP quality score. It looks like a verdict on the address, and it is tempting to read it as a verdict on the provider that sold it to you. If you are choosing residential proxies for data collection or testing, a column of high scores feels like a warning and a column of zeros feels like a guarantee.
Neither reading holds up well. A fraud score is one vendor’s estimate of how risky a visitor from that address is, built for a fraud team deciding whether to approve a payment or a sign-up. It answers a different question from the one a proxy buyer is asking. This guide explains what goes into the number, why residential proxy IPs score the way they do, and what to measure instead.
What an IP fraud score is
An IP fraud score is a number, usually on a scale of 0 to 100, that an IP intelligence vendor attaches to an address. Higher means riskier. The vendor sells it to sites that want a quick signal at the moment a user logs in, checks out or creates an account.
Each vendor builds its own, from inputs like these:
- Connection type. Whether the address belongs to a consumer ISP, a mobile carrier, a hosting company or a business network. Hosting addresses start out riskier.
- Proxy and VPN detection. Whether the vendor believes the address is a proxy exit, a VPN endpoint or a Tor node, often from its own honeypots and traffic partners.
- Abuse history. Reports of spam, credential stuffing, card testing or scanning tied to the address, and how recent they are.
- Blocklist presence. Whether the address appears on public reputation lists.
- Activity across the vendor’s customers. How many accounts or devices the address touched recently, and what happened to them.
“IP quality score” is the same idea under a different name, and at least one vendor uses it as a brand. For this guide the terms are interchangeable.
Why scores from different vendors disagree
There is no shared standard. Each vendor weighs the inputs differently, sees a different slice of internet traffic, and refreshes its data on its own schedule. The same address can score low with one service and high with another on the same afternoon.
The gap is widest on the exact question proxy buyers care about. Look up one set of residential exits with two detection services and it is common for one to flag a large share of them as proxies and the other almost none. Neither was broken. They simply had different data about which home connections carry proxy traffic.
So a single score from a single vendor is a weak basis for judging a pool. Compare two or three vendors before drawing any conclusion, and treat agreement between them as more meaningful than any one number.
What a score says about a residential proxy IP
A residential proxy exit is a real home connection carrying traffic it did not originate. That has two consequences for its score.
First, a well-sourced residential IP will often score better than a hosting IP, because its connection type is a consumer ISP. That part of the score is genuinely useful: it is one of several ways to catch hosting addresses sold as residential, a problem covered in depth in how to spot datacenter IPs sold as residential.
Second, the better a vendor’s proxy detection, the more likely it is to know that a given home connection has carried proxy traffic before. A rising score on a genuine residential IP can mean the vendor caught it, not that it is fake. And because exits are shared across customers, an address can pick up abuse history from someone else’s traffic, which is why IP reputation moves over time.
In other words, a fraud score mixes two things: “is this a real consumer connection” and “has this connection been seen doing proxy things”. The first matters a great deal when judging a provider. The second is true of every residential pool to some degree.
Why a fraud score is not a success rate
The sites you actually send requests to may not use the vendor you checked. Many run their own risk models, combine IP signals with browser fingerprints and request patterns, or barely look at the IP at all. A low score does not mean a target will let you through, and a high score does not mean it will block you.
What the target does is the only outcome you pay for. That is why Atlas measures success on real targets, site by site, with challenge pages counted as failures, rather than inferring quality from a lookup.
What Atlas measures instead
Atlas does not rank networks on any vendor’s fraud score. It measures the parts of the question that can be counted the same way for every network, across every exit it sees in a window. The authenticity measurements in the methodology are:
- DC ASNs. The share of unique exit IPs whose usage type is hosting or datacenter, on networks sold as residential. This is the connection-type part of a fraud score, isolated and counted across the whole observed pool.
- Authenticity. The share of unique exits that geolocate to the country the requests asked for. A pool sold by country is only worth what it delivers there.
- Blocklist presence. The share of sampled exits that appear on public reputation lists. It is shown where measured and kept out of the score.
Each of these is a pool-level share, not a per-IP score, and each uses one database and one rule for every network, Shifter included. You can see where each network lands on the networks page.
How to use fraud scores when evaluating a provider
Fraud scores are still useful as one input among several. Use them like this:
- Look at the distribution, not one IP. Pull a few hundred exits from a trial and score them all. A handful of high scores is normal. A pool where most addresses score high, across vendors, deserves questions.
- Split by connection type first. If the vendor reports a usage type, count how many exits are hosting addresses. That count is more stable across vendors than the score itself. Check the ASN of anything that looks off.
- Compare like with like. Score two providers’ exits on the same day, from the same countries, with the same vendor. Comparing a score from last month with one from today mostly measures the vendor’s data refresh.
- Check the country. Confirm that exits sold as one country geolocate there, ideally in more than one database.
- Confirm on your targets. Whatever the scores say, run your own requests and count what succeeds.
Doing it yourself
A practical routine for any trial:
- Send a few hundred requests through the provider to an endpoint that echoes the IP, recording each distinct exit.
- Look up every exit in two or three IP intelligence services, keeping the usage type, country and score.
- Work out the share of hosting addresses and the share outside the requested country. These are your most reliable numbers.
- Look at how the scores spread, and whether the vendors agree on which addresses are risky.
- Run a sample of your real workload and compare what succeeded against what the scores predicted.
If step 5 disagrees with step 4, trust step 5. The score was built to help someone else make a different decision.