You set your gateway to rotate on every request, run a job, and look at the log. The same residential address appears three times in an hour. The next day, many of yesterday’s addresses are back. Is rotation broken, or is that how pools work?
Some of it is how pools work, and some of it is a signal worth reading. If you rely on a rotating residential proxy network, knowing which is which tells you a lot about the pool behind the gateway. This guide explains why addresses come back, the two figures Atlas uses to measure it, and how to check a provider yourself.
Why repeats happen at all
A rotating proxy gives you a different exit per request or per interval, but it draws from a finite set of addresses that are online in your country at that moment. Random draws from a finite set repeat sooner than intuition suggests.
An illustrative example: say a thousand addresses are online in your target country and the gateway picks one at random per request. After only about forty requests, the odds are better than even that you have seen some address twice. This is the same arithmetic as the birthday problem, and it means a repeat on its own proves nothing.
Gateways also do not pick at random. They tend to favour addresses that are fast and have stayed online, which concentrates traffic on a stable core. Narrow targeting, by city for example, shrinks the set further.
Repeats within a day vs across days
Two kinds of repeat look similar in a log but mean different things.
Within a session or a day, repeats mostly reflect how many addresses are online in your slice and how hard you are pushing it. Send more requests per hour than the slice can absorb and repeats rise. That is capacity, and it varies by country and time of day.
Across days, repeats say something about the pool itself. Residential addresses change naturally: devices go offline, households restart routers, internet providers reassign addresses. A large, genuine residential pool therefore looks noticeably different from one day to the next. If the same set of addresses greets you every morning, the pool is smaller than it looks, or the gateway leans heavily on a fixed group of always-on exits.
That second pattern matters because targets keep reputation per address. Addresses that hit the same site daily build a history there, and a history is what gets an address rate-limited or blocked.
What pool churn measures
Pool churn is the share of exits in a window that were never seen in any previous window. It answers the question: how much of what I got today is genuinely new?
- High churn means fresh addresses keep arriving. For rotating workloads, that is usually what you want.
- Low churn means the pool is recycling what you have already seen.
Churn needs history to mean anything. In the first window ever measured, every address is new by definition, so the figure only becomes informative once a network has been tracked for a while.
Churn also has a cost for long sessions. Addresses that turn over quickly are addresses that disappear mid-session. If your work needs one address for minutes at a time, look at how long sticky sessions actually hold, a separate figure, rather than reading high churn as purely good.
What repeat IPs measure
Repeat IPs is the share of distinct exits in a 7-day or 30-day window that appeared on two or more days. It is not shown for a 24-hour window, which cannot contain a multi-day repeat.
A rotating residential pool should keep this low. A high figure is inventory that behaves like static addresses: the same exits, day after day, whatever the product page says about rotation.
For background on why residential pools turn over in the first place, see how providers build and refresh residential IP pools.
Reading the two together
Neither figure is enough alone. Read together, they describe the shape of a pool:
- High churn, low repeats. A large, living pool. Most of what you get is new, and little of it sticks around for days.
- Low churn, high repeats. A small, recycled pool. You keep meeting the same addresses, and few new ones arrive.
- High churn, high repeats. A stable core with a churning fringe. The gateway sends much of your traffic to a fixed group while new addresses cycle through the edges.
A third figure helps: requests per exit IP, the number of consecutive requests one address served before the gateway moved you on. It separates a product that genuinely rotates from one that quietly behaves like a sticky session, whatever the label says. The sticky vs rotating residential proxies post covers when you want each.
How Atlas measures it
Atlas records the exit address each target saw on every request, in five-minute buckets, for every network on the same schedule. Session and rotation settings stay at each provider’s defaults, so the figures describe the product as sold.
From that stream it computes churn against all earlier windows and repeats over 7 and 30 days. Because every network gets the same requests at the same times, a difference in churn or repeats reflects the pool, not the test. The exact definitions of churn and repeat IPs are in the methodology, and the unique IP counts for every tested network give the scale each figure applies to.
Doing it yourself
You can measure both figures for any provider with nothing but a log:
- Send rotating requests to an endpoint that echoes back the address it saw. Record the address and a timestamp for every request.
- Keep the rate and the country fixed, and run at the same hours each day.
- At the end of each day, build that day’s set of distinct addresses.
- Churn: count how many of today’s addresses appear in no earlier day’s set, and divide by today’s total.
- Repeats: after a week, count the addresses that appear on two or more days, and divide by all distinct addresses in the week.
- Run two providers side by side in the same window if you want to compare them.
A week of logs from your own workload is more convincing than any rotation claim, and it shows exactly how the pool will behave on the sites you care about.