Floppydata Floppydata

Residential Proxies Least Flagged as Proxies: Proxy Detection Rate Tested

Rayobyte and Floppydata had the fewest IPs flagged as proxies in our benchmark: 59.3% and 60.7% of a 300-IP sample. Their 95% confidence intervals overlap, so they share first place. The next provider, NodeMaven, had 77.7%.

We operate Floppydata and tested it with the same scripts, servers and schedule as the other 11 providers. The full data and scripts are public.

See the ranking
12 providers
300 IPs each
Scamalytics
5–8 Oct 2026
Residential Benchmark

The ranking

Residential Proxies Least Flagged as Proxies: Proxy Detection Rate Tested: all 12 providers
# Provider Proxy Detection Rate lower is better
1 = Rayobyte 59.3% 95% CI 53.7–64.7
1 = Floppydata 60.7% 95% CI 55.0–66.0
3 NodeMaven 77.7% 95% CI 72.6–82.0
4 Oxylabs 80.3% 95% CI 75.5–84.4
5 = ProxyEmpire 83.3% 95% CI 78.7–87.1
5 = Webshare 83.3% 95% CI 78.7–87.1
7 Decodo 85.0% 95% CI 80.5–88.6
8 SOAX 85.3% 95% CI 80.9–88.9
9 DataImpulse 87.0% 95% CI 82.7–90.3
10 = Evomi 96.0% 95% CI 93.1–97.7
10 = IPRoyal 96.0% 95% CI 93.1–97.7
12 Geonode 99.3% 95% CI 97.6–99.8

Shared first place (95% confidence intervals overlap) Floppydata (we operate it) = tied place

Proxy Detection Rate is the share of 300 sampled IPs per provider that Scamalytics flags as a proxy of any kind. A shared first place means the 95% confidence intervals overlap.

Check the numbers yourself: the anonymized raw data, scripts and full results workbook are on GitHub.

Key findings

How many of a provider's IPs were already known as proxies ranged from 59.3% to 99.3%.

  • Rayobyte and Floppydata share first place

    Rayobyte 59.3% and Floppydata 60.7%. The difference between them is within the 95% confidence intervals.

  • Then a wide gap

    The next provider, NodeMaven, had 77.7% of its IPs flagged: a gap of 17 percentage points.

  • Geonode is flagged most

    99.3% of its sampled IPs were known to Scamalytics as proxies.

  • Flagged is not blocked

    The rate says how many IPs one database knows as proxies. Whether a site blocks them depends on the site and its own signals.

How we measured

How we measured proxy detection

We checked a sample of 300 unique IPs per provider (3,600 in total), taken at random from the exit IPs of the infrastructure test on 5–8 October 2026.

  • Proxy Detection Rate. Each IP went to the Scamalytics API. An IP counts as detected when Scamalytics flags it as a proxy of any kind: a residential proxy, a datacenter, a VPN or a cloud network. The rate is the share of flagged IPs.
  • Who shares first place. Each rate has a Wilson 95% confidence interval. Providers whose interval overlaps the leader's share first place.

Limits

  • One database. The rate comes from Scamalytics. Another database would flag the same IPs differently.
  • A sample, at one time. 300 IPs per provider from 5–8 October 2026. Pools change; the same provider can score differently a month later.
  • Only aggregates are published. Individual IPs and raw Scamalytics responses are not published; the per-provider results are.
  • Raw data and scripts. The anonymized request logs, the test scripts and the results workbook are on GitHub, so every number here can be recomputed.

More from the benchmark

Download raw benchmark data

Anonymized request logs, the test scripts and the full results workbook. Check every number on this page yourself.

Frequently Asked Questions

Which residential proxies are least detected as proxies?
In our benchmark, Rayobyte (59.3%) and Floppydata (60.7%), tied within the 95% confidence intervals. Geonode had the most IPs flagged (99.3%).
What does Proxy Detection Rate mean?
The share of a provider's IPs that the Scamalytics database flags as a proxy of any kind. Lower means fewer of its IPs are already known as proxies.
Why do two providers share first place?
Rayobyte and Floppydata differ by 1.4 percentage points, less than the uncertainty of a 300-IP sample: their 95% confidence intervals overlap.
Does a low detection rate mean fewer blocks?
It helps, but sites use their own signals as well. A database flag is one input, not the decision.