Roam Residential Proxy Benchmark, July 2026: 10,000-request raw latency and exit-IP dataset

Raw per-request measurements from a 10,000-request benchmark of the Roam residential proxy network, run in July 2026. Published in full because per-request data is rarely released in this industry: vendor benchmarks are usually reported only as summary statistics. Method. 10,000 sequential HTTPS GET requests were issued through rotating residential exit nodes across five countries using plain curl, one request per proxy session, so that each row represents an independent exit IP draw. Timing comes from curl's own time_total and time_starttransfer counters. The full methodology is documented and reproducible at https://roamproxy.com/residential-proxy-benchmark and the dataset is released under CC BY 4.0. Columns. country (requested exit country), seq (request index within country), http_code, time_total_s (total wall-clock seconds), ttfb_s (time to first byte), exit_ip_anonymized, exit_geo (country the exit IP actually resolved to). Headline results. 99.67% success rate; 7,436 unique exit IPs across 10,000 requests; 98.9-100% geo accuracy against the requested country; median latency 2.0-2.7s depending on region. Privacy. Exit IP addresses are truncated (final IPv4 octet or trailing IPv6 groups replaced with xxx) so that individual residential hosts cannot be identified while the distinctness of exits remains measurable. No request payloads, destination hosts, or end-user data are included: the requests targeted a single IP-echo endpoint. Limitations. This is a single-vendor, single-vantage-point measurement of our own network and should be read as such; it is not a comparative study. Rows where the exit geo disagrees with the requested country are retained rather than dropped.