methodology

what we measure

proxy intelligence runs continuous measurement jobs against each provider's exit node pool. we track four primary signals: the number of unique ips seen in a rolling 30-day window, the share of ips that resolve to residential asns versus hosting or datacenter blocks, success rate of requests through each node, and how often the pool composition changes.

data collection

we purchase small plans from each provider under the same conditions any researcher would. all measurement traffic goes to endpoints we control, so no third-party sites are affected. each node is tested at most once per hour. we do not scrape, we do not abuse provider infrastructure, and we do not attempt to reverse-engineer internal routing.

transparency score

the transparency score (0–100) is a composite. higher is better. it rewards providers whose measured pool size is close to their advertised size, whose asn mix is genuinely residential, whose ips rotate predictably, and who publish clear documentation about their sourcing. it penalises mismatches between claims and measurements and a high share of datacenter or hosting-range ips presented as residential.

the exact formula is:

• pool accuracy (30 pts): measured unique ips vs. claimed pool size
• residential purity (30 pts): share of ips in genuine residential asns
• reliability (20 pts): 30-day success rate
• documentation (20 pts): sourcing transparency and opt-out mechanisms

asn classification

we use a combination of bgp routing data, asn registry metadata, and a manually curated list to classify each asn as residential, mobile, datacenter, hosting, or unknown. the classification is re-run weekly. we err on the side of caution: if an asn is ambiguous, it does not count toward the residential share.

ethics

we only use providers whose terms of service permit research use. we do not store the content of any responses routed through provider nodes. all our measurement endpoints return minimal responses. if a provider asks us to stop measuring them, we will.

the data we publish is aggregate and provider-level. we do not publish individual ip addresses, nor do we expose any personally-identifiable information.

limitations

our measurements represent the plan we purchased, not the provider's entire network. larger plans may behave differently. providers can and do change their infrastructure in response to measurement, which we treat as evidence of low transparency. all measurements have a margin of error; treat the scores as directionally correct, not precise.