Reliability
Scrapers need requests to complete before speed, targeting, or browser behavior matters.
Connection100HTTP reachability100Success rate55.1%
measuredA score appears only when every dimension with a positive weight has comparable evidence. Saved on this device.
Pool 99.7 · Workload 28.5 · trust and session clean.
The report has enough execution coverage for publication review.
Single declared worker vantage region: us-east. Declared concurrency tiers: 6-20.
Missing metrics are listed explicitly instead of implied.
2 providers · ranked by proxy index score · click a row to open the full record
Strong baseline proxy health for scraping transport
The proxy selection performed well on canonical connection, identity, HTTP reachability, geo, and ASN checks.
Leads the cohort on geo-accuracy (99.6%) and ASN grade A, 0.2 points clear of Bart ISP.
Open NSTProxy's page →Six buyer dimensions, each built only from measurements with real spread — not the commodity plumbing checks. Grades are 0–100; a dash means we do not yet score that dimension for this line.
Mean of equal-weighted workload-fixture success and browser-engine success (when present), scaled to 0-100. Baseline plumbing checks do not raise this grade.
E-commerce product fixture success rate 0.39 ratio · JSON workload fixture success rate 0.39 ratio · Large HTML fixture success rate 0.38 ratio · Listing page fixture success rate 0.39 ratio · SERP fixture success rate 0.39 ratio · Browser engine success rate 0.18 ratio
P95 latency banded 100 at <=200ms falling linearly to 0 at >=2000ms, averaged with throughput MB/s graded as min(100, throughput_mbps * 10) when present.
P95 latency 912.55 ms · Throughput 111.11 Mbps
Static ISP pool grade: mean of endpoint-to-exit coverage, repeat stability, exit distribution balance, and exit-IP entropy. Stable reuse is expected for a static endpoint and is not penalized as failed rotation.
Unique exit IP count 10 count · Repeat rate 0.99 ratio · Top exit IP share 0.10 ratio · Exit IP entropy 1 score
Geometric mean of enrichment-derived trust signals scaled to 0-100: usage-type accuracy (claimed vs observed), the inverse of high-risk IP share, and country geo accuracy. A zero load-bearing signal cannot be averaged away. Enrichment inputs require at least five unique exits, at least 80% coverage, and one labelled vendor source.
Usage type accuracy 1 ratio · High-risk IP share 0 ratio · Country geo accuracy 1 ratio
Sticky-session hold rate scaled to 0-100. Requires session-persistence attempts.
Session hold rate 1 ratio
Not graded — No normalized benchmark pricing snapshot was available. Provider-stated prices may still appear as unscored review context.
The measurements behind the dimension grades above — what we ran and what it returned.
Scrapers need requests to complete before speed, targeting, or browser behavior matters.
Latency and transfer speed determine queue depth, worker cost, and retry pressure.
Rotating proxy buyers need a pool that is broad enough and not concentrated on a few exits.
HTTP 200 is not enough; buyers need to know whether representative pages return usable content.
Browser-based scraping depends on proxies surviving the same engine stack buyers use.
Regional and network targeting breaks when provider metadata is stale or incorrect.
27 of 36 tracked buyer metrics are measured in this public payload.
How the line handled representative scraping targets — direct evidence behind the workload success dimension.
6 public trend series are available for this report.
The public report exposes the eligibility gate, buyer dimensions, metric definitions, sample design, target-class rollups, and redaction policy. It does not expose credentials, endpoint hosts, raw observations, or raw attempt errors.
Redacted fields: endpoint.id, endpoint.host, endpoint.port, endpoint.proxySourceId, endpoint.tags, proxy credentials, raw observations, raw attempt errors
This issue is one window on the field. Open a provider's standing record for the full history, or read the methodology behind every number.