Proxy tier benchmark › Targets

Do you need residential proxies for the Arkansas business entity search?

State business entity registries · minimum viable tier: residential · measured 2026-08-18

Yes. The Arkansas Secretary of State corporation search is the single target in this study of 32 where a proxy tier is the difference between nothing and everything. Unproxied and both datacenter configurations returned the expected page 0% of the time; adding a browser fingerprint raised that to 30%; a residential exit returned it on 10 of 10 requests (100%). Measured 2026-08-18.

What was measured
URL https://www.sos.arkansas.gov/corps/search_all.php
Content assertion Corporations — never verified against a live page
Arms 6
Rounds per arm 10
Requests 60
Measured 2026-08-18, 15:38–18:09 UTC
Benchmark host DigitalOcean droplet, New York (nyc3)

A request counts as a success only if the expected content is present. An HTTP 200 carrying a challenge page, an empty body or the wrong page is a failure, because it is not a page you can use.

This measures whether a tier works, not long-run stability under sustained load. Ten requests per arm, at a two-second floor, never approach a per-IP rate limit — which is the thing a rotating pool exists to defeat.

Results by arm

Arm Success p50 p95 Bytes per success Outcomes
direct (no proxy) 0% 360 ms 4,388 ms n/a blocked 10
datacenter, one pinned exit 0% 1,012 ms 7,201 ms n/a challenge 3, blocked 7
datacenter, rotating exit 0% 919 ms 3,154 ms n/a challenge 3, blocked 7
anti-detect, one pinned exit 30% 1,075 ms 5,337 ms 18,718 success 3, blocked 7
anti-detect, rotating exit 20% 1,013 ms 3,237 ms 18,718 success 2, blocked 8
residential, rotating exit 100% 3,395 ms 5,486 ms 18,718 success 10

What protects this target

Amazon CloudFront, refusing by network rather than by client signature.

Thirty-nine of 60 requests came back HTTP 403 with a 919-byte body whose text the probe captured as "ERROR: The request could not be satisfied. 403 ERROR The request could not be satisfied. Request blocked." — CloudFront's stock refusal. Every unproxied request drew it, seven of ten on each datacenter arm, and seven and eight of ten on the two anti-detect arms.

Six requests, all on plain datacenter arms, came back HTTP 405 Method Not Allowed with a 2,551-2,553 byte body matching a captcha signature. A GET that returns 405 is not a client bug; it is a WAF answering with a status that most scrapers will file as a coding error and stop investigating.

The residential arm was served the real page — 18,718 bytes, ten times out of ten, in a median of 3,395 ms. Anti-detect over datacenter exits reached it 3 and 2 times out of 10, which is the fingerprint helping and not being enough. The decisive variable here is the exit network.

robots.txt status

The URL measured is allowed by the host's robots.txt.
Status allowed
File https://www.sos.arkansas.gov/robots.txt
Crawl-delay 5 s — obeyed by the harness
Fetched as the benchmark's own User-Agent
Note robots.txt 821 bytes via benchmark_ua

Practical setup

This target needs a residential exit and rotation is not the issue — the residential arm rotated per request and scored 100%, and the datacenter arms scored 0% pinned and 0% rotating alike. Select the residential pool (see residential) with US exits, and respect the site's declared Crawl-delay: 5 — Arkansas is the only target in this study that publishes one, and the harness obeyed it.

At 18,718 bytes per successful page, one thousand pages move about 0.019 GB, which across the published residential pack ladder ($9.00/GB at 5 GB down to $5.00/GB at 250 GB) is $0.09-$0.17 per 1,000 pages.

Limits of this result

  • The search page was measured, not a submitted entity query.
  • Ten rounds per arm. The residential arm's 100% is 10 of 10, which is consistent with a rule that admits residential exits but does not establish stability over a sustained collection run.
  • Ten rounds per arm in a single three-hour window on 2026-08-18. Anti-bot behaviour drifts; a result older than a few months is a historical record, not a current one.
  • One exit geography. Every request left from a US datacenter host or a US exit, so nothing here describes how this target answers traffic from elsewhere.
  • The unproxied arm ran from a clean, previously unused cloud IP with no accumulated reputation. It is a best case a production scraper will not sustain.

Run date: 2026-08-18 · Full benchmark report · Harness and raw data · All 32 targets

The 60 rows behind this page are in results/six-arm/raw.csv; filter them with grep '^[^,]*,us_ar_sos_search,' raw.csv.

Test your own target

The result above describes one URL on one day. Paste your own URL into the live tester and see the real response before creating an account.

Test your target See plans