How to evaluate a proxy provider
A repeatable test plan for latency, success rates and regional coverage.
Read the engineering noteSCRAPETECH / ENGINEERING
Explore the sample formats, work through a measurable evaluation, and check your own recorded results. Everything here is available without an account.
RECORDED MEASUREMENTS → CLEARER COMPARISON
Paste your recorded results to compare success rates and latency by provider and region. This tool analyzes the CSV; it does not connect to proxies or measure your connection.
Required: provider, region, status, latency_ms. Use ok / failed / timeout or HTTP status codes from 200 to 599. Successful results need a positive latency. Maximum 5,000 rows / 1 MB.
The preloaded measurements are synthetic. Mean, P50 and P95 use successful requests only; failed requests remain in the success-rate denominator. Percentiles use the nearest-rank method. A blank latency means there were no successful requests.
METHODS & CHECKLISTS
Three practical guides, published 17 October 2026.
A repeatable test plan for latency, success rates and regional coverage.
Read the engineering noteHow to define fields, handle missing values and check duplicate records.
Read the engineering noteA practical checklist for formats, identifiers, refreshes and acceptance checks.
Read the engineering noteDOWNLOADABLE REFERENCE FILES
These small datasets are synthetic reference material. Example domains are placeholders; the files do not represent client records or a live collection.
Five records with an intentional duplicate. Fields: sku, title, price, currency, in_stock and source.
Three example organic results with query, position, title, URL, locale and result type.
Eight illustrative requests across two fictional provider labels. Includes a failed request to show how the denominator works.
Bring a sample URL and the fields your workflow actually needs.