Methodology

Sources and evidence

Every price, plan limit and feature is taken from the provider's own public pages or documentation. Each fact records the page it came from and the date it was last checked; product pages list these sources next to the numbers. We only read publicly accessible pages, with a collector that names itself and a contact address, obeys robots.txt and never logs in or bypasses access controls; ScrapeSelectBot describes it. A product's prices are as current as the oldest source they cite.

Freshness

• Fresh — the source was checked within the last 24 hours.
• Recent — checked within the last 7 days.
• Stale — older than 7 days, or never verified. Stale data is flagged wherever it is shown, and stale products are left out of recommendations.

Change detection and review

A collector fetches every official source page daily and stores a snapshot. Pricing pages are parsed into plans, prices and limits; differences from our catalog become pending changes. On documentation pages it checks that the text we quote is still there; when a quote disappears, the facts it supports stop being re-verified until a person looks. Nothing is published automatically: a person compares each change with the live page, and only confirmed changes update the catalog and appear on the change log. A page that can't be fetched or parsed three days in a row raises an alert, and we investigate instead of guessing.

Cost estimates

Providers bill differently — credits, requests, bandwidth, browser time or dollar budgets. We translate your workload (successful pages per month, page size, JavaScript share, anti-bot difficulty, countries) into what each provider actually bills, then find the cheapest plan that covers it, including overage and top-ups where documented.

Unknowns such as success rate and how many pages hit protected sites are handled as three scenarios — low, expected and high — so every estimate is a range. The assumptions behind each scenario are shown with the result. Estimates are not invoices; the provider's checkout is authoritative, and taxes and negotiated discounts are not included.

Fit Score

Recommendations first exclude products that cannot do the job: a required capability is not documented, no published plan fits the workload, even the low estimate is more than 50% over budget, or the pricing is stale. The rest are scored out of 100:

• Features — 30: required capabilities documented; deductions for missing proxies, self-managed anti-bot work and a team's maintenance capacity.
• Budget fit — 25: within budget scores 15 plus up to 10 for being close to the cheapest option; over budget scores progressively less.
• Capacity — 15: published concurrency against your peak.
• Geography — 10: country targeting where you need it.
• Data freshness — 10.
• Stability — 5: currently the same for every product, so it does not change the order.
• Commercial terms — 5: free trial, monthly billing, pay as you go.

Every point comes with a reason you can check. Components without data score the same for everyone and are labelled that way.

What never affects rankings

Affiliate commissions, sponsorships and whether a provider has a relationship with us are not inputs to any score. See the disclosure.