How Proxy Costs decides whether a price is knowable

Quotable answer

Figure: 710 plan records from 47 providers. The buyer-facing price states are 473 published, 98 conflicting, 110 not published, and 29 not verifiable.

As of: 2026-07-25T17:42:41.156Z, the generation time recorded by the current proxy-plans snapshot.

Source: /api/v1/latest.json. Each plan in that file carries its first-party evidence URLs, evidence retrieval times, raw source state, and buyer-facing price state.

What this site compares

The current scope is public residential, datacenter, ISP or static residential, and mobile proxy plans. Quote-only offers and rows whose public price cannot be read remain in the dataset. Scope and row counts come from /api/v1/latest.json, as of 2026-07-25T17:42:41.156Z.

For traffic-priced plans, the buying question is the cheapest published configuration that covers the requested traffic before it expires. The site then divides that checkout total by the traffic the buyer expects to use.

Plans sold per IP, by concurrency, as unlimited bandwidth, or through a configurator stay in those units. The current snapshot marks 136 of 710 plans comparable per GB and leaves the other 574 outside that comparison. Source: /api/v1/latest.json, as of 2026-07-25T17:42:41.156Z.

How a price is collected

Collection starts with the provider's own pricing page, pricing API, checkout surface, documentation, terms, or help material. A third-party review, a competitor table, a search snippet, and memory are not accepted as price evidence.

The raw source_state records how collection happened. vendor_api means a first-party structured endpoint. vendor_page_parsed means a public page or its embedded data was parsed. vendor_docs means the figure came from first-party documentation. vendor_page_manual means a readable public page was transcribed because it did not yield a stable structured record. Manual does not mean estimated.

Every accepted evidence record keeps the field observed, provider URL, retrieval time, and supporting text. The current 710 rows all retain at least one first-party URL and retrieval time. Source: /api/v1/latest.json, checked against the 2026-07-25T17:42:41.156Z snapshot.

The four price states

Published means a first-party source supplied enough numerical terms to record the price. It can come from an API, parsed page, provider document, or manual reading of a public provider page. Published does not automatically mean comparable per GB.

Conflicting means two or more first-party sources disagree about a price or a term that changes what the buyer gets. The current snapshot has 98 such plans. The row keeps every reported value and every source link, and it cannot set a cheapest-price claim while the contradiction remains unresolved. Source: /api/v1/latest.json, as of 2026-07-25T17:42:41.156Z.

Not published means the public provider material gives no reproducible numerical price. The current buyer-facing state contains 110 rows: 103 whose raw source_state is not_published and 7 quote-only rows whose public page could be read but supplied no fixed amount. The value is null, not zero, and the row stays visible with the provider link. Source: /api/v1/latest.json, as of 2026-07-25T17:42:41.156Z.

Not verifiable means a price may exist, but the public surface available to this collection could not establish it. In the current 29 rows, the page was blocked or the provider exposed only a configurator without a fixed plan to quote. Source: /api/v1/latest.json, as of 2026-07-25T17:42:41.156Z.

The rules behind every calculation

A checkout figure has one of three bases recorded by the pricing engine reviewed on 2026-07-26: the provider published the exact total, the total was computed from that tier's published rate and volume, or the total was computed from a published package plus a published overage rate. One tier's rate is never applied to another tier's volume.

When several configurations cover the request, the engine selects the least expensive published configuration, not merely the smallest tier above the request. The selected package size, unused traffic, checkout total, and calculation basis remain attached to the answer.

Effective cost per used GB is checkout cost divided by requested traffic, not included traffic. If a buyer asks for 40 GB and must buy 100 GB, the denominator remains 40 GB and the unused 60 GB stays visible.

Expiry windows remain separate plan records. Per-IP, unlimited-bandwidth, concurrency, configurator, and quote-only plans are never forced into a per-GB number. These rules are implemented in the pricing engine reviewed on 2026-07-26 and are reflected in /api/v1/latest.json.

Freshness and failed refreshes

The public snapshot has its own generated_at time, and each evidence record has its own retrieved_at time. They answer different questions: when the dataset was assembled and when that provider statement was observed.

The configured collection cadence is daily and the public freshness limit is 48 hours, reviewed in site configuration on 2026-07-26. /api/v1/health.json reports current, stale, or unavailable and names the dataset time used for that decision.

If a refresh fails, the refresh runner keeps the last accepted records in the date-stamped snapshot and marks them stale. It does not turn the current clock into a new observation time. This behavior is defined by the refresh runner reviewed on 2026-07-26.

What happens when the answer is unknown

An unknown value stays null and renders as not published or not verifiable. It is never converted to zero, unlimited, false, or the most convenient nearby number. A plan with an unknown required limit does not pass a filter for that limit.

A row is not removed because it is awkward. A missing price remains in the provider count, a blocked page remains named, and a conflict retains both values and both links. The incumbent comparison pattern is to keep the table looking complete. This site lets the table say that the answer is unknown.

The site runs no proxy performance tests. A third-party success rate can be shown only with the tester, source, and observation date, and it remains separate from the price. Free proxy lists are outside the dataset scope as of the 2026-07-25T17:42:41.156Z snapshot.

Corrections

Send the plan ID, disputed field, provider URL, published value, and the time you observed it to data@proxycosts.com. A correction is made against first-party evidence, regardless of who reports it.

A provider correction and a reader correction require the same record: a public provider source and the figure or term it supports. Private assertions can identify a question but cannot replace public evidence.

This methodology was materially reviewed on 2026-07-26. A rule change updates this date and the machine contract or pricing logic that implements the rule.