Skip to main content

Where our facts come from

Sources & crawler policy

Every fact we publish is anchored to a registered source. This page explains how sources are classified and scored, the rules our crawler follows before it ever fetches a page, and how each fetch becomes an immutable snapshot in a verifiable evidence chain. For how those sources turn into published facts, see the methodology.

Source types

Sources are registered with a type and an authority score. The type records what kind of evidence it is; the score (below) records how much weight it carries. The counts shown are from our live register.

Source typeWhat it isRegistered
RegulatorNational telecom regulators and the ITU — the authority of record for network-code assignments and licensing.0
Provider officialA provider's own site, documentation, or terms — authoritative for that provider's plans and policies.191
Standards bodyIndustry standards organisations (e.g. GSMA) for definitions and technical baselines.59
App storeOfficial app-store listings, used for device/app facts and version details.0
Partner APIStructured feeds from partners who supply data under an agreement.0
Editorial testFacts derived from our own controlled testing, always labelled as test-derived.0
User reportReports from the public — the lowest authority, corroborated before anything is published.0
Third partyReputable third-party references, used to corroborate rather than as a sole basis.0

250 sources registered, 179 with a reviewed crawl policy currently permitting crawling.

Authority score

Each source carries an authority score between 0 and 1. It is kept deliberately separate from freshness: a regulator’s filing is highly authoritative even if it is a few months old, while a brand-new user report is fresh but low-authority. A regulator or standards body outranks a provider’s marketing page, which in turn outranks a third-party mention or a user report. When sources conflict, the higher-authority source prevails and the conflict is recorded rather than hidden.

Robots & terms review before crawling

A source cannot be crawled just because it exists in the register. Crawling is disabled by default and only enabled once a policy has been reviewed:

  • Robots and terms reviewed. We record when a source’s robots.txt and terms were reviewed. Crawling stays off until that review has happened and explicitly permits it.
  • Identified and polite. Our crawler identifies itself with a descriptive user agent, honours robots.txt, applies a per-source rate delay and a single-request concurrency cap, and enforces byte and time limits.
  • Scoped fetching.Only the path patterns a source’s policy allows are fetched, on the interval that policy sets.
  • Safe by construction. The fetcher blocks private, reserved, and metadata addresses, re-validates on every redirect, and pins each connection to the address it validated — so a source can never be used to reach internal infrastructure.

Snapshot & evidence chain

Every fetch that yields new content is stored as an immutable snapshot: the retrieved content, a content hash, and the time it was fetched. Identical content is de-duplicated by hash so we do not re-process unchanged pages, and each snapshot is preserved so the exact text a fact rests on can always be re-read.

A published fact then points at a specific evidence excerpt inside a specific snapshot, together with a locator, how it was extracted, and a confidence. That chain — source, snapshot, excerpt, verification, freshness — is what lets any fact on this site be traced back to the document it came from.

Corrections & takedowns

If a fact is wrong, out of date, or a source should not be crawled, we want to know. Providers can claim their brand to submit corrections through the review process, and anyone can report an issue for a fact via its page. Correction requests are reviewed on the same evidence-and-approval basis as any other change — nothing is altered without a source and an audit trail.