Sources & Updates

Signal sources for news/info sites

Media Bias/Fact Check (MBFC)

Media Bias/Fact Check (MBFC) is an independent media-watchdog project that rates news and information websites on political bias and factual accuracy. WikiSignals imports MBFC’s publicly published ratings dataset and uses it as one input into a domain’s credibility profile.

What we import

For each domain in MBFC’s public dataset, we store:

• Factual reporting rating — MBFC’s six-level scale: Very Low, Low, Mixed, Mostly Factual, High, Very High.
• Credibility rating — MBFC’s three-level scale: Low, Medium, High.
• Bias rating — MBFC’s left–right political-bias categories (plus Pro-Science, Conspiracy/Pseudoscience, Satire, Fake News).
• Link to the MBFC profile page for the domain, so readers can see MBFC’s full write-up.

We only accept a profile link that actually points to mediabiasfactcheck.com; anything else in the source data is treated as a data-quality discrepancy and is not displayed.

MBFC credibility score

We convert MBFC’s Factual Reporting and Credibility ratings into a single 0–1 estimate, weighting factual reporting more heavily than credibility, since it reflects a more direct assessment of accuracy:

MBFC score = (Factual-Reporting score × 0.7) + (Credibility score × 0.3)

Factual Reporting	Credibility: Low	Credibility: Medium	Credibility: High
Very Low	0.00	0.15	0.30
Low	0.14	0.29	0.44
Mixed	0.28	0.43	0.58
Mostly Factual	0.42	0.57	0.72
High	0.56	0.71	0.86
Very High	0.70	0.85	1.00

Update cadence

MBFC ratings are refreshed monthly against MBFC’s published dataset. New domains, rating changes, and any data-quality issues from each monthly run are listed on our Changelog page.

Limitations

• MBFC does not rate every website in our database; a domain with no current MBFC listing simply has no MBFC score.
• A small number of old bias categories (Extreme Left, Extreme Right, Least Biased) are no longer used by MBFC or us.
• A handful of entries in MBFC’s source data point to a specific article or section path (e.g., “facebook.com/handle”) rather than a top-level domain; these are skipped and logged rather than imported as a publication.

---

Wayback Machine (Internet Archive) Snapshot Data

For each domain, we record how extensively the Internet Archive’s Wayback Machine has archived it: the date of its first known snapshot, the date of its most recent snapshot, and the total number of snapshots on record.

Where the data comes from

We query the Internet Archive’s public “sparkline” endpoint — the same data source that powers the archive-history badge in the official Wayback Machine browser extension. It is a different, general-purpose archival-history lookup, distinct from InternetArchiveBot’s citation-repair tools (which track dead links used specifically as Wikipedia references, not a domain’s overall archive history).

What we store

• First snapshot date — recorded once and never overwritten afterward, even if a later check is incomplete or returns an unusual result, so this date reflects the earliest capture we’ve ever confirmed.
• Most recent snapshot date — refreshed every update cycle.
• Total snapshot count — refreshed every update cycle.

Update cadence and a few things worth knowing

• This data refreshes monthly, but not all ~75,000 domains are checked in a single pass. Archive.org’s response time per domain varies considerably — small sites typically respond quickly, while heavily-archived domains with long histories can take noticeably longer — so the monthly refresh runs in small automated batches over the course of the month rather than all at once. A given domain’s figures may therefore be a few weeks newer or older than another’s at any moment, even though both are refreshed roughly monthly overall.
• A new total that comes back noticeably lower than what we had on record is treated as an unreliable reading and is not saved — we keep the last good value instead. Archive counts should only grow over time, so an unexplained large drop almost always indicates a temporary or partial response rather than a real change.

---

Wikipedia (English) External-Link Counts

For each domain, we record how many times it is cited as an external link within English Wikipedia articles (citations, references, and external-links sections specifically — not Talk pages, user pages, or other non-article namespaces).

Where the data comes from

We use two complementary sources:

• Quarry, a Wikimedia tool that runs SQL queries against a replica of Wikipedia’s own database. A monthly Quarry query pre-computes external-link counts for the most-cited domains and subdomains (see how Wikipedia indexes external links for the technical basis). This covers the large majority of domains quickly, at no live-query cost.
• The live Wikipedia exturlusage API (the same mechanism behind Wikipedia’s own Special:LinkSearch tool; background: Help:Linksearch, Wikipedia:External links). Quarry’s pre-computed data only reports domains above a minimum citation count, and only this live query can confirm a domain has zero English-Wikipedia citations, so we use it for every domain Quarry doesn’t cover, and as a check on any Quarry number that looks suspiciously low compared to what we had on record.

A technical note on accuracy

Quarry’s pre-computed data is organized by domain in a specific way: ordinary domains (e.g. nytimes.com) are tracked as a single unit, while domains under a shared two-part suffix (e.g. bbc.co.uk, under the shared .co.uk suffix) are tracked individually using a more specific lookup, so that a site like bbc.co.uk gets its own accurate count rather than being lumped in with every other .co.uk site.

Update cadence and a few things worth knowing

• Like the Wayback Machine data above, this refreshes monthly in automated batches rather than all at once, for the same reason: domains that fall back to the live API vary in how long they take to fully count.
• External-link counts can legitimately rise or fall over time, since Wikipedia articles are continuously edited and citations are sometimes removed, replaced, or consolidated. A moderate change from one month to the next is normal, not an error.
• For the relatively small number of extremely heavily-cited domains, the live API count may be based on a bounded number of queries rather than an exhaustive count; in those rare cases the figure shown is a reliable lower bound rather than a guaranteed exact tally.

---

WikiSignals docs (written with the assistance of Claude Science):

• Design System
• Domain Data
• Sources & Updates