Bias methodology

What "bias" means on this site

We use "bias" in a specific, narrow sense: the patterned editorial choices a source or perspective makes when telling the same underlying story. That includes:

• Which actors are named first, last, or not at all.
• Which actions get described with active verbs vs. passive verbs.
• Which casualty figures are foregrounded.
• Which context is included and which is left out.
• The certainty level of the headline.
• The choice of value-loaded vocabulary ("terrorist", "freedom fighter", "civilian casualty", "collateral damage").

Bias in this sense is real and observable. It does not mean "the source is lying" — most credible newsrooms do not lie. It means they emphasise.

Source bias vs. article bias vs. framing bias

• Source bias. The structural orientation of an outlet — funding, country of origin, audience, editorial line. A source has a bias profile that persists across stories.
• Article bias. The framing of a single article. An outlet with a measurable institutional bias can still produce a well-framed article on a given day, and vice versa. We tag the source, we read the article.
• Framing bias. The pattern across many articles on a topic. This is what the perspective comparisons on the site are trying to surface.

The six perspectives

For each canonical event we generate framings from the following perspectives:

• Neutral. A fact-first reconstruction. The base layer the others modify.
• Western.US/NATO-aligned framing — security alliances, deterrence logic, "our allies", "our forces".
• Iranian. Tehran-aligned framing — sovereignty, resistance, external interference, sanctions burden.
• Israeli. Tel Aviv-aligned framing — existential threat language, proxy network, regional destabilisation.
• Global South. Non-aligned framing — economic spillover, food and fuel prices, displacement, regional autonomy.
• Pro-peace. Diplomatic/humanitarian framing — civilian cost, de-escalation options, the gap between the rhetoric and a settlement.

These are generic framings — characterisations of how a perspective tends to present this type of event, not what a specific outlet wrote today. The framings are clearly labelled wherever they appear.

State interest, editorial stance, headline framing

A useful diagnostic when reading any conflict story is to separate three layers:

1. State interest. Whose interests would be served by the story being told this way?
2. Editorial stance.What does this outlet's funding, audience, and history tell us about its likely framing?
3. Headline framing. What does the specific wording on the page do? Which actor is the subject? Whose action is the verb?

Loaded language

Specific word choices that signal a perspective:

• Naming choices."Militants" vs "combatants" vs "terrorists" vs "fighters" — different connotations, often describing the same group.
• Casualty wording."Killed" vs "died" vs "eliminated" vs "martyred" — encodes responsibility.
• Action verbs."Struck" vs "attacked" vs "retaliated against" — encodes legitimacy.
• Modifiers."Alleged", "reported", "confirmed" — encodes certainty.

Omission risk

The most common form of framing bias is not what is said but what is left out:

• Casualty figures from the other side.
• Prior provocations from the actor being defended.
• Civilian impact when reporting a military operation.
• Diplomatic options when reporting a military action.
• Historical context that complicates the preferred narrative.

The comparison view tries to make omissions visible by showing what each perspective tends to leave out, not just what it says.

Attribution quality

A high-quality report attributes every contested claim to a named source. A low-quality report uses constructions like "sources say", "it is understood that", or anonymous "officials" without specifying which government's officials.

Limitations of the framing analysis

• The six perspectives are generalisations. Real outlets are more varied than "Western" or "Iranian" framings imply.
• The framings are generated by an LLM from the canonical event record. They are interpretive analysis, not independent verification. See the AI use policy.
• We do not currently score individual articles for bias. The framing analysis is event-level.
• Bias detection is not the same as fact-checking. An article can be unbiased in framing and still factually wrong, and vice versa.

Examples

The compare page shows worked examples of how the same base event reads under each framing. The comparison is for reader literacy — it is not an accusation against any specific outlet.