Fact Checking vs Source Intelligence

Fact checking and source intelligence are often treated as the same activity. They are not, and the difference determines what a system built around either one can honestly claim.

Fact checking produces a verdict

A fact check takes a claim, investigates it, and publishes a conclusion: true, false, misleading, unsupported. It is editorial work. It requires judgement, expertise and accountability, and when done well it is valuable.

It also has structural limits. A verdict is a single point of view issued at a moment in time. It compresses a body of evidence into a label, and the label travels further than the reasoning behind it. Readers who disagree with the verdict have no way to inspect the path to it, and readers who agree stop looking.

Source intelligence produces context

Source intelligence asks a different question. Not “is this true” but “what is the evidential situation around this claim”:

• who stated it, and when;
• what they were describing;
• which other sources reported the same thing independently;
• which sources contradict it;
• whether apparently separate reports trace back to one origin;
• what was known before it, and what changed after.

The output is not a label. It is an arrangement of the material such that the reader can see the shape of the evidence.

Why the distinction matters for software

A system that issues verdicts is making claims of its own, and it inherits every obligation that comes with them: it must be right, it must be accountable, and it must be defensible when contested. Automating that is a promise very few systems can keep.

A system that arranges context makes a smaller and more honest promise: that the material shown is real, that its provenance is accurate, and that the arrangement does not distort it. That promise is keepable, and it is genuinely useful.

This is why NWS.ONE describes itself as providing contextual signals and cross-source analysis rather than determining truth. The distinction is not modesty. It is a statement about what the system is actually able to be responsible for.

What context looks like in practice

Take a single claim appearing across coverage. Useful context includes:

• the earliest appearance — where it entered the record;
• the independent reappearances — sources that report it without deriving from the first;
• the derivative reappearances — sources that repeat the original, which look like corroboration and are not;
• the contradictions — sources reporting something incompatible;
• the surrounding chronology — what happened before and after that bears on it.

A reader who sees this arrangement can often reach a conclusion faster than a verdict would allow, and more importantly, can see why.

The two are complementary

None of this argues against fact checking. Some claims need adjudication by people willing to stand behind a conclusion, and no amount of arrangement substitutes for that.

The argument is about division of labour. Software is good at assembling and arranging evidence at a scale no person can match. People are good at judgement. A system that assembles well and refuses to adjudicate is playing to the strength it actually has.