How scoring works

We take a debate transcript, pull out every factual claim, and check each one against the web — on two separate axes.

Axis 1

Veracity — is the claim true?

Every factual claim is researched against the web and scored 0–100, with the sources it relied on attached. That one score is read two ways — absolutely (true exactly as worded) and substantively (true in its gist). Two readings of this axis, not two axes.

92% 64% 12%

Axis 2

Omissions — what was left out?

A claim can be entirely true and still mislead, by leaving out the one fact that changes what it means. So claims that check out get a second, separate look, and each finding names a missing fact and cites a source. There is no score on this axis at all.

⚠ 2 omissions Context checked

The two never mix — an omission is never subtracted from a score. Everywhere on this site, purple means omissions: a purple badge, rule or heading is always about what a claim left out, never about how accurate it was.

1. Every claim gets a score out of 100

0 is clearly false. 100 is clearly true. The colour tells you the band at a glance.

0–19
False

20–49
Mostly false

50–79
Mixed

80–100
True

Every score links the sources it came from, so you can check the working yourself.

2. Two ways to read that score, because “wrong” means two things

Take the claim “Our plan created exactly three million jobs”, when the real figure is 2.6 million.

Literally 30% — “exactly three million” is not what happened.
In substance 85% — millions of jobs were added, so the point stands.

3. A speaker's score is the average of their claims

100%25%50%33%= 52%

Opinions don't count. Nor do claims nobody can check either way — they're shown, but they neither help nor hurt.

4. Claims are judged by the day they were said

Facts change. A claim that was true in 2015 is scored as true, not marked wrong because the world has moved on since. Every verdict shows the date it was judged against.

5. ⚠ The second axis: what was left out?

A claim can be completely true and still mislead you, by leaving out the one fact that changes what it means. So true claims get a second, separate check.

“Exports have grown 40% since 2019.” True — and here is what you weren't told:

Missing baseline

2019 was the worst export year in a decade. Measured from 2018, the same figure is a 3% rise.

100% ⚠ 1 omission

That is the pairing to look out for: findings appear beside the score, never inside it. The claim is still 100% accurate, and its score still says so.

There is no “integrity score”. Whether something is misleading is a judgement you can reasonably disagree about, and a number would pretend otherwise. You get the specific missing fact and a source, and you decide.

6. ⚠ One kind of omission we don't count against anyone

In a debate, arguing one side is the job. Not making your opponent's argument for them is not dishonesty, so it isn't flagged. What is flagged is leaving out something that changes what your own claim means. In an interview or a speech there is no opponent to make the missing case, so that allowance doesn't apply.

Back to the full instructions → — the same system with the rules spelled out, worked examples, and a sample debate you can click through.