How scoring works

In a debate, arguing one side is the format. A speaker is not obliged to make their opponent's case, and treating that as an omission would flag every statement in every debate and make the whole thing meaningless. So each candidate omission has to pass one test: does it change what the speaker's own claim means?

That allowance belongs to a debate, because an opponent is in the room to make the missing case. In an interview or a speech nobody is, so it is withdrawn — a fact left out there is simply one the audience never hears. The motion and each speaker's side are recorded either way, and are optional either way: they are what lets us say which direction an omission favours.

Motion This house believes remote work harms productivity.
An opponent of the motion says: “Company X reported a 15% productivity increase after going remote.” The figure checks out — it is true.

1. It has to matter. Would knowing it change a reasonable listener's conclusion about this claim? Every true statement omits something; almost none of it matters.
2. It has to be cited. Every omission names a specific fact and links a source that was actually found when researching it. Anything the app cannot trace to a real page is discarded rather than shown — an uncited accusation is worse than no finding at all.
3. It has to undercut the speaker's own claim, not merely support the other side, as above.

Every finding above passes one test: the missing fact changes what the claim means.

Where we know which side the speaker was arguing, we ask one more thing — would that fact have damaged their own case? In the Company X example above it would: the missing detail deflates the very figure the speaker was leaning on.

Those are the ones a speaker's profile counts. One on its own proves nothing — omit enough facts by accident and some will land against you by chance. So the profile reports what share of someone's omissions were the damaging kind, and reports nothing at all until there are enough of them to be worth reading.

A speaker who says nothing false, but whose omissions keep turning out to be the damaging ones, is what this app calls Accurate · One-sided. That is a pattern, not a motive — whether it is selection or coincidence is for you to decide.

The two answers read together

Side by side, the accuracy score and the context findings sort speakers into three groups. A profile names the group its speaker falls into, and Speakers ranks each:

• Accurate · Forthright — Accurate, and checked for what they left out
• Accurate · One-sided — Say nothing false, and what they leave out always damages their own case
• Inaccurate · Incomplete — Inaccurate even read generously, and leaving things out as well

Most speakers are in none of them. Each group has a bar to clear — enough claims examined, and a clear enough result — and Speakers states what that bar is. A name missing from a list is not a verdict about that person.

Every finding is filed under how the missing fact did its work. You will meet these labels on a claim, on a speaker's profile and on the home page — always in the omissions purple, never on the green-to-red veracity ramp, because these are not ranked. Cherry-picking is not worse than a missing baseline; they are different, not ordered.

• Cherry-picking The favourable data points presented as the picture.
• Missing baseline A change quoted with no baseline, so its size cannot be judged.
• Selective timeframe A window chosen because it flatters — the trend reverses just outside it.
• Subgroup selection True of a narrow subgroup, stated as though true generally.
• Count vs rate A raw count where a rate is the meaningful measure, or the reverse.
• Quote mining A quotation stripped of the passage that gave it its meaning.
• Missing context Context missing in some other way.