Pundits don’t keep score.
Avow does.

What is an Avow score?

It's one number, from −100 to +100, showing how accurate someone's public predictions have been. Zero means they always hedged and never really committed. A high positive score means they made bold calls and kept getting them right. A negative score is a warning sign: their confident calls tended to be wrong, so you'd have done better doing the opposite of what they said.

Where do the predictions come from?

From the pundits' own public words: TV and podcast transcripts, columns, posts, and books. We save every prediction with the exact quote, the date, and a link to a stored copy of the original. Each person also gets a public list of everywhere we looked, so you can see what's covered and what isn't.

How do you decide whether a prediction came true?

First we write down what would count as right or wrong, before we look at the result, so the standard is fixed in advance. Then three AI judges grade it against what actually happened, and we take the middle score. The judges are all Claude Opus, run three separate times with no shared memory, so they can't influence each other. Deadlines are strict: coming true too late still counts as a miss.

How is the number calculated?

For each call we weigh how sure the person was against what actually happened. Being confident and right earns the most, being confident and wrong costs the most, and hedging does a little of both. Bigger, higher-stakes claims count for more, and we hold new names near 0 until they've built a real track record, so a few lucky calls can't inflate a score. The AI judges grade each call, then one fixed formula, applied the same way to everyone, turns those grades into the score, and we show how much each call moved the total.

Isn't this cherry-picked?

This is the risk we worry about most. We don't just collect someone's worst moments. We go through their whole archive with a set method, check a sample of what we might have missed, and show every prediction, the good and the bad. When we haven't gathered enough to be fair, we mark the score "provisional" instead of pretending it's final. That is what the "small sample" tag means.

What if you got something wrong? Can the pundit respond?

Yes. Every prediction has its own page and a full history you can inspect. Anyone, including the pundit, can challenge a quote, a result, or how we read their confidence. We log the dispute, check it again, and if something was off we fix it and the score updates automatically.

Want the full detail? Read the methodology.