How hedwi.gg works

Claims answer claims

A discussion starts from an opening claim: the main result of a paper, or the question you want answered. Anyone with an account can add a claim that supports another claim or attacks it, and each relation names the claim it answers.

Write one statement per claim, so that the next person can answer exactly that statement. Adding claims is free.

Strength: is the claim still standing?

Every claim that has been answered gets a strength between 0 and 1. Nobody votes on it: it comes only from the claims that support and attack it, and from theirs in turn.

Try it on the right. A claim with nothing against it and good support rises towards 1; a strong attack pulls it down; when support and attack are equally strong, the claim stays in the middle. A weak attack barely moves anything, which is why answering an attack is often the best way to defend a claim.

Strength is not the chance that a claim is true. It says how the claim fares in the debate as it stands today, and it changes every time someone adds to it.

Every contribution is a new version

A version is the whole debate at one moment: its claims and relations. Each contribution creates a new version on top of the one you were reading, and versions are never changed or deleted. If you take a claim out, it disappears from the next version and stays in the earlier ones.

When two people build on the same version, the debate splits into lines of argument, like branches. You can read any version, compare two of them, and build on an older one when you think the debate took a wrong turn.

Hold the version you stand behind

In each discussion you hold one version, with a confidence from 0 to 1 that what you newly accept there will hold up. On hedwi.gg, every contribution holds your new version straight away, with the confidence you choose in the same box. Your confidence stays private until the hold is scored.

About two weeks later the debate answers: hedwi.gg looks at how what you accepted turned out and compares it with your confidence. That comparison is the skill of your hold. A hold is good when you were closer to what happened than having no view (0.50) would have been, and bad when you were further. Being confident and right counts most; being confident and wrong costs most.

What a hold is about depends on what you changed. A new claim counts in full, by how strong it turned out. A new relation between two claims, or a claim or relation you took out, counts a quarter as much. A relation is judged by whether it kept its weight, so whether the claim behind it held up and nobody removed it, never by which way it moved the claim it answers: pointing out a correct attack never counts against you. Something you took out is judged by how it fared in the lines of argument that kept it. If nobody reacts, it scores the same as having no view.

You only ever bet against your own forecast: nobody else wins or loses, and no credits move. Your own later contributions do not count towards your score, and each person's lines of argument count once, however many they open.

Judgment and the p-index

Your judgment gathers the skill of all your scored holds into one number: how well you read debates over time. It starts close to zero and needs several holds to move far, so one lucky call is not a record. hedwi.gg shows it on your profile after five scored holds, always with the number of holds behind it.

Your p-index is how much your holds count when hedwi.gg weighs who stands behind each line of argument (its backing). It grows with good judgment and with the number of holds that judgment rests on. New accounts, and accounts whose calls go against the debate, have a p-index near zero, so a crowd of fresh accounts cannot push a line of argument up.

Your authorship is the other half of your record: how the claims, relations and removals you wrote held up, and how strongly people with a high p-index built on your versions. Credits are separate: they can be spent, while judgment, p-index and authorship can only be earned.

Structured reviews and editors

On a paper or a preprint you can write a structured review in three questions: what the paper concludes, whether the results support it, and whether the methods support those results. Your answers become claims in the debate, and reviews are paid from the discussion's review bounty.

Journal editors can read a report of a debate, with the strongest claims across every line of argument, the objections still open and the backing of each line, and accept a specific version for publication.