How we collect data, and why

On each fact page you can give your own confidence in a claim, next to the AI's.

This page says exactly what that collects, what it does not, who else can see it, where it lives and why.

The collection is meant to be an example of the method the site argues for: say exactly what is measured.

1. What we collect when you vote

Each vote is one row in a database on our own server, with these fields and nothing else:

• •A row number the database assigns.
• •Which claim you voted on.
• •Your answer, a number from 0 to 100 in steps of 5.
• •The time the vote was recorded.
• •A salted SHA-256 hash of a random id your browser made for this site.
• •A salted SHA-256 hash of your IP address.

The random id is used to allow one vote per claim per browser, and to show you your own answer when you come back.

The same id is used on every claim, so your votes on different claims share the same hash.

The IP hash is kept with each vote so that a burst of votes from one address can be spotted later.

It also limits how many vote submissions one IP address can send: at most 20 per hour.

For that limit, each submission also writes the IP hash and the time to a second table, and entries older than a day are deleted the next time anyone votes.

The salt is a secret that was made on our server and is kept only there.

A hash cannot be turned back into the address directly, but there are few enough possible IP addresses that someone holding both the salt and the table could test guesses, so the hash is a pseudonym, not anonymity.

Reading a fact page without voting stores nothing on our server.

In your own browser, a vote leaves three things:

• •hs_vid in local storage: the random id, created the first time you submit a vote.
• •hs_voted_ followed by the claim's short name, in local storage: the answer you gave on that claim, so the page can show it.
• •A cookie named hs_vid with the same random id, set by hbar.science after a vote is recorded, readable only by the site's server and not by scripts on the page, set to last two years (Chrome and some other browsers shorten that to about 13 months).

Local storage stays until you clear it, or until your browser does (Safari can clear it after seven days without a visit).

Clearing this site’s data in your browser removes all three, and your vote stays counted under the old id.

2. What we do not collect

• •No names.
• •No email addresses.
• •No accounts or logins.
• •No raw IP addresses in the vote database, only the salted hash.
• •No browser details, device details or location in the vote database.
• •No advertising, analytics or tracking scripts anywhere on hbar.science.
• •No cookies before you vote, except on an unlisted access page for unpublished material, which sets a one-day cookie if you enter its key.

3. What others in the chain can see

The site is hosted by Vercel.

Every request to hbar.science, including a vote, goes through Vercel’s servers, which see your IP address and the details your browser sends.

A vote is passed on to our server by code that runs on Vercel, so the vote itself also passes through Vercel.

That code runs in Vercel’s Washington, D.C. region in the United States, so your IP address, id and vote pass through the US on the way to our server in Finland.

Vercel keeps its own request logs under its own policy, which we do not control: https://vercel.com/legal/privacy-notice.

Our code on Vercel does not write your vote, your id or your IP address to any log.

Vercel’s own records of each request include its address, which names the claim, and your IP address, but not your answer or your id.

The name hbar.science is looked up through Cloudflare’s DNS, in DNS only mode, without its proxy.

Your own DNS resolver, usually your internet provider or a public resolver, sees that you looked up hbar.science.

Cloudflare usually sees only that resolver, though some resolvers pass on the first part of your IP address.

Neither sees the pages you read or your votes.

Your browser never connects to our data server directly; only the site’s code on Vercel does, with a secret key.

4. Where it lives, and for how long

The vote database runs on a server we rent and run ourselves, at Hetzner in Helsinki, Finland, in the EU.

It is not a cloud database service.

Votes have no expiry date in the database.

The database is backed up once a day to the same server, and each backup is deleted after about two weeks.

So a removed vote can remain in backups for about two weeks.

Our server writes no log of requests, and it never sees your IP address as a connection: only Vercel connects to it.

Your IP address reaches our server only inside the forwarded vote, and it is hashed there before anything is stored; request and error logging is switched off, so it is not written to any log.

To remove your vote, write to hello@hbar.systems.

We cannot find your vote by your name, because we never had it, so include the hs_vid value from your browser's local storage for hbar.science (in the browser's developer tools, under storage).

There is no remove button on the page yet.

5. Why we collect it

Each fact page shows an AI confidence score for the claim.

Your votes give the public’s confidence in the same claim, shown as data: count, median, mean and the spread of answers.

The point is to set the two side by side, claim by claim.

Later, for claims that get settled, we plan to score both the public and the AI for calibration with the Brier score, so each can be measured against what turned out to be true.

That scoring is not built yet.

6. What any website can learn about you

Everything above is what we choose to keep.

Most websites can learn much more from your browser alone, without cookies or a vote: your device and operating system, screen size, timezone and language, the fonts you have installed, and a canvas fingerprint from how your graphics hardware draws a hidden image.

Together these can be enough to tell one browser apart from millions of others: in large studies between about a third and over four fifths of browsers were unique (Eckersley 2010; Laperdrix et al. 2016; Gómez-Boix et al. 2018).

A tracker present on several sites can use this to follow one browser between them.

We will publish a page that shows you your own dossier: what your browser gives away to any site.

It is not live yet.