Quantified Intuitions

What it is

A training site rather than a forecasting platform, built by Sage, the charity behind Fatebook. Everything here exists to shorten the feedback loop: on a real forecasting platform you learn whether you were right in eighteen months, and here you learn in ninety seconds.

Pastcasting is the part worth the visit. It serves you a question that already resolved, presented as it stood on a chosen vantage date — the question text, the background, and the comment thread people wrote at the time — and asks for your probability. The original crowd forecast is on the page but hidden behind a "Show" button. To stop the future leaking in, the site ships Vantage Search, a search engine that only returns results published before the vantage point, falling back on the Wayback Machine for archived pages, which is why it is slow. The questions are already-resolved Metaculus and Good Judgment Open questions, and the site links the original after you have committed.

Scoring is a log score relative to the crowd forecast at the vantage date: match the crowd exactly and you score zero, beat it and you score positive. The site states, correctly, that the rule is strictly proper. There is a leaderboard, a charts page, and a multiplayer mode.

Two other tools sit beside it. Calibration asks trivia questions and wants a confidence interval rather than an answer, scored by the Greenberg rule — a bounded score that pays for an interval both containing the answer and being narrow, with a deliberately gentler floor on challenge questions so one disaster does not wipe out a session. The Estimation Game is a monthly themed quiz you play in a team, with a leaderboard per month; the September 2026 edition was called Digital literacy and ran 23 to 30 September, and the archive on the same page goes back month by month for years. Anki with Uncertainty turns a flashcard deck into calibration practice.

Availability

Free, worldwide, no identity check. Signing in is a Google account and nothing else — the calibration tool, the estimation game questions and your charts all redirect to a Google sign-in, so an account with that provider is effectively required to use anything but pastcasting's front page.

There is no terms of service and no privacy policy. Both paths return 404. For a free training tool with nothing at stake that is a small thing, but it is worth knowing before an employer's security review asks.

Pricing

Free, with nothing to buy and nothing to subscribe to. Sage funds it as a charity and links a donation page. The source is MIT-licensed.

Markets & resolution

Nothing resolves here, which is the entire point. Every question already has an answer before you see it, and what the site withholds is your access to it: the crowd forecast is hidden until you ask, search results after the vantage date are filtered out, and the link to the original question appears only after you submit. If you already know the answer to a pastcasting question there is a checkbox to say so, which takes it out of the pool.

That makes the platform's relationship to the rest of this category an unusual one: it is downstream of Metaculus and Good Judgment Open rather than a competitor to them, and it exists because their resolved archives are a training set nobody else was using.

Integrations

None to speak of. There is no public API, no export and no documented way to get your own history out — the routes under /api/v0 are the app's own, session-authenticated and undocumented, and an unauthenticated call to them is refused. The practical route to the data is the MIT-licensed source rather than an endpoint.

The repository is actively committed to: the most recent commit on the default branch when this card was written was 1 September 2026, and the repository is not archived. It is also a two-line README with no setup instructions, so running your own instance means reading the Next.js and Prisma configuration yourself.

Limitations

It does not build a record anyone else will read. A pastcasting rank is a rank at a training exercise on questions whose answers exist; it is not a forecasting track record, and presenting it as one would be a category error. If you need a record for a research context, that is Metaculus or Good Judgment Open, and it takes years.

Calibration training has a known ceiling: a forecaster who becomes perfectly calibrated on trivia intervals has acquired one of the two halves of a good forecast and not the other. Nothing here trains the half that consists of knowing something.

Pastcasting has its own leak problem that no search filter fixes, and the site is honest about it: you may simply remember what happened. The checkbox helps the question pool more than it helps you.

Sage says explicitly that it is still gauging by usage whether pastcasting is a useful addition to the forecasting ecosystem before committing more time to it. That is a candid thing to publish and also a statement about how much to depend on the site.

Alternatives

Fatebook is the same organisation's tool for the opposite problem — real questions about your own future, scored slowly. Metaculus and Good Judgment Open are where the questions here come from, and where a record counts for something. Manifold if you want a fast feedback loop on live questions and can live with profit rather than a proper score as the measure.