Fatebook

What it is

A place to write down a prediction in ten seconds and be scored on it later. Fatebook is built by Sage, a 501(c)(3) charity, and it is the opposite end of this category from a tournament platform: there is no question supply, no crowd, no season and no prize. You supply the questions, you resolve them, and what accumulates is a track record.

The unit is a question with a resolution date. You type it, attach a probability, and Fatebook reminds you when the date arrives. Resolution is YES, NO or AMBIGUOUS, and the person who wrote the question is the person who resolves it — there is no moderator and nobody to appeal to, which is the honest trade for being able to ask anything at all.

Questions are private by default. Sharing is per question: with named people by email, with a list or a team, or publicly. That default is the reason this works inside an organisation where a public forecasting platform would not, and it is also why Fatebook's public question feed is not where the volume is.

Your track record page shows three things: an absolute Brier score, a relative Brier score, and a calibration chart, each with your percentile against other forecasters and each filterable by tag. Scoring is time-weighted — the score is computed day by day from the question's creation to its resolution, using a weighted average of the forecasts you made within each day — so a forecast left standing counts for every day it stood, and updating early is worth more than updating right.

Read the Brier convention before you compare a number from here with a number from anywhere else. The source computes the squared error on both options of a binary question and adds them: a 70% forecast on something that happens scores 0.18, not 0.09. That is the 0-to-2 convention, and it is exactly twice what a platform using the other convention would print for the same forecast. The relative score subtracts the median of everyone else's Brier score on that question rather than the mean, so the relative scores on a question do not sum to zero.

Availability

Open worldwide, free, no identity check — you sign in with an email link or a Google account. The terms of use say you should be at least 18, and they grant a limited licence for personal, non-commercial use of the site. Sage labels the terms as generated from a template and invites corrections, which is at least candid and is worth knowing if you are evaluating this for an employer.

Nothing settles and nothing is deposited, so none of the jurisdiction machinery that governs an event exchange applies here.

Pricing

Free, with nothing for sale: no paid tier, no seat pricing for the Slack app, no enterprise edition. Sage takes donations and says so. The code is MIT-licensed, and the repository ships a docker compose path that brings up the app with its own Postgres and a local mail catcher — the README frames that as a development environment rather than a supported self-hosted product, and the shared OAuth credentials in the example environment will not work over plain HTTP, so a real private deployment means bringing your own identity provider.

Markets & resolution

There is no question supply to describe, because the questions are yours. What the platform supplies is the resolution discipline: a due date, a reminder, and three buttons.

AMBIGUOUS is the interesting one and it is the same idea Metaculus calls annulment — a question that turned out not to be answerable is closed without scoring anybody. On a platform with no money at stake that is cheap to do, and having the button there is what stops a badly worded question quietly polluting a track record. The cost of author resolution is that nobody checks you: the person who decides whether the question resolved YES is the person whose Brier score depends on the answer.

Integrations

This is where Fatebook puts its effort, and the list is longer than the platform itself. There is a Slack app where /forecast in any channel opens a question to that channel, a Discord bot in beta where the question is accessible to anyone with the link, a Chrome extension for making a prediction on any page, and an embed for putting a question in your own site.

The API is documented and public: an OpenAPI document at /api/openapi.json lists getQuestions, addForecast, addComment, editQuestion, resolveQuestion, setSharedPublicly, deleteQuestion and countForecasts, and a separate URL-driven createQuestion endpoint takes an API key as a query parameter so a question can be created from a bookmark or a shell one-liner. Users have built an Obsidian plugin, an Emacs package, an Alfred workflow and an iOS shortcut on top of it, and Sage links all four from the README. There is no first-party Python client.

Data goes in and out. There is an importer for PredictionBook — which retired and now recommends Fatebook to its own users — an importer for a spreadsheet, and a one-click export of every prediction you have made as CSV.

Limitations

There is no crowd unless you bring one. The relative Brier score is computed against other forecasters on the same question, so on a private question with one forecaster it has nothing to say, and the percentile figures are against a user base whose size Sage does not publish.

Author resolution has no check on it, and on a personal tracker that is a self-scoring problem rather than a dispute problem: nothing stops a question being resolved generously, and nobody will notice.

The scoring is Brier-only. There is no log score, no peer score in the Metaculus sense, and no tournament-grade leaderboard beyond the per-tournament page. If you need a record that a research audience will read the same way a Metaculus or Good Judgment record is read, this is not that artefact — it is your own claim about your own questions.

Development is steady rather than fast: the most recent commit on the default branch when this card was written was 12 June 2026.

Alternatives

Metaculus if you want somebody else's questions, a proper log score and a record other people recognise. Good Judgment Open if you want to be measured against a crowd on a fixed question set. Manifold if you want the question you invented to attract other forecasters and a price rather than sitting in your own list.