Confido

What it is

Forecasting software you run rather than a site you join. Confido is a workspace — rooms, members and questions — built by the Confido Institute, a non-profit whose stated subject is uncertainty communication rather than forecasting competition. You invite the people you want, ask questions they can quantify, and the app collects probabilities and distributions, aggregates them, resolves them and scores them.

Three question types, and the second and third are what mark it out from the rest of this category. Forecasting questions are the usual thing: an outcome nobody knows yet with a clear resolution later. Knowledge questions have an answer that already exists, so a group can practise calibration and find out immediately how overconfident it was. Implicit belief questions — will venue A suit us better than venue B — never resolve at all, and exist to get a distribution of opinion out of a room without the first speaker anchoring everybody else. That last one is not forecasting and the institute does not claim it is; it is why the product's other listed use is running meetings.

The score is a rescaled Brier that points upwards, and this is the first thing to get straight. The code computes (0.25 − (outcome − p)²) × 4: the range is −3 to 1, a 50% forecast scores 0, and higher is better. Every other Brier score in this catalogue is lower-is-better. A Confido score is not comparable with a Metaculus, Good Judgment Open or Fatebook number without converting it, and the sign convention will catch anybody who skims.

Automatic scoring is binary questions only — ask for a numeric range and the scoring function returns nothing. The institute says as much in its own tournament guide and tells you to export the CSV and compute your own. Calibration curves are computed separately and are the feature the personal-use documentation is built around.

Availability

There is no public instance to sign up to, and that is deliberate rather than an omission. The website offers a demo instance and a contact form; the supported route is building and running the app yourself, which the repository documents as a Gradle build whose run task serves the compiled front end on port 8080. The institute's second selling point, after ease of use, is that running it locally gives you full control of the data — pitched at government and large institutional users.

So the answer to "can I use this" is not a jurisdiction question at all. There is no account, no identity check, no geoblock and nothing to deposit, and also nowhere to join: somebody has to host it first.

Pricing

Free, released under AGPL-3.0, with no paid tier advertised anywhere on the site. The institute is a non-profit and also sells nothing on the page — workshops and calibration trainings are the other half of what it does, arranged through a contact form.

Note what AGPL-3.0 means if you are considering this for an employer: a modified version offered to users over a network has to be offered as source. That is the strictest licence on any card in this category, and it is a deliberate choice by a non-profit rather than an accident.

Markets & resolution

There is no question supply. The questions are whichever ones your workspace writes, which the institute's own material illustrates across a wide spread — a geopolitical question with a 2030 horizon, an internal promotion, a venue choice, and a run of personal questions about whether you will be glad you went to the party.

Resolution is administrative: whoever runs the room enters the outcome. The part worth knowing is the score time, a field in a question's schedule. A prediction is scored only if it was submitted before that moment, which is how the app stops somebody forecasting after the answer is obvious. Three conditions have to hold for a prediction to appear on the scoreboard at all: the question must be resolved, it must have a score time configured, and the prediction must predate it. A room with sloppy schedules produces an empty scoreboard rather than a wrong one.

Integrations

None, in the sense this catalogue usually means. There is no public API, no SDK and no webhook — the scoreboard endpoint the front end calls is internal. What there is, is a CSV export of predictions, which the institute's own guidance treats as the escape hatch for everything its automatic scoring does not cover.

The rest of the integration story is that you have the source. It is Kotlin multiplatform, AGPL, and about five megabytes of repository.

Limitations

The maintenance signal is the weak one. The most recent commit on the default branch when this card was written was 1 September 2025 — over a year old — against 111 commits in total, and the build still declares itself version 1.0-SNAPSHOT. The website and the institute are active; the code is not moving fast.

Scoring is thin next to the platforms in this category that make scoring the product. One rescaled Brier for binary questions, no peer or relative score, no leaderboard beyond a per-room scoreboard, and nothing for numeric questions. If what you want is a record that travels, this is not it; if what you want is your organisation's own forecasters ranked against each other on your own questions, it is exactly it.

And there is nothing to evaluate without deploying it. You cannot read a live question set, check a public track record or see how well its users are calibrated, because by design none of that is public anywhere.

Alternatives

Fatebook is the hosted version of the same idea — private questions, Brier and relative Brier, a Slack app — if you can accept somebody else holding the data. Metaculus is open source too and sells private instances as a service, which is the commercial answer to the same requirement. Prediki sells confidential questions on a hosted platform by the seat, if a purchase order is easier for you than a server.