What to use instead of INFER

The product

INFER was a crowd forecasting program aimed at people who make decisions with public money. You answered geopolitical, technology and policy questions with a probability, wrote a rationale arguing for it, and were scored — the aggregate was used as an input to research rather than sold back to you as a product.

It changed hands twice without changing platform. The archive's own overview, read on 19 September 2026, gives the sequence: Foretell at Georgetown's Center for Security and Emerging Technology, then INFER at the University of Maryland's ARLIS, then the RAND Forecasting Initiative. The platform was built and operated throughout by Cultivate Labs, which is why the site looked the same across three sponsors and why a forecaster's history survived each handover.

The record it left is specific, and the archive publishes the totals: 3,433 forecasters, 200,901 forecasts, 409 questions published and 298 scored, 183,752 rationales, across 7 seasons whose leaderboards are labelled 2020 through 2026. Standings are ranked by relative Brier score — lower is better, and the number is relative to the crowd, not absolute. The question archive carries all 298 scored questions with how each resolved and how the crowd's aggregate moved over the life of the question; six are marked as not resolved, all of them multiple-select questions where no listed option occurred.

Then it stopped. The banner on every page of the archive reads that the RAND Forecasting Initiative has concluded, that this is a permanent read-only archive, and that accounts, forecasting and comments are no longer available. infer-pub.com now answers a 301 redirect to randforecastinginitiative.org, which was still the case on 19 September 2026.

One thing we could not establish. The archive names no closure date, and RAND's own programme page returned a 403 to us rather than a statement. The date on this page — 27 July 2026 — is the latest resolution date in the archive's own question table, not an announcement, and it is the most defensible end date available from a primary source. The programme's About and FAQ pages are gone from the archive as well, both 404 on 19 September 2026, so nothing about its remit beyond the overview page is checkable today.

What to use instead

The habit and the record are two different losses, and they point at different places.

If what you want is the same scoring and the same seriousness, Good Judgment Open is the nearest thing running. Challenges are scored by relative Brier score against the crowd median, which is the metric you already think in; questions are geopolitical and resolve against named sources or credible open-source evidence; and the top of the leaderboard is a recruiting ground rather than a wall of names, which gives the score an outlet INFER's did not have.

If what you want is a track record that a stranger can read, Metaculus publishes yours under Baseline and Peer scores, with a longer question set and tournaments carrying real prize pools. Its scoring is log-score-based rather than Brier-based, so your instincts about what a given number means will need recalibrating for a few months.

If what you miss is the questions, none of these will publish yours on a policy sponsor's timetable — but Manifold will publish it in a minute, because anyone can open a question and anyone can resolve their own. It is play money in mana, the crowd is much broader than INFER's and much less specialist, and the discipline is entirely self-imposed. It is the only card here where the supply of questions is the users' problem rather than an editor's.

If the scoring was never quite enough, Futuur runs every question twice, in play-money Ooms and in USDC or USDT, on politics, macro and science — a way to find out whether your calibration survives having a position. Real money is barred in the US, the UK, Canada and much of the EU, so check the terms against where you live before assuming it applies to you. The rest of the prediction market venues category is the same trade at different prices.

What you give up

A named consumer for the crowd's answer. INFER's aggregate went into research that informed policy analysis. That is the whole difference between forecasting and a scoreboard, and none of the four replaces it: Metaculus and Manifold publish to whoever reads, Good Judgment Open feeds its own organisation's commercial work, and Futuur's consumer is the market price. If you forecast because someone with a decision to make was reading, that reader is gone and no card brings them back.

A question set written by somebody with a stake in the answer. 409 questions over seven seasons, narrow, resolvable and sourced, on subjects a sponsor needed answers to. Metaculus comes closest and its agenda is its own; Manifold's volume is far larger and its median question is not remotely that; sports and crypto questions dominate on the venues. You will spend real time filtering to find what INFER simply published.

Your standing, permanently. The leaderboards are frozen and the metric does not port. A relative Brier score is measured against one crowd on one question set; arriving at Good Judgment Open with a good INFER record means arriving at zero scored questions, and the number you earned there cannot be compared with the number you earn here. Five years of work is now a page you can link, and nothing more.

The rationales. 183,752 of them were written and the comment threads are not in the archive — comments are explicitly among the things no longer available. For most serious forecasters the argument under a question was the reason to show up, and that is the part of the record that was not preserved.

Migration notes

Save your standing today, not eventually. The archive has no export button, but it does not need one: the leaderboard page embeds its data as JSON in a script tag with the id leaderboard-data, the table rows are server-rendered, and the question index is plain HTML. One curl of the leaderboards and questions pages puts your rank, your scored-question count and your relative Brier on your own disk. "Permanent" is a plan, not a guarantee, and archives with no budget line behind them have a way of becoming DNS entries.

Keep the same username. The archive lists forecasters under the handle they chose, so a stable handle across Good Judgment Open, Metaculus and Manifold is the only thing that makes a link to your INFER standing evidence about you rather than about a stranger with a good Brier score.

Your individual forecasts are not there. Only leaderboards and question-level crowd trajectories were preserved. If you kept your own spreadsheet of forecasts and updates — many forecasters did — it just became the only copy, and it is worth normalising into something you can carry: question text, close date, your probability, the date of each update, the outcome.

Do not read your new score as a continuation of the old one. Relative Brier against a crowd median, Metaculus's Peer and Baseline scores, and Manifold's published Brier measure three different things. The first month on a new platform is noise from a different distribution, not evidence that you got worse.

Start with questions that overlap. The fastest way to find out whether a platform suits you is to forecast something you already forecast at INFER and have an opinion about — a macro release, a conflict milestone, a technology-diffusion threshold — rather than a sports market you would never otherwise have opened.