*Radiant, our tool for mapping the forecasts behind a decision, is in open beta. You can sign up today. It is free while it's in beta. No waitlist, no credit card. Read more, check out* [*a sample map*](https://radiant.metaculus.com/projects/557bab12-0f70-4b32-8316-a2f15b9379d9)*, or* [<u>*get mapping!*</u>](http://radiant.metaculus.com) ## **What is Radiant?** [Radiant is a visual map](https://radiant.metaculus.com/) of the predictions behind decisions. Boxes hold the numbers. One might carry a live Metaculus forecast, another a base rate you build from historical data, another a colleague’s best guess with their name on it. You can also drop in a box that holds no number at all, just a title and some notes, for when you’re still sketching. Where a number comes from a Metaculus question or a base rate, the box keeps that source attached. [<img height="269" width="438" src="https://substackcdn.com/image/fetch/\$s_!SvmP!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8ce800ee-9515-496b-9249-737b7cb6df5c_438x269.gif" />](https://substackcdn.com/image/fetch/\$s_!SvmP!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8ce800ee-9515-496b-9249-737b7cb6df5c_438x269.gif) If you’re unsure, give a range (“somewhere between 15 and 40, probably around 22”). Any formula that reads that box carries the whole range forward instead of rounding it to a single number. Your team adds their own numbers to the same boxes, and every view stays visible instead of overwriting the last one. You find out where you disagree, then spend the meeting on the one question that actually moves the decision. [<img height="269" width="438" src="https://substackcdn.com/image/fetch/\$s_!hLmE!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9100824d-2858-4b6e-bab6-401366d54a10_438x269.gif" />](https://substackcdn.com/image/fetch/\$s_!hLmE!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9100824d-2858-4b6e-bab6-401366d54a10_438x269.gif) Arrows show how one thing affects another. Think of them as notes vs. calculations; arrows let you sketch your thinking, and drawing one never changes a number. When you want a node to actually depend on its inputs, write a formula (or have Radiant draft one from the arrows you’ve drawn), and the map will do the math. [<img height="269" width="438" src="https://substackcdn.com/image/fetch/\$s_!4U-f!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7a69d93e-fb4f-4678-9313-f0a6fbbca2c4_438x269.gif" />](https://substackcdn.com/image/fetch/\$s_!4U-f!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7a69d93e-fb4f-4678-9313-f0a6fbbca2c4_438x269.gif) Not sure where to start? Describe your question in the chat, and Radiant’s AI assistant will draft a first map for you to work from. [<img height="269" width="438" src="https://substackcdn.com/image/fetch/\$s_!CI9i!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F905be4e5-b8ac-4658-8dca-aba1ff01dae4_438x269.gif" />](https://substackcdn.com/image/fetch/\$s_!CI9i!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F905be4e5-b8ac-4658-8dca-aba1ff01dae4_438x269.gif) ## **Why Radiant exists** Being right about the future matters more every year. “We’ll be ready by Q3” is a forecast; so is “we can hire for that,” “costs will fall,” and “those regulations will change.” Every plan is constructed from predictions like these. But typically, none of them get a probability, so nobody can check them, and nobody notices when one stops being true. Some live in a spreadsheet only one analyst understands, others are on slide 14 of a deck from 2 months ago. Most are trapped in people’s heads. Take a climate funder choosing between two programs. One pays off if grid batteries become much cheaper by 2030. The other pays off if a particular bill passes. Each program officer has an implicit sense of the probabilities, and neither writes it down. The board hears two confident pitches, but what it never hears is that the choice comes down to a probability estimate where the officers disagree by about 30 percentage points. ## **Why we’re building this** For over a decade we've collected forecasts, scored them against what actually happened, and learned what makes a question resolvable. Here’s our [**track record**](https://www.metaculus.com/questions/track-record/). More recently, we've been doing the same with [**AI forecasters**](https://www.metaculus.com/futureeval/). Radiant takes that discipline to the questions that never get asked cleanly: the ones buried inside a strategy, a grant decision, or a policy. On a map, a vague worry becomes something that can resolve. It also makes a decision legible. Anyone who wasn't in the room can open the map and see what the call rests on. So can any AI agent you connect over MCP. For questions of global importance, we think that's the whole point. ## **Where Radiant has been used** Center for Civic Futures brought Radiant to policymakers from 12 states to map how AI will reshape the workforce. ARIA's Horizon Scanning Study Group used it to map bottlenecks in fields like bioengineering and earth-sensing. A major funder also built avoided-warming estimates with it. We're excited about these experiments and that others now have the chance to test out Radiant for themselves! ## Start today Radiant is new, so expect some rough edges. If you hit one, tell us; that’s how we’ll be able to make it better. Credits are free while it’s in beta. Start a map of your own, or send us a decision you’re actually facing and we’ll map it with you. Radiant’s documentation covers how each piece works, including what it won’t do yet; [**read that here**](https://radiant.metaculus.com/docs). ### [<u>**Get mapping →**</u>](https://radiant.metaculus.com/) ### [**Book a working session →**](https://radiant.metaculus.com/book) Build something? Post the link in the comments. ## **Further reading** Breaking a hard question into forecastable pieces has a long history. Here are a couple good places to start: * RAND & Metaculus, "[Integrated Strategic Forecasting: A New Methodology to Enhance Decision Advantage](https://www.rand.org/pubs/perspectives/PEA3490-1.html)," on merging scenario planning with crowdsourced forecasting to sharpen policymakers' decision advantage. * Forecasting Research Institute, "[Conditional Trees](https://forecastingresearch.org/research/ai-conditional-trees)" (2024), on building networks of questions with experts and superforecasters.