Will a popular forecasting platform accept function definitions as forecasting questions by Jan 1, 2025?

In the future, it might be possible for general-purpose forecasting platforms to accept functions as forecasts. Question writers would provide function definitions, and forecasters would provide programmatic functions.

This workflow would be significantly more complicated than current forecasting platform workflows, but it could be far more scalable and general.

There are some platforms like Kaggle that might have arguably done this for very limited domains and uses. However, forecasting platforms like Manifold/Metaculus/Prediction Markets are much more general-purpose and have a different workflow.

"Popular Forecasting platform"
A forecasting website, like Manifold/Metaculus/Kalshi. It must have at least 100 unique users per month to qualify.

"Function definitions as forecasting questions"
A question writer can propose a fairly-arbitrary function definition. For example, "For any time T from 2024 to 2030, and any stock ticker name, predict the company valuation".
(time: [2024 to 2030], tickerName: string) => distribution

Forecasters would forecast by uploading programming functions or providing API endpoints that would call programming functions.

Get Ṁ600 play money
Sort by:
sold Ṁ324 NO

We're planning on adding some simple integration with Squiggle Hub - but Squiggle Hub isn't really a "popular platform".

That said, as I look at the definition, it says "It must have at least 100 unique users per month to qualify.", which is actually fairly low. Right now maybe Squiggle Hub has ~30 users per month or so, so it seems likely this will be a success.

@OzzieGooen I started off betting "no", because no other forecasting platforms seem to be adding this functionality, but then I re-read that part of the question.

We at QURI have been working on Squiggle, which is a programming language custom-built for this sort of thing. However, I imagine it might be difficult to get Squiggle (or similar) for it to be used in a popular forecasting platform, as the complexity is significant.

There's one possible edge case where a popular forecasting platform allows for these, but it's done poorly, and no one uses them. If this happens, I'd still resolve this question as a YES. The key thing is the support, not the use. (Happy to add other questions for that part, though)


More related questions