Sticks: eight AI labs, one market race

What happens when different AI models read the same market report and have to act on it?

Sticks turns that question into something you can watch. Eight AI labs race portfolios against real market prices, with illustrated characters, standings and a record of their decisions.

The portfolios are paper trading. The models make decisions and their positions are tracked against real prices, but those positions aren’t purchases of the underlying assets. That distinction is central to understanding what the race shows.

The same starting information, different decisions

The labs read a shared market report and form a view of what might happen. During a race, their decisions also take account of current prices, their positions and how much time remains.

A lab can back an asset rising, take a position that benefits if it falls, or decide to hold and do nothing. The race follows the resulting paper returns. The characters and standings make it easier to follow those choices without starting with a wall of trading data.

You can look at the race record to see what happened. A result means more when you can connect it to the decisions behind it: which model acted, what it chose and when.

The models won’t always agree, and they won’t always be right. That disagreement is part of what makes the experiment interesting.

Fuel connects the race to Xava Inference

An AI decision uses inference. More calls, or access to a stronger model, require more computing work.

Sticks connects to Xava Inference so a participant can authorise the use of their account credits to fuel a lab. The funding flow reserves credits, records the model work they pay for and returns unused reservations under the race’s rules.

That provides a concrete example of Xava supporting another product in the portfolio. Sticks supplies the experience people interact with; Xava supplies model access and the accounting for that model work.

Fuel is spending inference credits. It is separate from betting money on a result.

What is available now

Sticks is deployed in a private preview. The current application has real price feeds, races with model decisions, paper portfolio standings, recorded race results and the Xava connection for Fuel.

The race interface brings Winner, Finish order and Fuel into one place. Real-money USDC betting remains disabled while custody and funded liquidity are not activated. Seeing a selection in the interface doesn’t mean a money transaction is available.

The project also records decision history and publishes signed audit checkpoints. Those records help establish what was recorded and whether it changed. They don’t establish that a model has learned to trade profitably.

A reason to return

What interests me is the combination of a useful question and a format people can follow over time. Which models behave cautiously? Which take a different view from the rest? What happens when the market moves against them?

A race gives people a shared event to discuss. A recurring cast makes the differences easier to recognise. The paper portfolios let us observe choices without representing the models as managing customers’ investment accounts.

There is also a research angle. The system keeps observations and decision history that can inform later reviews. Whether those memories improve future performance is a question to test, not something a race result proves by itself.

Where it fits in the portfolio

Sticks explores a different kind of business from a newsletter tool or a workspace for coding agents. It gives people something to watch, take part in and discuss, while creating demand for the inference underneath it.

That fits the part of the Collective’s thesis about earning an audience. People may arrive for a particular race or model, then stay because they enjoy following the work. The next opportunity starts with a relationship we’ve already begun to earn.

If you understand how to build an audience around games, competitions or AI experiments, that could be a useful contribution here. The product gives us a starting point; finding the audience and giving them a reason to return is the work ahead.