Roadmap
Who decides what "best" means. Today Fugal writes the scoring rule. That should not last, and this page says what could replace it.
Who decides what "best" means
Not started
A competition that pays out on a scoring rule is only as good as the rule, and today Fugal writes it. Nothing on this page is being built yet, and there are no dates.
What the rule decides
For every question, a router weighs each model's odds of being right against its price. How much a point of accuracy is worth in money cannot be measured, only chosen. On the subnet, four choices like it decide what every miner builds:
| Choice | What it sets |
|---|---|
| The value of accuracy | How much extra spend one more correct answer is worth |
| The question mix | Which kinds of question count, and how much |
| The model menu | Which models a router may call, and at what prices |
| The payout | The champion's share, and the margin needed to take the seat |
Whoever sets these decides what the network builds. A token vote would only hand that to the largest holder.
One candidate: decision markets
Futarchy, proposed by the economist Robin Hanson, is summed up as "vote on values, bet on beliefs". For each proposed change, two markets price the network with the change and without it, and the change goes ahead if "with" is worth more. It suits this subnet because the effect of each change, accuracy and spend on new questions, is measured again every round.
The caveat. The market prices a token, which is only a stand-in for the network's value. Thin markets misprice, paid metrics get gamed, and futarchy has little track record. It is a candidate, not a choice, and where it would run is undecided.