Each round has both surveys and markets on the same base claims. Markets pay more, later, while surveys pay less, sooner. Markets pay out proportionally to winning shares. Surveys pay fixed amounts to top ranks in each batch.  (They are also different forecasting methods, and we will be comparing strengths and weaknesses.)

Markets pay based on actual outcome once known, and so are worth twice as much in total. However, we expect only 100 to be tested by direct replication, and will not know which until DARPA reveals them, probably mid-2020.  

Surveys pay each round, using a peer prediction mechanism with bias correction to rank each participant for each of their forecasts. These peer prediction scores are computed based on 

1. an inferred distribution of the true outcome using all contributed predictions 
2. a bias estimation and correction procedure to correct the scoring step.

With high probability, the top rankings will track ground truth. But we won’t know until the contest ends. Therefore these are worth less. 

Details in the rules, and summarized below.