Users provide their sentiment with respect to certain stocks (bullish/bearish) and they get rewarded in function of the accuracy of their predictions and the amount of SHP tokens held. There will be a reputation score system that will constantly update itself as the user provides sentiment feedback.

For a more in-depth explanation and an example refer to page 34 of the White Paper:



In equation 1, Rcurrent is the users current reputation score, calculated weekly based on how accurate the provided sentiment was. R is linearly modulated by a scalar constant, m (equation 2) positively for each correct prediction and negatively for each incorrect prediction. That is, for each correct prediction the reputation score is increased or decreased by multiplying the user’s previous reputation score, Rprevious, by m, and adding
this to Rcurrent. This is repeated for incorrect predictions, but by reducing R by R · m instead of increasing it. The maximum value of R is 1, and the minimum value is 0, as stated in equation 3.

The impact of predictions on a user’s reputation score is perhaps best demonstrated by example (let us assume that m = 0.05):

• A user has R = 0.5, and provides 6 correct and 3 incorrect predictions.
Their new R is (0.5 + (6 ∗ 0.5 ∗ 0.05)) − (3 ∗ 0.5 ∗ 0.05) = 0.575. This represents an increase of 7.5 percentage points for all rewards received by this user for correct predictions.
• A user has R = 0.2, and provides 6 correct and 3 incorrect predictions.
Their new R is (0.2 + (6 ∗ 0.2 ∗ 0.05)) − (3 ∗ 0.2 ∗ 0.05) = 0.23. This represents an increase of 3 percentage points for all rewards received by this user for correct predictions.
• A user has R = 0.5, and provides 3 correct and 6 incorrect predictions.
Their new R is (0.5 + (3 ∗ 0.5 ∗ 0.05)) − (6 ∗ 0.5 ∗ 0.05) = 0.425. This represents a decrease of 7.5 percentage points for all rewards received by this user for correct predictions.
• A user has R = 0.2, and provides 3 correct and 6 incorrect predictions.
Their new R is (0.2 + (6 ∗ 0.2 ∗ 0.05)) − (3 ∗ 0.2 ∗ 0.05) = 0.17. This represents a decrease of 3 percentage points for all rewards received by this user for correct predictions.

As the example above illustrates, changes in a user’s R are linear - that is, the relative change in R is dependent only on the accuracy of the user’s provided sentiment, and not the previous value of R.