L-01 ChannelFollowingUpdateRule is different from whitepaper formula

Summary

When calculating trendPortion (or m(t) in whitepaper), base number absGradient is divided by preExpScaling parameter multiplied by 2. This is a clear difference between implementation and whitepaper

Vulnerability Details

m(t) aka trendShare has the following formula in the whitepaper (page 6, eq (12))

In a more simple term:

However, in ChannelFollowingUpdateRule.sol, abs(gradient) is divided by preExpScaling * 2 beforehand:

So m(t) is implemented like the following:

Impact

• Incorrect weight calculation

• preExpScaling parameter is unknown to end users, and it might cause confusion or unexpected behavior

NOTE

Although it's difference between implementation and whitepaper, it can be mitigated by setting all preExpScaling parameter to 0.5e18

Tools Used

N/A

Recommendations

Fix whitepaper and note about this silent extended implementation and unknown parameter preExpScaling

L-02 Missing negative weight checks when calculating weights with scalar kappa

Summary

In most of the Rule weight calculations, negative weight check is missing for params with scalar kappa

Vulnerability Details

In AntimomentumUpdateRule.sol:120, negative weight check is done if kappa param is a vector:

However, the above check is missing when kappa param is scalar:

Similar issue observed in ChannelFollowingUpdateRule, DifferenceMomentumUpdateRule, MomentumUpdateRule and PowerChannelUpdateRule

Impact

• Incorrect weight calculation

• Inconsistent weight update behavior between scalar and vector kappas

Tools Used

N/A

Recommendation

Use the same check for scalar and vector kappas