Questions about marginalization in BatchFixedLagSmoother

[EricX-Zhao]

When using BatchFixedLagSmoother, if the robot remains stationary for a period of time (e.g., the first few seconds of Euroc MH03 dataset), the factor graph in the sliding window evolves as follows:

Before reaching the sliding window limit, there are approximately 40 factors, and these factors are observable from X0 to X5,

When X6 is added and X0 is marginalized, many factors associated with X0 are removed, and because the robot is stationary, there are few new factors.

By the time X10 is added, only two factors remain, leading to insufficient constraints.

My question is: after marginalizing a variable, do I need to re-add the measurement data of the smart factors associated with every variable in the sliding window?

For example, after adding X6 and marginalizing out X0, all the factors related to X0 will be removed. However, if these factors are still observable by the variables within the sliding window X1,X2,X3,X4,X5,X6 do I need to re-add these factors?