negative path estimate in UMX cholesky

Dear forum,
I conducted a Cholesky model with two variables in two time-points, using the uxACE command, and then conducted a model fitting procedure with the umxModify command. The model was an ADE model.
The results I got (a screenshot is attached) make perfect sense in terms of the overall explained variance by genetics. Meaning, if I compute the variance explained by both A and D for each variable (after squaring each path), I get the same amount of explained genetic variance as in the univariate models for each variable.

The only thing I am worried about is that one of the paths is negative (the path A to the third variable), even though the phenotypic correlations between all variables are positive (and meaningful).
What does it mean?

Thank you very much,
Lior