Interpreting cvx_precision

Hello everyone,

I have a convex optimization problem which does not belong to any of the LP, QP, SOCP, SDP, GP categories and it has a nonlinear objective. When I use CVX for my problem, it selects SeDuMi as a solver and cvx_status is solved at the end. CVX documentation says that it considers three different tolerance levels \epsilon_{solver}, \epsilon_{standard} and \epsilon_{reduced} which can be set by cvx_precision. When a problem is solved, does \epsilon_{standard} give an error bound for the optimal value or is it rather a bound on Euclidean distance from a true optimizer?

Thank you for your help.