Since for CC2 the left eigenvectors are obtained by solving a different eigenvalue problem then for the right eigenvectors, it can in rare cases happen that they don't match together, e.g. because for high lying states different roots were found.
One can check for this by restarting the calculation and requesting just right eigenvectors, but than one needs at the end and with a tighter convergence threshold (conv=<number> in $excitations), and then doing the same for the left eigenvectors (add 'leftopt' in $excitations'). If the enough right and left eigenvectors converged with matching eigenvalues one can restart the calculation for oscillator strength with the original settings.
The alternative is to let the program ignore the problem. This is done if you add 'no_strict' in the $ricc2 input section.