Did you divide the transmission by the Purcell factor (dipolepower/sourcepower) or multiplied it? It should be divided. If the result is correctly renormalized, there might be two causes:
1: the dipole power may not be accurate when the dipole is located in lossy material, or in small mesh size. Please search dipolepower in Lumerical website to get more information, as the current forum has difficulty with a website link.
2: When you use broadband dipole to extract excited mode information, caution ay be taken that, the mode profile is frequency dependent.
I guess your result might be related to the mode profiles that are at one frequency but the decomposition is for broadband.