The problem occurs when a nucleus is produced (e.g. by radioactive decay) in the highest of the excited levels contained in the G4LEVELGAMMADATA database. 

For instance, Pb-212 can decay to the 415,272-keV level of Bi-212, which is the highest one reported in the G4LEVELGAMMADATA database.
 
Although the level energy is exactly the same in the two databases, the difference (excitation-maxLevelEnergy) calculated in G4ContinuumGammaDeexcitation::CanDoTransition() is positive (5.1459947e-12) probably because of rounding errors. The condition   
if (excitation <= _levelManager->MaxLevelEnergy()) at line 133 
is NOT met, causing CanDoTransition() to return "true", instead of "false". Therefore, a ContinuumGammaDeexcitation takes place instead of a DiscreteGammaDeexcitation and gamma rays are produced with the wrong branching ratios (e.g. the 176.7-keV line is produced at 1.76% instead of 0.05%).

Possible solutions:
1) hack the database $G4RADIOACTIVEDECAYDATA, so that the excitation energy is slighly lower than the level energy reported in $G4LEVELGAMMADATA. This is necessary only if the highest excited level is populated.

2) add a tolerance in the "if" block of G4ContinuumGammaDeexcitation::CanDoTransition(), line 133, e.g.
if (excitation <= (_levelManager->MaxLevelEnergy()+1*eV))