Hello.
There is an incorrect level evaluation in G4NeutronHPInelasticCompFS. Due to inconsistency in level structure between data in G4NDL and inelastic gammas specified in text files under $G4NEUTRONHPDATA/Inelastic/Gammas in some cases level number is shifted by 1. For example inspect (n,n')-reaction on carbon: if exit channel it==2 (i.e. MT=52) G4NeutronHPInelasticCompFS::CompositeApply deexcitates nuclei with 4.43 MeV instead of 2 gammas - 3.21 and 4.43 MeV, if it==3 we have 3.21 and 4.43 MeV instead of 9.64 MeV.

Please, consider the replacement at http://www-geant4.kek.jp/lxr/source/processes/hadronic/models/neutron_hp/src/G4NeutronHPInelasticCompFS.cc#L374 .

This if-block:

if ( eExcitation < theGammas.GetLevel(iLevel)->GetLevelEnergy() ) 
{
   find = true; 
   iLevel--; 
   // very small eExcitation, iLevel becomes -1, this is protected below.
   if ( iLevel == -1 ) iLevel = 0; // But cause energy trouble. 
   break;
}

with this:

G4double level_tollerance = 100*keV; // maybe too high?
if ( std::abs(eExcitation - theGammas.GetLevel(iLevel)->GetLevelEnergy()) < level_tollerance )
{
   find = true; 
   break;
}
else if ( eExcitation < theGammas.GetLevel(iLevel)->GetLevelEnergy() ) 
{
   find = true; 
   iLevel--;          
   // very small eExcitation, iLevel becomes -1, this is protected below.
   if ( iLevel == -1 ) iLevel = 0; // But cause energy trouble.
   break;
}