Hi @regwan,
The negative values occurring around these hours make sense - have you looked at this thread?
HB uses the enhanced 2-phase method for daylight/radiation simulation, which from my understanding is the root cause of negative lux / irradiance values. My rough understanding is the sun patch and sun beam simulations aren’t guaranteed to fully align, especially with coarse radiance parameters. Have a look at other posts from @justinfmccarty and @mikkel for more information.
The full tutorial by Sarith here is a big help in understanding the calculation methodology.
https://www.radiance-online.org/learning/tutorials/matrix-based-methods
My only potential cause for concern, without actually testing, is that the sky dome oddities reported by @lionpeloux for certain weather files could be resulting in misaligned sky patch intensity and sun beam vectors. I’m likely wrong with that concern and I’m sure someone like @chris or Mikkel could easily correct me - my knowledge of how sky patches and sun beam vectors are produced for radiance is close to none.
On the Radiance modifiers, I would expect them to have an impact, but that impact could be relatively minor. Are you explicitly modelling a ground surface? I’d assume the ground reflectance will be one of the most significant variables you could change to see an effect. I would set up a simple model to test.