Video encoder settings are a kind of alchemy.  The math is there.  The metrics are there.  It’s all very cut-and-dry and measurable and matter-of-fact.  Except it’s not.  Does it look good?  No, but, does it look ‘good?’  What looks good for this might not (absolutely won’t) for that.  File too big for the gain?  How many GPU frames you losing in real-time encoding?  And then you add two separate platforms, with two separate sets of requirements and two separate perception expectations at the end.  It’s an applied science in the same way that politics is an applied science: that is to say, not at all.  It is a game in and of itself.