You can’t expect the best advice if you only tell half of the story. Just saying…
So it sounds as if you need to treat each rack as two separate items for measurement - even though for display purposes you can lump the readings together in emonCMS (which is why I asked about that).
One reading, if the voltages were a long way apart (i.e. at the edge or outside your acceptable accuracy band) would be enough to say you needed to monitor all three, but alas the converse isn’t necessarily true. It’s going to depend largely but not totally on how your total loads are balanced - and to a lesser degree on your neighbours on the same local substation transformer.
Every c.t. ever made is inaccurate at low current - it’s inherent in the way the magnetism in the core works. The only difference is what ‘low’ means as a percentage of the rating, and that depends on the grade of the c.t. A ‘revenue’ grade will go proportionately lower, and cost disproportionately more.
I can’t read the rating of your breakers, so I don’t know the current you expect to measure - but as you say “a few hundred watts”, I’m guessing 3 A max? You’re definitely pushing the limits with the 100 A c.t., both from the c.t’s point of view and the ADC’s – you’d be better off with the SCT006 - the two I tested (the report is in ‘Learn’ - along with one for the SCT-013-000) started getting iffy at 50 mA for linearity, but were still OK for phase error.
You realise that you can measure the sum of the currents on the same phase by putting the c.t. around more than one phase conductor?
That’s the difference between a “ring core” and a “split core” c.t. Ring core are inherently more accurate and they don’t have the possibility of introducing an unknown and variable air gap in the magnetic circuit, and that can have a huge detrimental effect on the accuracy.
You’ll probably still need to change the 4th (high sensitivity) c.t. input on the emonTx - especially if you choose the SCT006.