We have deployed the 3phase firmware (GitHub - openenergymonitor/emontx-3phase: emonTx 'approximate' 3-phase firmware) on a few emonTX and the values measured against another meters and equipments are always different. In one case we are measuring the power production of a 3phase solar inverter which displays the current generated power and the values do not match with the emonTX.
If you calibrated the voltage and the current accurately, and at the point where you had all 3 c.t’s on the same phase, were they all reading the same value of power and was that correct when compared to your Inverter? (Steps 4 & 5 on page 8.)
If everything at that point was correct, I can only think that the phases of the c.t’s are out of step with the voltage by one phase. This would give you exactly half the power you expect on all three phases, provided that all three phase voltages were identical. If the voltages differ by a small amount, then the power will also differ from exactly half by a small amount.
The final check at Step 6 explains exactly what happens when a c.t. is on the wrong phase - you see exactly half the power and negative. You would correct the negative power by reversing the c.t, and from what you have written, that is exactly what you are seeing.
Provided that you are using the same c.t’s and a.c adapter as when you calibrated the emonTx, I cannot think of another explanation.