I've been doing some experiments with the "sign of active power has changed" (REVAP1) interrupt, and am curious about the timing. In the attached trace I had a ~160W resistive load flowing through the CT, and a ~2300W resistive load wired in the other direction through the same CT. I then turn the 2300W load on, causing it to swing from a ~160W load in one direction to a ~2140W load in the reverse direction (I got lucky and happened to turn it on close to a zero-crossing).
The RED trace shows the current input and you can see it swing from an upward zero-crossing to a downward zero-crossing as the second (bigger) load is switched on. The YELLOW trace is the /IRQ0 signal out of the device. You can see it fired 37.6 msecs after the reversal. This is a 50Hz system, and the meter is setup such that full-scale deflections would occur at approximately 6266W. I'm running the device in line cycle accumulation mode with a count of 1000, giving me LENERGY events every 10 seconds.
There's not much detail in the 7816 datasheet as to exactly when the sign of the power is noted, and compared to the previous sign. I vaguely recall reading in the datasheet for one of your other devices that the sign is noted and compared every time the 48-bit accumulator hits its threshold and triggers an increment or decrement of the WATTHR register. Is that the same algorithm the 7816 uses? Given my reversed load is roughly 1/3rd of full-scale, I would have expected the WATTHR register to be ticking over much faster than the lag suggested by the trace.