Not sure I completely follow the question but an existing reference current sink is provided by Rt at 1M8 which is -8.3 uA.

So yes a reference current is subtracted but, being at the very bottom of the picture, it may not be obvious.

"It may need an offset subtracted to account for two pairs in parallel."

It depends on whether you want it calibrated for one or both channels driven at reference current.

The added offset from both being driven in parallel is a log domain measure of the power sum.

The input trims can be reduced so that reference current driving both channels is 0V out if that's desired.

Or they can be calibrated so that a single channel produces 0V and both driven equally produce +18 mV.

I think JR is pointing out that the actual voltage at the Ct node increases when both inputs are driven.

For one thing the value of the timing capacitor is usually doubled when two detectors are in parallel and current-summed to provide the same timing of a single detector.

This is true for the original 2252 and 4301 as well as the 43XX-series because each IC has its own current sink.

In the detector I show It is set by a single, common, resistor.

Secondly, the Ct node is, in the log-domain, the square of the log of the absolute value of the input so rather than rising 9 mV (at 3mV/dB scaling) to indicate the 3 dB power summation of equal inputs, the actual measured increase is 2X that or about 18 mV.

As JR has pointed out earlier the detector provides about half of an RMS computation.

When coupled with a VCA having 6 mV/dB scaling the log-domain square root is taken relative to the 3 mV/dB response of the detector.

The exponential response of the VCAs control port converts the computation back into the linear domain.

If a log-domain result is desired, exponential conversion is not required, but the 6 mV/dB scaling of the detector output would need to be divided by 2 to take the square root to technically provide the log-domain RMS value of the input.

The main improvement of this detector over some of the earlier "303 types" is the added level-shift that it isn't one Vbe up and subject to a larger temperature mismatch between channels.

The 2252 was a one-Vbe-up design and that was corrected in the 43XX-series.