Anyone know of an analog version of a sliding correlator/covariance circuit?

I would like to have a circuit with an output that tells me if two voltages are correlated.

I smashed my head for several days and the closest thing I can think of is to choose a frequency spectrum that only the two signals share and then use high pass filters to find when the two signal's derivatives are the same polarity and use this truth table to gate an integrator fed by the "dependent" variable so that only when the two signals are changing in the same direction is the signal in question weighted. The problem is that I know this is not a true correlation computation, and I find my idea impractical. That's why I'm asking if anyone with experience knows of methods for finding this.

Can't you simply scale down one of the voltages to the level of the other one and use a comparator with dedicated hysteresis?

I believe the problem with scaling for my purposes is that one unit might only change by 30ppm/C and the next unit with the same resistor part might change at 50ppm/C. My signal will be dependent on this change. Both are bad, but if I scale for one case, it won't necessarily work for other cases. I should explain what I'm trying to do then.

I have a voltage divider for feedback to a regulator, but one of the resistors in the divider has a large enough temp. coefficient that it will create unacceptable error in output voltage with temp. variation. This resistor is 50 Mohm and 1kV+ working voltage, so buying a better part is somewhat price limited. To get around this, I am inserting a temp. independent resistor in series with temp. dependent resistor so that any variation in voltage across this independent resistor at very low frequency (thermal time consant) will be because of my regulator adjusting the voltage from the temp. drift of the large temp. coefficient resistor. I want to compare this slow-varying voltage to a temperature-only dependent voltage source so that when my regulator temperature compensates based off of the change in voltage, it won't trick itself to think the temperature is changing in the opposite direction and become unstable. Basically, I need the temp. sensor to compare to so that I know when its my regulator/temp. compensation circuit changing the voltage vs. when its the temperature doing it.

- - - Updated - - -

This is what I was thinking of trying to do:

If I had a correlation circuit that took the voltage from the temperature independent resistor as 1 input and the voltage from a temperature sensor (thermal voltage of semiconductor or thermistor) as my other input and put out a voltage value from -1 to 1, I could use this to drive a comparator with a threshold at some % close to 1 or -1. If the correlation is close to 1 or -1, I would gate an integrator on and start integrating the voltage from the temperature independent resistor. If the correlation starts to go back towards 0, the gate is turned off and the integrator is no longer increasing/decreasing its output voltage even though both signals may be changing in opposite directions or just one of the signals is changing while the other is stable. This amount of voltage on the integrator would be used to drive something like a current source that adjusts my regulator's reference voltage. If i have a change in output voltage but no change in temperature, I will know to ignore this change because it is not temperature related. Any load changes will be corrected for sooner than my low frequency, low bandwidth correlator would see since my voltage regulation bandwidth is in the kHz range while the thermal time constant is < 1 Hz.