The normal layout is well described in the following illustration on Wikipedia. One stage for every decade you want to divide, with decreasing resistor values for every new stage. The total resistance for the new stage must equal to twice the resistance of a single resistor on the previous stage.
Kelvin-Varley example from Wikipedia:
However, because you can't easily get for example 400 Ohm resistors, and because of resistor tolerances and the importance of matching the total resistance, it seems customary to select a value that's slightly higher, then connect a shunting resistor and a trim-pot in parallel with the whole stage, to get the resistance down to whatever is necessary for the previous stage.
What would be wrong with simply using the same value of resistors for every stage, and using a shunt with a trim-pot to get the combined resistance down to the required range again:
There are at least three benefits here:
- The tedious job of hand-selecting 11 closely matching resistors for each value is reduced to sorting a lot of the same value into bins.
- No need to buy hundreds each of odd value resistors when you only need 11 of each. The parallel and serial shunt is from E24 and does not have to be matched with everything since they are trimmed out.
- It's more modular. You can construct any number of these and chain. (Of course, the accuracy is still limited by the first segment.)
Perhaps there is some hidden drawback with this scheme. Will it be more sensitive to temperature? More noise?