I agree that Hamon would be helpful, but I think you can do it even more... Hamon-y. Keep in mind that his approach was the ratio of N resistors in series to those same resistors in parallel, giving an accurate ratio of N^2. There are ways to get accurate ratios of N, but they seem a little harder.

The error in the ratio is based on the square of the sum of the individual errors wrt the average, but you don't need to measure all that. If you know the resistors are within 1% of their average, you know the ratio is within 0.01%. And 0.1% resistors, if you can manage or select them, gives you a ratio at 1ppm. That's the magic.

So a very accurate resistive ratio of 4 or 9, with some good high-res voltage measurements across them, giving current ratios, seems helpful somehow.

IOW, I think you can avoid having to know the individual resistor values, other than their relative values to get some idea of the ratio accuracy. Use them in series, use them in parallel, and use the ratio. This also avoids having to know the absolute voltages, because you're only interested in their ratio. Everything pretty much lands on the accuracy of the reference current measurement.