The higher the value of the shunt resistor is, the lower the errors due to the amplifier input offset will be. Because I am going to place the shunt resistor above the system load, using a larger resistance value will not affect the ground value seen by the system.

I think I can tolerate a maximum voltage drop of 10mV. If I am powering my device with a single cell lipo, that would leave plenty of head room. It will make it a bit harder to test the full life of coin cells, but I think that is acceptable - you want a tolerance on that anyway, right?

At a max load of 100mA, and a max voltage drop of 10mV, I am looking at a 0.1ohm shunt resistor.

At my lowest target current (5uA) this resistor would give me 0.5uV ... which is just over 2xVos for the MAX9922t (the nominal value anyway). That's cutting it close, but it should be good enough.

0.1% tolerance resistors cost ~$3.5 per thousand, and they are rated for watts. I am going to have to calibrate these things anyway, so I am not going to opt for a perfectly trimmed resistor when I can get a 0.5% tolerance resistor for $0.29 (the FC4TR100DER).

Assuming a 24 bit ADC and a 3.3v reference voltage, the LSB of the system will be <0.2uV.