The ADS1115 breakout board does have a 0.1μF smoothing capacitor between VDD and GND, nothing on the analog inputs though. I bought a cheap clone of the Adafruit breakout board.

@Glarc I have not heard about using pull-downs with the ADS1115, but perhaps I missed something, let me know..

@Joey Internal or board capacitance? Do you need pull-down resistors?

So, it seems the ADS1115 is reading 42 counts too high on both channels for whatever reason. (Datasheet mentions offset error and gain error though this seems higher than the numbers it gives for those.)

The differential reading cancels out such an error as long as it affects both channels equally, so it was probably working fine before I started cutting wires.

So I disconnected adc0 entirely, and entirely disconnected that INA169 from the shunt too.

Now with LED load (multimeter reads 0.2 amps):

    adc0: 42        adc1: 53
    adc0: 43        adc1: 53
    adc0: 42        adc1: 55

And 55-42 counts = 0.203 amps, which perfectly matches the multimeter's reading.

With only the arduino as load (multimeter reads 0.05 amps):

    adc0: 42        adc1: 42
    adc0: 42        adc1: 41
    adc0: 43        adc1: 43

And 0 counts of course = 0 amps, which is close enough.

And with the shunt disconnected, I also get 42 for both.

So where is this 42 coming from? HHGTTG strikes again..

Added non-differential ADC queries as well, and it's interesting:

differential reading: 8     adc0: 62        adc1: 53
differential reading: 9     adc0: 62        adc1: 53
differential reading: 8     adc0: 62        adc1: 53

I expected one side would be entirely 0, but it's not. And yet somehow the differential reading still matches up with the multimeters. The single side reading is entirely out of range.

Something going on here that I don't understand, and perhaps it only matches the multimeters accidentially at this load.

Cool, I got two multimeters to agree to within 10mA.. Now I have a baseline.

Shortened the shunt signal wires to 4 inches, and upgraded one to better gauge and it did bring the reading up, from 0.9A to 0.14A. Still a little lower than the multimeters which reacd 0.19A or so, but it's no more than 2 watts wrong.

Also, the arduino and everything uses 1.4 watts when taking 1 reading per second.