Puck.js magnetometer as a compass ?

One thing I notice from that is that the values in c vary by +/- 200ish

However the values for min/max are anywhere from 1000 to 1600 apart, so basically however you calibrated it, it must have been near a much bigger magnetic field somehow so the min/max values (and hence the center) are way out.

If you do vmin = vmax = new Vec3(Puck.mag()); when the Puck is on the paper as you had it, and then rotate it slowly by 360 degrees I'd expect that then vmin/vmax will end up around 400 apart, and at that point you'll get a much better reading.

Getting the compass calibrated is a right pain - it's why even on phones where they've got sometimes billions in R&D budget the phone still occasionally asks you to rotate it in circles to recalibrate.

If you know the field strength where you are (looks like ~160ish for you?), you can effectively just keep track of the center point (rather than min/max) and then if the reading is ever more than 160 away from the center point you just update the center point.