Why does the scale reading depend on the position of the object on the scale and how to equalize that? I have a KERN DS 100K0.5 industrial platform scale, range up to 100 kg, resolution 0.5 g, more or less like this:

The specifications say reproducibility 0.5 g, linearity 1.5 g.
When I stand on the scale, each time the reading is different, in a range of over 50 grams. So the reading changes within 0.1%. That might sound good, but when I put a large 5 kg jar on another scale with 0.1 g accuracy I get almost the same reading each time, only different by 0.1 g max, so can achieve 0.002% reproducibility, and my aim is to achieve the same 0.002% reproducibility while weighing myself (for the purpose of studying the rate of body water loss through perspiration and breathing over time).
I notice that the reading depends on the way I stand on the scale, if I hold my arm just a few cm forward for example, the reading is already different.
How can I equalize the readings so it would not depend on the way I stand on the scale? I tried sitting down curled on it, to make myself contained in a sphere closer to my center of gravity, but it didn't help much.
I tried not only using the scales averaging "animal weighing" mode, but also repeating that measurement 10 times, and taking the average after rejecting highest and lowest value, but the standard deviation of that heavily averaged measurement (n-1, divided by square root of the number of measurements) is still about 5 grams, and the process takes a long time.
How do the forces of my body gravity act on the scale differently if I have my arm a few cm forward? Maybe be not the average, but the highest or lowest of a series of 10 measurements would be most close the my actual mass?
How can I make it work more like with the glass jar, that each time I put it on the scale the reading is the same?
If not with this scale, what other kind of scale and/or weighing technique could achieve reproducibility close to 0.002% when weighing myself? The absolute value does not need to be accurate within 0.002%, only the difference between two consecutive measurements have to be accurate within 1 gram, so for example if I weight myself, drink a glass of water, and weigh myself again it will tell the mass of drank water within 1 gram.
 A: I do not know if this will solve you problem but I've seen the following done (in a kinesiology lab). Two "footprints" are drawn on the standing plate so the subjects always stand more or less at the same place and in the same position (arms along the body).
It is possible that, by moving on the scale or by changing your mass distribution by extending your arms, you end up producing a displacement of the spring or springs inside the device that is not completely vertical.  (Obviously this depends on the balance).  The response of the spring to counterbalance the weight can thus also contains a twist (in addition to a purely vertical displacement), and the twisting part is not captured by the scale reading (which often measures vertical displacements.) This is how it was explained to me that standing in different spots can produce different readings (although in this case the scales had several years of usage).
In the case of very heavy objects, it was also possible to even out the pressure on the balance by placing the heavy object on a plank.
BTW, the readings can also be affected by simply standing on your toes, which places uneven deformation pressure on the actual plate, hence the need to even out the pressure on the plate.
