You might ask
@sparker.. Re-zeroing at the test weight does not seem like the ideal solution to me. But then, I'm at 'zero' when it comes to experience with scale tuning.
I'm willing to try and explain my intuition though.. if incorrect, it wouldn't be the first time this year I was.. ahhh, suboptimal in my reasoning. That said:
With the poises (?) at zero, adjusting height pins the left side of a X-Y plot of weight .vs. error. After doing this (as you have), if you find that the error increases as you add more weight - what is happening may be due to a non-linear response from the scale. I.e., the more weight you add the greater the error (distance) between a plotted line of actual weight .vs. indicated weight. The gap is error, but by starting all lined up at zero any error then will increase (or decrease) in magnitude as measured weight goes higher. In a plot, the two lines get farther apart as you move to the right.
How does that happen? There are weights in the bottom of the tray holder.. and if you removed one 'bb' from that spot and re-zeroed, you might find that the scale indicator now lines up. Again - I'm trying to sort this out in my head with no real-world experience to back it up so.. grain of salt, right? (Looking for someone with experience to chime in!)
Principle at work here is that with three of four variables fixed:
1) Test weight remains 35gr
2) Pivot point (blades) remain at the same spot
3) Poises set the same
What would cause the pointer to go down 0.1gr? At least one answer is less weight at the tray.. though we did adjust the pointer at zero again.
This should change the 'slope' of the line - a plot of measured weight .vs. error. What my weary brain can't tell you from experience is if you want to a) add weight and re-zero or b) subtract weight and re-zero.
Hopefully in the above ramble lies a principle of operation that'll get us to the right answer.
?