Dear developers,

grading from multiple input fields in Formulas and its Parts may lead to confusing behavior, although mathematically it is correct. The relative error is calculated from a vector of inputs, by definition as:

`_relerr`

= `sqrt(sum(map("*",_d,_d)))`

/ `sqrt(sum(map("*",_a,_a)));`

Now imagine that absolute numbers are quite different in input 1
and input 2. An example below illustates that, requiring numbers
1.e-6 and 100. Setting grading to `_relerr`

< 0.01 (default) means that a user can enter approx. ±1 in **ANY**
field. You would expect 1% relative error for every field, but the
relative error can be enormous, in the first input 0.5/1.e-6.

I found a partial workaround by redefining grading, but you have troubles with 0 values:

sum( map( "<", map("abs", map("/",_d, _a) ),0.01))/len(_a)

I suggest that relative error should be evaluated from every field
separately in the same way as you do it for a single answer. It would
not matter whether you group answers into Parts or leave them alone
separately, the grading would be the same. By my opinion, this should
be the default relative error in Formulas/Parts in multiple fields. There are likely
several solutions, either by introducing new functions or overloading
current functions with lists/members.

Best regards and thank you for your time. Vit