> I got a problem and it is not solvable by the fit weight as explained in the
You got that wrong.
> I got a bunch of data points, which each represents a series of tries:
> 50 tries with 30 fails is represented by 0.6 (60%) the other parameter of
> the point is a sort of percentile but doesn't really matter.
What exactly does "is represented by" mean here? Are the actual
numbers, 50 and 30, present in the data visible by gnuplot, or not?
> What I'd like to have is, that, while the least square process, the residue
> of every data point is weighed by the amount of tries it represents.
Which is *exactly* what the z_error column would do for you, if only you
> (The weigh explained in the gnuplot help for "fit" just alters the y axis
> value, I don't think that solves my problem.)
You misunderstand the documentation, leading to incorrect conclusions.
No, the weight doesn't just modify the y axis values.
What you need to do is crack open your statistics textbook and learn
about counting statistics and their intrinsic error, so you can
understand why the solution to your problem is something like
fit f(x) 'data.data' using 1:2:(sqrt($2)) via ...