Removes the float noise that inverting a pixel coordinate introduces.
The error is absolute rather than relative — it comes from arithmetic across
the whole domain, so a bar near zero on a chart that runs to 1200 is wrong in
the same decimal place as one near the top. Rounding is therefore sized from
the domain span and not from the value: keep ten digits below the span's
leading digit, which is far more precision than any plot's pixel resolution
can encode and far less than the noise floor.
Removes the float noise that inverting a pixel coordinate introduces.
The error is absolute rather than relative — it comes from arithmetic across the whole domain, so a bar near zero on a chart that runs to 1200 is wrong in the same decimal place as one near the top. Rounding is therefore sized from the domain span and not from the value: keep ten digits below the span's leading digit, which is far more precision than any plot's pixel resolution can encode and far less than the noise floor.