Hi,
I'm having trouble interpretating the calls.cns files derived from running cnvkit. The weight is supposed to be a measure of how reliable the result is right? But what is considered being a small weight value and what would be a "good" one? Right now all my weight values are below zero but varies between 0.02-0.6.

The weight value doesn't correspond directly to a p-value. It's either the sum (current & recent CNVkit versions) or mean (some earlier versions) of the weights of the bins that the segment covers. Those weights indicate the estimated standard deviation of bin log2 values in a (possibly hypothetical) pool of normal samples used as the reference -- in the reference .cnn file this is the "spread" column. Reference bins with spread > 1.0 are dropped automatically under the assumption that they're unreliable.

(If your segment weights are all below 1, you might be using an older version of CNVkit. Consider updating to take advantage of recent improvements.)

If you want to quantify the reliability of a segment, try the segmetrics command, in particular the --ci (confidence interval) or --sem (standard error) options.