You're missing the point of the parent: There is no 'one true' metric. If you use different metrics (actual lines, logical lines, gzip'd size, etc) you may well find different correlations.