Techniques for automated determination or correction of count bias are based on nucleic acid base content on a finer grained scale than a bin of interest in a target sequence. The techniques include obtaining a target sequence with bins where relative abundances indicate a condition and raw counts Hj of reads, from a subject, which start at each locus j. A partition indicates a fine-grained window at a position relative to a current locus and multiple strata indicating different base contents. Each locus is attributed to one stratum k(j). An expected count of each stratum, E(k), is determined based on Hj for j belonging to the stratum and a number of loci in the target belonging to the stratum. A copy number of a bin is based on a sum of E(k(j)) in the bin. Output data indicates condition of the subject based at least partly on the copy number.