(A) Grid corrections reduce noise in the data. Shown are distributions of coefficients of variation (CVs) of internal wild-type controls for 1906 plates across the dataset (coding and non-coding deletion mutants) before and after correction. The median CVs before and after correction are 0.098 and 0.027, respectively. So our normalization approach typically reduces the noise by about fourfold. (B) Plot of the statistical power (1 – chance of non-rejection of wrong null hypothesis) vs. standardized effect size (difference in means divided by standard deviation) using median number of replicates per long intergenic non-coding RNA (lincRNA) gene and condition (n = 9) and number of replicates for control conditions (rich medium, median = 173.5). The large number of replicates enables statistical detection of differential growth with low false-negative rate (type II error rate). The two curves show the statistical power before correction for multiple testing and after Bonferroni correction (which represents a worst-case scenario). A 5% difference in fitness, which is a very subtle effect and approximates the standard deviation of our method, is detected with a chance of 83% (no correction) or 24% (Bonferroni correction) (left dashed line). A stronger 10% difference in fitness is detected with ~100% power (no correction) or 98.5% power (Bonferroni) (right dashed line). Two-sided Student’s t-tests and a standard deviation of 5% were used throughput with a significance threshold of 0.05. (C) Four control conditions (rich and minimal media, each with or without phloxine B), were included in most of the ~30 batches acquired over 2 years. Boxplots of Pearson correlations for technical repeats within one batch (orange) vs. repeats of the same condition across all batches (blue). While correlation within the same batch is consistently higher, this effect is small considering the biological signal as illustrated by the distribution of pairwise correlations across all conditions (green).