Annual “hottest-year-on-record” headlines rely on global temperature-anomaly indices built from weather-station data that are incomplete, geographically biased and contaminated by non-climatic artefacts (urbanization, poor siting, instrument changes). Automated homogenization algorithms, especially NOAA’s Pairwise Homogenization Algorithm, are widely believed to overcome these challenges. However, in practice they often rewrite station records on a daily (or monthly) basis. Adjustments applied to the same record often change from one day to the next and rarely match documented station events. Meanwhile, current statistical homogenization techniques blend urban warming into rural series. The resulting homogenized network therefore converges on a more urbanized average rather than removing bias. In contrast, it is shown that a rural-only Northern Hemisphere reconstruction that avoids automated homogenization shows 38 % less warming than standard estimates and also implies a more cyclical alternation of warming and cooling periods since the late-19th century – a period that coincidentally marked the end of the Little Ice Age.
