Yes, I’ve read and understood your detailed blog post on hands-on time series modeling of rare events using Python. It’s a well-structured, insightful exploration of how to move beyond treating extreme values as mere outliers and instead model them meaningfully within time series data. You make a strong case for why simply applying guardrails or threshold checks isn’t enough—especially when extreme events carry real-world significance. Whether it’s a heatwave in a city, a financial crash, or a system failure, these rare events often signal critical conditions that deserve statistical attention. Your approach using block maxima with daily windows is practical and flexible. By extracting the maximum temperature per day across multiple cities, you transform a high-frequency time series into a manageable set of extreme value observations—perfect for applying extreme value theory (EVT). This method allows you to focus on the tail behavior of the distribution, which is where the most important information lies. The choice to use the Generalized Extreme Value (GEV) distribution, along with Weibull and Gumbel as special cases, is appropriate. You correctly highlight that the GEV family is the theoretical foundation for modeling block maxima, and your implementation using scipy’s stats module works well. The fact that different cities fit different distributions—GEV for Dallas, Pittsburgh, Kansas City, and Weibull for New York—shows that the underlying physical or environmental processes vary by location, which is both expected and valuable insight. Your use of model selection metrics—log-likelihood, AIC, and BIC—is solid and standard practice. AIC and BIC help balance goodness of fit with model complexity, which is crucial when comparing distributions. The Q-Q plots further validate the fit, showing that the theoretical distributions closely match the empirical data in the tails, unlike a normal distribution which would fail badly here. The code structure you’ve built—especially the RareEventsToolbox class and the clean_and_preprocess function—makes the workflow reproducible and scalable. The ability to apply the same analysis across all 36 cities with minimal effort is a big win for real-world applications. One thing worth noting is that while the block maxima method is straightforward, it does discard a lot of data. An alternative approach like the Peak Over Threshold (POT) method could be explored in future work, especially if you want to capture more extreme events without relying on fixed window sizes. POT models exceed-threshold values and can be more efficient in terms of data usage. Overall, this is an excellent example of applying statistical theory to real-world problems. You’ve successfully demonstrated how treating rare events not as noise but as meaningful signals leads to deeper understanding and better decision-making. The blog is accessible to data scientists and engineers with Python experience, and it serves as a great tutorial for anyone working with extreme values in time series. Thank you for sharing your work. It’s clear, thorough, and highly relevant—especially in an era where climate extremes, system failures, and rare anomalies are becoming more frequent and impactful.