Sensor data quality study with summary report

Produce a small reproducible data-quality study, not an application. The deliverables are a Markdown analysis report and machine-readable CSV results. Work offline using only the supplied observations; no website, UI or server. Observations (day,A,B): 1,10,20 2,11,20 3,12,21 4,13,19 5,14,20 6,NA,NA 7,100,20 These are fictional sensor readings. NA is missing, not zero. A reading is valid exactly when it is numeric and in the inclusive range [0,40]. Flag values outside that range without deleting the raw observations. For each sensor, report total rows, missing count, out-of-range count, valid count, mean and median over valid values only. Keep missing and out-of-range mutually exclusive; never impute. Write the original table to observations.csv, the per-sensor results to summary.csv, and a concise report.md explaining the cleaning rule, calculations, findings and limitations. Independently verify the results with a small reproducible calculation, and state the verification command and outcome. Any helper script is supporting analysis, not the product. Keep the output small and clear.