Coding and AI classes in North East England

The learner computes each complete year's mean temperature at Durham from its twelve monthly values, skipping 1883 because a month is missing. For a reference period of thirty years, the program sorts the annual means and places two cut points: one between the 10th and 11th coolest years and one between the 20th and 21st. Anything at or below the first cut is cold, anything above the second is warm, and the rest is normal. Then it labels the thirty years from 1991 to 2020 against three different reference periods.

Every row is correct, and together they tell a clear story: Durham's recent years sit well above the older normals. Against 1961 to 1990, all five years from 2021 to 2025 are warm as well, and 2022, at 10.57 degrees, is the warmest complete year in the whole record, while the coldest is 1892 at 7.23. But the last row shows the trap. Terciles built from the same years they classify must split ten, ten and ten, whatever happened. A report that says "a third of recent years were warm" without naming its baseline could be hiding a big shift.

Two programming details matter. The cut points need a rule for where exactly to cut between two ranked years; we take the midpoint, and other rules move the boundary slightly, so the rule must be written down. And a year sitting exactly on a cut needs a tie rule too. The learner writes tests: labels against their own baseline must come out ten, ten and ten, and shifting every year up by a degree must never make any year colder.

Always checking the yardstick behind a label is a habit a North East teenager builds by coding through 2026. The longer argument is in why teenagers should still learn to code in 2026.