AI and programming classes in Elgin

The Met Office publishes monthly records for its former Nairn weather station, which moved site in 1998 and closed at the end of 2014. The learner takes the longest stretch with no missing rainfall, January 1931 to June 1996, all at the original site: 786 months. Each month becomes a letter: L if its rain was in the lowest third for that calendar month, H if in the highest third, M otherwise. The result is a 786-letter string with 272 L, 263 M and 251 H. Now weather questions become text searches. How often did three high months come in a row? That is a search for HHH.

On this data the saving is real but modest: KMP made about 10 to 12% fewer comparisons than the naive method. The bigger difference is behaviour. The naive search kept re-reading earlier months, 1.51 comparisons per month for LLLLLH, while KMP read each month once, in order, and never went back, 1.33 comparisons per month including the extra steps along its table. To see the worst case, the learner builds an invented string of 786 L letters: searching it for LLLLLH costs the naive method 4,686 comparisons and KMP 1,567. Real weather rarely looks like that, which is why the honest summary is that KMP guarantees good behaviour rather than always winning by a mile. The longest dry run in the real record was six months, and the longest wet run five.

Understanding how a search really works keeps Elgin teenagers in charge of the AI tools they use, and learning to code is how that understanding is built in 2026. The longer argument is in why teenagers should still learn to code in 2026.