Coding and AI classes in Basildon

The learner downloads the resident count for each of Basildon's 584 output areas from Nomis and writes quicksort in Python, adding a counter every time a value is compared with the pivot. In the order the file arrives, taking the first value as the pivot, the sort finishes after 4,520 comparisons and never recurses more than 15 levels deep. That is close to what theory promises for a good case, about n times log n, here roughly 5,367. The program checks its answer against Python's own sorted list every time.

Now the learner sorts the list again, after it is already sorted, still using the first value as pivot. Every split is lopsided: the pivot is always the smallest value left, so almost everything lands on one side. Comparisons jump from 4,520 to 68,341, fifteen times more, and the recursion reaches 240 levels. With larger lists that depth can crash a program. The fix is tiny: take the median of three values, the first, middle and last. On the same sorted list that takes 3,773 comparisons and 9 levels.

The learner then notices something the textbooks mention in passing. The worst case should be about n squared over two, 170,528 comparisons, yet the program only reached 68,341. The reason is duplicates: only 241 of the 584 counts are different, the value 302 alone appears 13 times, and the program groups values equal to the pivot together so they are never sorted twice. The learner also records that the 584 area counts add to 187,542, slightly different from the published borough total of 187,571, and so always quotes the published total.

Testing code on its worst case, not just its usual one, is a sound reason for Basildon teenagers to keep learning to program in 2026. The longer argument is in why teenagers should still learn to code in 2026.