Coding classes in Staffordshire

The learner downloads Johnson's Preface to a Dictionary of the English Language from Project Gutenberg, splits it into words, and keeps each different word once: 2,376 of them, from 9,654 words of text, sorted alphabetically like a real dictionary. The program then looks up every one of the 2,376 words two ways and counts how many times it has to peek at the list, which programmers call probes.

Binary search always peeks at the middle of the part still possible and throws away the half that cannot contain the word. Interpolation search guesses the position from the word itself: a word starting with "s" should sit about 70 per cent of the way through, one starting with "c" about 10 per cent. To make that guess the program turns the first six letters into a number between 0 and 1, in the way a decimal turns digits into a fraction.

On evenly spread random numbers, interpolation search is brilliant: 3.55 probes on average, far better than binary search. On Johnson's real words it is worse than binary search on average and dreadful in the worst case, taking 110 probes to find "courting". The reason is that letters are not evenly spread. In the preface, 238 different words start with "s" and 215 with "c", but only one with "x" and two with "z". A guess that assumes every letter has an equal share lands in the wrong place, and near crowded stretches like "co" it creeps forward only a few words at a time.

This is the real lesson: an algorithm's speed depends on the data it meets, and the promise of a faster method is only as good as its assumptions. Binary search makes no assumption about how words are spread, so it never does worse than 12 probes here. A careful programmer either checks the data first or uses a hybrid that falls back to halving when the guesses stop helping.

So a young person in Staffordshire should learn to code in 2026 to test a clever method on real data before trusting the promise. The longer argument is in why teenagers should still learn to code in 2026.