Coding and AI classes in Livingston

The learner downloads every road, path and cycleway mapped on OpenStreetMap across Livingston: 9,827 ways, the newest edited on 27 September 2026. For the 96 ways edited since 1 July, the API's version history shows what each looked like on that date. Nineteen did not exist then and are set aside, leaving 9,808 ways present in both snapshots. Each way is hashed with SHA-256 over its tags and its list of points, the hashes are paired and hashed upwards, and after 14 levels each snapshot has one root hash.

The two roots differ, so something changed. The program then compares the two children of the root, follows only the child whose hash differs, and repeats at every level. Wherever both hashes match, that whole branch, perhaps thousands of roads, is skipped in one step.

Of the 75 changed ways, 40 had their shape altered and 50 their tags, some both. Two more had been edited during the summer but ended up exactly as they started, and the tree correctly treats them as unchanged, since it compares content rather than edit counts. The saving depends on how clustered the changes are: one change costs about twice the tree height, while changes spread evenly through the data share fewer branches and cost more. Even here, with 75 scattered edits, the tree needed under a tenth of the comparisons.

Being able to verify changes gives Livingston teenagers real authority over the AI tools they use, a practical reason to learn to code in 2026. The longer argument is in why teenagers should still learn to code in 2026.