Hot, uneventful month. Watched most of the World Cup final with Dad even though neither of us had a real rooting interest — France won, he was happy about it for reasons I didn't fully follow, something about a team he liked in the 90s. Spent most evenings on a side project that's going nowhere: trying to get a small model to generate slightly-varied training examples for itself instead of me hand-writing every toy dataset. Not working well. The variations it generates are either trivial (barely different from the original) or nonsensical (technically different but useless as training data). There's a real tension there I don't understand yet — between "different enough to be useful" and "different enough to actually still be correct." Filed it away as a hard problem, not an urgent one. Going back to more basic architecture experiments for now. Senior year starts in three weeks.