sketch_naive_bayes_vs_majority_bench.nx

sketch_naive_bayes_vs_majority_bench.nx -- classification accuracy bench. CLAIM TO VALIDATE: Naive Bayes (Maron 1961) outperforms a majority-class baseline when features actually carry signal about the class. Without real per-feature likelihood, classification = majority class. Test: a streaming binary classification task where class 1 has features {1, 2, 3, 4} with high probability and class 0 has features {5, 6, 7, 8} with high probability. Some overlap to make it non-trivial. WORKLOAD: Train: 200 samples per class (400 total). Test: 100 fresh samples per class. Majority baseline: always predicts the more frequent class (both 50% here so any constant prediction → 50% accuracy). NaiveBayes should hit >=80% accuracy by learning the feature distribution. MEASUREMENT: ACCURACY axis: NB correct count vs majority correct count.