Because StyleMatch does not compare the passage with itself. The Style score is cosine similarity to an author profile aggregated across independent works. Only an embedding compared directly with the identical embedding should equal 1.00. The larger number beside a result is Affinity, not a probability of authorship; 0.92 does not mean “92% written by this author.”
Style matching and source identification are different tasks. The current ranking asks which author’s broad profile is nearest to the excerpt; it does not perform quotation lookup. Short excerpt boundaries can also move an embedding, while a single author profile compresses variation across novels, essays, speeches, and decades. Rank 1 therefore means “nearest profile under this model,” not “verified source author.”
The current index compresses each writer’s works, periods, and registers into one centroid. That average can erase an author’s less typical modes and favor broadly central profiles. High-dimensional embedding spaces also produce hub writers: profiles that sit near many otherwise unrelated passages and therefore recur in the top results. Repetition reflects this ranking geometry, not a hidden shortlist. The library is also broader than the live index: a writer needs usable original-language sources before entering honest comparison.