CHiPS achieves 93% authorship accuracy without transformers or tokenization

CHiPS (Character Histograms and Positional Signals) introduces two complementary fingerprints of writing style: CH-SVM, a character-histogram classifier based on one-character marginal distributions, and FFT12-LR, a positional-signal classifier that converts selected characters and punctuation into impulse trains and extracts Fourier/Welch spectral descriptors. A leak-safe decision-level fusion variant (CHiPS-F) and an optional top-5 listwise reranker are also presented. The method sidesteps all heavy NLP—no tokenization, syntactic parsing, pretrained language models, or transformer fine-tuning—and avoids character bigrams or longer n-grams in the histogram component, ensuring a transparent, restricted feature set.

On a locked grouped ROST split (400 files from 392 source-text groups, 10 authors, source-text-level evaluation with grouped 5-fold model selection), CHiPS-F achieves 0.9310 accuracy and 0.9341 macro-F1. A matched unrestricted TF-IDF SVM using character 2-5-grams reaches 1.0000 accuracy on the same data, confirming the authors do not claim state-of-the-art performance; instead, the work explores how far restricted, transparent character evidence can go under strict leakage control. On a larger ROSTories-cleaned corpus (1,248 files, 1,240 source-text groups, 19 authors), CHiPS-R achieves 0.8919 accuracy and 0.8708 macro-F1.