Rank-Turbulence Delta and Interpretable Approaches to Stylometric Delta Metrics

The deeper contribution isn't just two new metrics but the token-level decomposition that lets researchers audit which specific words drove an attribution decision, a property classical Burrows's Delta has always lacked despite decades of use in forensic literary scholarship.

This sits largely disconnected from the recent LLM-focused coverage on Modelwire, though the interpretability thread is a genuine point of contact. The ORCA framework covered in 'Structural interpretability in SVMs with truncated orthogonal polynomial kernels' (April 16) pursues a similar goal in a different domain: making a classifier's internal reasoning legible without retraining it. Both papers are responding to the same pressure in applied machine learning, namely that a correct answer without an auditable rationale is increasingly insufficient for scholarly or high-stakes use. Stylometry is a field where attribution decisions carry real consequences, from literary fraud disputes to legal proceedings, so interpretability isn't a nice-to-have feature bolted on for optics.

The meaningful test is whether these metrics hold attribution accuracy on shorter, noisier texts, such as social media or disputed historical documents, where classical Delta degrades badly. If the authors or independent replicators publish results on sub-1000-word samples within the next year, that would signal genuine practical reach beyond controlled literary corpora.