We investigate the scaling properties of implicit deductive reasoning over Horn clauses in depth-bounded Transformers. By systematically decorrelating provability from spurious features and enforcing algorithmic alignment, we find that in sufficiently deep models with a bidirectional prefix mask, implicit reasoning approaches explicit CoT performance across graph topologies and problem widths, though CoT remains necessary for depth extrapolation.
@article{arxiv.2605.04330,
title = {The Scaling Properties of Implicit Deductive Reasoning in Transformers},
author = {Enrico Vompa and Tanel Tammet},
journal= {arXiv preprint arXiv:2605.04330},
year = {2026}
}