Reproducibility kit for "Bond-dimension scaling of a local-refinement advantage over hyperoptimized tensor-network contraction on Sycamore-like topologies": data, headline contraction trees, and verification scripts.

This deposit accompanies a manuscript submitted to Quantum on a refiner that takes a contraction tree returned by a state-of-the-art hyper-optimizer and reduces its floating-point cost while preserving the contraction order's correctness. It contains the four CSV tables behind every figure in the paper (the 300-row χ-sweep that establishes refiner_fT ≤ hyper_fT across three topologies, the 300-row PLS-vs-logit dynamics ablation, and 63 cells of cost-model validation), 20 deposited contraction trees as .npz cells, the regenerated reference figures (PDF + PNG), and a small Python package whose python -m tnc_reproducibility verify command runs nine end-to-end self-checks (schema, headline monotonicity, predicted-vs-measured FLOP-ratio agreement, tree self-consistency to ≤ 10⁻¹³ bits, and figure regeneration) in a few seconds. A full algorithmic specification (SPEC.md) and the open-source cotengra baseline regeneration code are included; the CUDA/C++ implementation of the refiner is held back for a follow-up release. Code MPL-2.0, data CC-BY-4.0. © 2026 NeuroTechNet S.A.S.