Verification artefact package for "After the Euclidean Highway: Hyperbolic Expert AI as the Next Innovation"

This deposit contains the verification artefact package for the HySAT (Hyperbolic Structure-Aware Training) placement principle developed across six expert small-language-model training projects (17,954,911 cumulative training samples; approximately 317,000 cumulative optimizer steps; six complete training runs; zero NaN events). The package enables independent verification of the manuscript's load-bearing claims — Proposition 1 (HySAT stability), Proposition 2 (batch-size pair-formation regime), Proposition 3 (convergence rate), Lemma 1 (Stability-Dominant placement), the cross-project hyperbolic activation matrix (six SLMs by three loss components by batch regime), and bf16 Lorentz-manifold preservation at inner product minus one — without access to retained commercial-deployment model weights. Per the Option C hybrid-release framework, per-step numeric traces, evaluation aggregates, hyperparameter configurations, a matched four-arm ablation, a toy replication harness, and the manifold-preservation verification are openly released; commercial-deployment weights and proprietary training corpora are retained under a gated research-access protocol.