Einstein–Karahan Framework v128 — Expanded Discovery, Origin Audit, and Short-Range Gravity Test Architecture

This expanded v128 release presents the major post-v127 developments of the Einstein–Karahan Geometric Unification (EK-GU) framework, covering the Master Action V2 architecture, origin audits, weak-field and cosmological null tests, direct short-range-gravity constraints, and a prospective same-face torsion experiment with explicit systematic-error control. The V2 branch introduces a differential-weight effective action in which the Einstein–Hilbert and R² sectors acquire distinct modulus dependence, yielding the scalaron relation mχ² = F/(6G) while preserving explicit local stability conditions. A higher-gauge completion identifies a projective transgression and a conditional compact R/Z zero-mode structure, but the compact C3 field ontology itself remains an explicit global assumption rather than a source-native derivation. Post-freeze origin audits further show that the R² coefficient, vacuum unit operator, and cosmological scalaron state datum remain independent inputs; no dark-energy scale, dark-matter replacement, or cosmological coincidence is claimed as derived. The short-range-gravity branch reconstructs the fixed-α = 1/3 Yukawa profile from available Eöt-Wash source-level information. The external 4.4 meV benchmark remains allowed within the tested interpolation families, but no statistically significant laboratory detection is found. A prospective microscale torsion architecture is then developed around a buried m = 3 density pattern beneath an unchanged external science face. Finite-geometry integration, Fisher analysis, finite-element mechanics, seating, electrostatic patch-field transfer, and vibration/common-mode tribunals identify the principal failure modes and their quantitative control requirements. For the current consolidated simulated geometry, the 4.4 meV, α = 1/3 benchmark produces a projected Yukawa torque of approximately 3.94 × 10^-23 N m at 25 μm, with a projected two-sigma coupling threshold α_min ≈ 0.239. A four-axis common-mode and witness architecture reaches a simulated 10,000:1 end-to-end vibration rejection, leaving approximately 41% of the frozen coherent-bias budget. A frozen injected-signal virtual experiment subsequently recovers the benchmark signal and reaches an anytime-valid, sequential 5σ-equivalent likelihood threshold at day 121, while an independent null control produces no corresponding 5σ crossing within 2000 simulated days. These final virtual results validate the recovery and falsification pipeline under the stated numerical and systematic assumptions. They do not constitute experimental evidence for a new force or a 4.4 meV particle. The release therefore separates derived structure, conditional extensions, prospective experimental sensitivity, fail-closed results, and remaining source gaps explicitly. The accompanying public validation archive reproduces the promoted numerical claims while leaving the private candidate-generation and exploratory research workflow undisclosed.