Calibrated torsion-balance spectra can be mapped to bounds on deterministic and stochastic departures from standard gravitational sourcing rules in the weak-field Newtonian regime. Using one-sided spectra, we formulate a linear calibrated angle-equivalent output-noise budget and the finite-time matched-filter/Cram'er–Rao bound for a known torque template. For stochastic models, the calibrated standard noise budget—thermal, Newtonian, environmental, imprecision, and backaction terms—is subtracted from the observed angle spectrum to define a residual spectrum, which is then converted to an equivalent residual torque spectrum using the calibrated torsional susceptibility. This residual provides an upper bound on any additional stationary stochastic torque noise. Only in the Markovian white-noise limit can this frequency-resolved bound be compressed to a single torque-diffusion coefficient ; colored or non-Markovian models must be compared with the full residual spectrum. Page–Geilker branch discrimination and Fedida–Kent mixture-equivalence tests address different physical sourcing questions, but after their competing predictions are projected onto torsion-balance torque templates they become the same statistical problem of matched-filter template discrimination. For a representative room-temperature Cavendish benchmark, the resonant thermal angle ASD is , whereas the measurement-added SQL is . For the torsion-balance search of Yan , the reported sensitivity at implies a conservative observed-floor bound when that public floor is interpreted as white torque noise. These expressions provide a common interface between calibrated torsion-balance data, deterministic template tests, and stochastic semiclassical-gravity searches without claiming a direct test of the full relativistic semiclassical Einstein equation.