We develop a cross-scale consistency test of whether the late-time source of cosmic acceleration behaves as a cosmological constant within general relativity. Rather than testing the expansion history alone, the framework combines background expansion, linear matter growth, and Weyl lensing as three correlated closure conditions of the same physical hypothesis. The main novelty is the organization of these sectors into a single covariance-aware inference architecture: cross-sector correlations are retained after cosmological and nuisance-parameter marginalization, a conditional decomposition identifies whether a failure originates predominantly in the background or in the growth–lensing sector, and models specified at the action level can be evolved and projected into the same observable test. We validate the construction with a controlled Asimov calculation and show that injected dynamical-dark-energy and modified-gravity departures are correctly assigned to different physical sectors. As an action-level example, we study a restricted thawing branch of exponential quintessence; after profiling over the matter density and fluctuation amplitude, the benchmark gives = 0.91 for _Q = 0.30 and = 7.41 for _Q = 0.50, with a conventional Asimov reference crossing near _Q . These values are not observational constraints, and the result is a reproducible method for testing late-time cosmological-constant closure and for diagnosing which physical sector drives a statistically significant failure, with a direct route to future mock-calibrated multi-probe analyses.