Symmetry-Driven Coherence Preservation in Open Quantum Systems: From a Negative Single-Qubit Specificity Audit to a Locked Two-Qubit Decoherence-Free-Subspace Benchmark

We evaluate symmetry-driven coherence restoration (SDCR) strictly within standard finitedimensional Gorini–Kossakowski–Sudarshan–Lindblad (GKSL) dynamics. Operationally, the term denotes suppression of future coherence decay; no reversal of already irreversible information loss is claimed. The study contains two locked computational stages. First, an independently reproducible single-qubit benchmark with equal Pauli rates is preserved as a negative specificity result: the declared target improves the basis-specific coherence area under the curve (AUC) from 1.2273 to 2.1617, yet loses to the norm-matched null (2.4638), channel permutation (3.1842), and every random-axis null. Second, we execute a FULL-profile two-qubit challenge on the logical sector HL = span{|01⟩,|10⟩} with exchange Hamiltonian HJ and collective/differential dephasing operators L± = (Z1 ± Z2)/√2. The target retains L+ and suppresses L−. Because PLL+PL = 0 and [HJ,PL] = 0, the target is an analytically exact decoherence-free-subspace positive control. The numerical implementation is nevertheless subjected to 512 equal-rank/equal-spectrum random nulls, 96 dissipative-gap-matched nulls, a 64-case paired bootstrap with 20,000 resamples, 96 held-out cases, Choi-positivity audits, global unitary covariance tests, and independent propagator cross-checks. The target coherence AUC is 10.0000, the random-null 95th percentile is 9.17915, the paired-bootstrap 95% interval for target minus the per-case best null is [0.33546,0.58186], and the held-out win fraction is 1.000. All ten core gates and four numerical audits pass. The admissible conclusion is limited: the declared selector is specific inside this synthetic standard-GKSL/DFS benchmark. The result does not establish HLV, Spiral-Time, FFGFT, new quantum dynamics, hardware performance, or experimental coherence restoration.