We present Hvala, a linear-time ensemble approximation algorithm for the Minimum Vertex Cover problem. Hvala combines a maximal-matching 2-approximation, a bucket-queue maximum-degree greedy method, and the degree-1 weighted-reduction Hallelujah heuristic of a companion work. It applies redundant-vertex pruning to each candidate cover and returns the smallest resulting cover. We prove unconditionally that Hvala runs in O(n+m) time and space and attains worst-case approximation ratio at most 2 on every finite simple graph. Conditional on the strict pointwise inequality of the Hallelujah heuristic, the bound improves to |S| < 2·OPT(G) on every graph. Empirically, we evaluate Hvala on three independent studies totaling 252 instances: the 109-instance NPBench collection, 130 real-world graphs from the Network Data Repository, and 13 hand-crafted adversarial graphs, including crown graphs, odd cycles, high-girth cubic graphs, and bipartite-expander graphs. On the 173 instances with published reference values, the mean approximation ratio is 1.015, the maximum is 1.192, and every reported ratio falls below 1.414. The benchmark raises a natural open question, not proved here: whether Hvala admits a uniform sqrt(2)-epsilon bound on broad restricted graph classes, such as bounded-degree, bounded-treewidth, bounded-clique-number, expander-like, or power-law graph families. An all-graphs sub-sqrt(2) bound would contradict P != NP by the Khot–Minzer–Safra hardness barrier. The algorithm is publicly available through the hvala Python package.