Matrix Code Conjugacy asks whether two matrix subspaces are related by one simultaneous change of basis. A recent average-case algorithm reaches a fraction when the code dimension equals the matrix size, but a general code basis carries an additional unknown coefficient-space action. We bypass that action rather than recover it. A nonzero generator of a one-dimensional trace hull defines the homogeneous functionals . A transverse moment selects a nondegenerate complement of the hull, and trace duality turns the moments into basis-independent homogeneous matrices inside the code. The pair transforms only by ambient conjugation and a known scalar weight.
For every odd prime power , odd , and , we obtain a deterministic partial search-and-decision algorithm that is correct for at least a fraction of uniformly random -dimensional first codes, against every second input. Its bit complexity is ; the certified fraction is . The proof counts the actual correlated projection law of , including its endpoint atoms, and never models it as an independent random matrix. A direct corollary gives the same scale in the independent-uniform ordered-tuple model. The result excludes characteristic two and even , and it does not by itself yield a general Matrix Code Equivalence algorithm.