Biorthogonal eigenvectors of the Holte carry matrix and cascade-free enumeration

For $k$-summand base-$N$ addition, the carry process is a Markov chain on $\{0,\ldots,k-1\}$ whose transition matrix--the Holte matrix $T$--has eigenvalues $\{N^{-j}\}_{j=0}^{k-1}$, all simple and independent of $N$. We give the complete biorthogonal eigenvector system. The left eigenvectors factor as $\sum_i u_j[i] x^i = c_{k,j} (x-1)^j A_{k-j}(x)$, where $c_{k,j} = |s(k,k-j)|/k!$ involves unsigned Stirling numbers and $A_n(x)$ is the Eulerian polynomial. The right eigenvectors satisfy $\sum_i \binom{k-1}{i} v_j[i] x^i = (1+x)^{k-1-j} Q_j(x)$, where the quotient polynomials $Q_j$ have palindrome symmetry $x^j Q_j(1/x) = (-1)^j Q_j(x)$ and converge to $(1-x)^j$ as $k \to \infty$; for $j \le 3$, we give explicit closed forms in terms of $k$. The cascade-free avoidance count satisfies $a(L) = (\sqrt{d})^L U_L(x)$ (Chebyshev polynomial of the second kind) whenever the restricted transfer matrix has dimension $d \le 2$; we prove this is sharp: for $k$-summand addition, Chebyshev form holds for $k = 3$ and fails for $k \ge 4$. The proof uses oscillatory matrix theory to establish non-vanishing of all spectral residues. The characteristic polynomial of the restricted transfer matrix is determined in closed form by a Stirling-weighted Lagrange interpolation at the Holte eigenvalues. Two systems with binary carry state spaces are shadow-equivalent if and only if they share the pair $(N, d)$. The general classification for $k$-state systems reduces to the characteristic polynomial of $T$.