Scope
O30 studies a model whose data are assumed to lie on a Veronese cone in $\operatorname{Sym}^2(V_\rho)$. This is hypothesis (H0), not a consequence of the measured $6\to3$ collapse. Hypotheses (H1)–(H3) supply the inter-block identification, weight exchange and fixed projected symmetric square.
Under those assumptions, a Veronese root satisfies $\beta_j=\varepsilon_j\overline{\alpha_j}$ with $\varepsilon_j\in\{\pm1\}$. Both branches belong to the conditional model.
Exact spectral conditions
With the average taken over the full declared index set, the model covariance is diagonal with entries proportional to $(\tfrac14,\tfrac12,\tfrac14)$ if and only if
$$\langle\varepsilon r^4e^{2i\varphi}\rangle=0,\qquad \langle r^4e^{4i\varphi}\rangle=0.$$
The factor two in the diagonal entries is the square of the $\sqrt2$ coefficient in the symmetric-tensor basis. Diagonal entries alone are not eigenvalues when either off-diagonal cancellation fails.
Countermodels and zeros
Unweighted second-harmonic decorrelation is insufficient; the paper records explicit countermodels. Null points remain in the common $N_{\mathrm{total}}$ normalisation, with $\varepsilon_j:=+1$ harmlessly assigned because it is multiplied by $r_j^4$. Renormalising only over non-zero points rescales the whole covariance.