# Argument selection for T659 This table stores one quantity: the ordered real roots of the Riemann xi Jensen polynomials in the coefficient convention of Griffin, Ono, Rolen, Thorner, Tripp and Wagner, arXiv:1910.01227v3, equations (1.1)–(1.2). The function is xi(s) = s(s−1) pi^(−s/2) Gamma(s/2) zeta(s)/2, and xi(1/2+z) = sum gamma_j z^(2j)/j!. The polynomial variable is X. Roots keep their negative signs and are ordered increasingly, counting multiplicity. All roots in the selected block are certified simple. Every free parameter is a genuine discrete index. There are no continuous couplings, sampled input values, or convention selectors. * **d, degree (discrete_index):** every degree from 1 to 8. Degree 1 gives the coefficient-ratio control, degree 2 gives the discriminant control, and degrees 3 through 8 provide higher-degree root geometry. Degree eight is also the endpoint of the small-degree hyperbolicity theorem in Griffin, Ono, Rolen and Zagier, arXiv:1902.07321v2, Theorem 2. This motivates a coherent breadth of degrees without requiring every individual polynomial to have a separate application. * **n, coefficient shift (discrete_index):** every shift from 0 to 20. Shift is an integer index of the central coefficient sequence, not a continuous argument rounded to integers. Starting at zero includes the unshifted family; the consecutive prefix shows early changes in the roots at each degree. The endpoint 20 is a noncanonical computational cutoff. This short prefix does not claim to probe the asymptotic Hermite limit. Extending only the shifts of a low-degree slice would lose useful degree breadth, while much higher degree would increase cancellation and root-isolation costs. * **k, real root index (discrete_index):** every k from 1 to d for each selected pair (d,n), in increasing numerical order. Selecting all roots is necessary to represent each polynomial's finite root geometry and to certify that no roots were missed. Negative roots are not replaced by magnitudes. The general definition counts multiplicity, while the complete selected block has simple roots. These independent degree and shift indices are the conventional two-index Jensen family in both sources, rather than a Cartesian product of unrelated simple real arguments. The finite rectangle is appropriate for this discrete family. Its cutoffs do not depend on decimal notation. They are neither a storage target nor an assertion that every member is individually famous. There are 168 polynomials and 756 root entries. The dry run measured 135.6 KB and a longest ball representation of 136 characters; every entry is written at 100 significant digits. The default enumeration is d outermost, then n, then k. The parameter domains remain d ≥ 1, n ≥ 0, and the available real root indices k ≥ 1, without defining the family by the current cutoffs. The completeness note specifies exactly d=1,…,8, n=0,…,20 and all k=1,…,d and explains the coverage choices. There is no second named quantity in a parameter. The family coverage and all eight existing tables named in numberdb-data issue 291 were read from the live database. T15 contains coefficient inputs, not these roots. Its warning about unproved interval wrapping precludes using it as a certified numerical dependency. The generator independently recomputes gamma_0,…,gamma_28. The 2019 paper's coefficients are eight times this sequence, a uniform factor that leaves roots unchanged. The two source conventions were checked directly, rather than inferred from an old proposal. Policy evidence: `agents/runs/20261101T195835Z-build-skills/manifest.json`. Both archived versions, `skill-local.md` and `skill-live.md`, were read and agree byte for byte, SHA-256 `a2b28d9953264d7b0ca62663785c2952190c5be6d6d3597fd695d9a54d606278`. The canonical source snapshot has the same digest. This build makes no independent critique acceptance; the controller must obtain a separate critique and its fingerprinted argument-review sidecar.