The generator transcribes published numerical determinations; it does not rerun a Monte Carlo simulation or a conformal-bootstrap optimisation. These estimates are not known to 100 significant digits. The source precision is retained, and every entry states whether it is transcribed or derived.
The four remaining leading components are calculated from $(\nu,\eta)$; for walks, $\alpha,\beta,\delta,\eta$ are calculated directly from $(\nu,\gamma)$ and $\omega=\Delta_1/\nu$. Exact rational arithmetic finds the extrema at the corners of each input box. Radii are rounded upward to two significant figures after allowing for rounding of the centre. This preserves numerical containment of the source error boxes, not a physical guarantee that the unknown exponents lie inside them.
The attached checks test a known two-dimensional Ising control and a deliberately inconsistent control, finite widths, all five scaling relations for every class, a common consistent point for each leading-exponent set, and all seven Ising relations with the published scaling dimensions. A separate comparison uses the Heisenberg bootstrap bounds in Table 1 of [16]. Independent literature comparisons cover the Ising Monte Carlo estimates in [10], all six leading XY and Heisenberg components in [11] and [12], and the XY correction exponent. For walks they cover $\nu$ and $\Delta_1$ in [13] and $\gamma$ and $\eta$ from the six-loop resummation in [14]. For percolation they cover the thermal and magnetic eigenvalues in [15]. The older Heisenberg $\beta$ comparison does not overlap and is recorded as a discrepancy; its physical cause is not established here. The Heisenberg and percolation correction exponents are transcribed from their primary sources, without an independent recomputation. The generator and these checks are also run on the stored entries.
