# With the attached generate.py in the current directory:
from generate import SpecificHeats
g = SpecificHeats()
g.value({'m': 6, 'n': 6}, 100)
list(g.enumerate()) # all rectangles 2 <= m <= n <= 6Exact integer cut counts are constructed by row transfer. The logarithm in $K_c$ and the partition-function moments are evaluated with real ball arithmetic [3] at 429 bits, including 96 guard bits. Every entry retains at least 416 bits of relative accuracy, about 125 decimal digits, before serialization to 100 significant digits. The generator recomputes every count and does not use the attached numerical outputs as inputs.
The independent checks enumerate the cycle space of each rectangle to construct its even-subgraph polynomial, then recover every coefficient of the zero-field cut polynomial by the high-temperature expansion. Bond-energy variances are recomputed exactly in $\mathbb Q(\sqrt2)$ and combined with a 600-bit ball logarithm. Each resulting enclosure lies strictly inside the interval represented by the stored decimal. The logarithmic-derivative formula is also evaluated independently of the generator. Direct enumeration checks the full two-variable cut counts for all nine rectangles with at most 16 spins. For every rectangle, separate two-variable transfer checks spin reversal and the specializations at $b=0$ and $b=1$; additional checks verify rotational symmetry and the zero-coupling mean and variance of $B$. The single bond and the four-cycle give known-answer controls. Reproducible checks, exact counting data and numerical evidence are attached.