Thorium-232 versus Riemann zeros — run 2026-10-17 Download compare.py, n-090_Th_232.endf.gz and zeros-first2000.txt into one directory. Install numpy and scipy, then run: python compare.py The script reads the compressed evaluation directly. Compare output with results.json. th232-swave.csv contains the selected 242 energies in eV. Source: NNDC ENDF/B-VII.1 neutron library https://www.nndc.bnl.gov/endf-b7.1/zips/ENDF-B-VII.1-neutrons.zip Member: neutrons/n-090_Th_232.endf; MAT 9040, MF2 MT151, LRF3. The compressed download preserves the full evaluation byte-for-byte upon decompression. Its uncompressed SHA-256 is in results.json. Zeros: Odlyzko's zeros1 table, using the existing project's first-2000 copy: https://www-users.cse.umn.edu/~odlyzko/zeta_tables/zeros1 https://github.com/kwnorton000-hash/spectral-verification Selection: first resolved region (up to 4000 eV), l=0, abs(J)=0.5; strictly positive energies only. Target Th-232 spin 0 gives J=1/2+. Negative-energy/external poles, p-wave blocks, and unresolved averages are excluded. Linear staircase unfolding and mean-normalized nearest-neighbor spacings match the prior uranium pipeline. GUE distances use its grid integral, with an exact analytical-CDF sensitivity check in the output. Nominal KS classifications: thorium passes the GOE threshold and exceeds GUE/Poisson thresholds; zeros pass GUE and exceed GOE/Poisson thresholds. Two-sample D=0.1410788382, nominal iid p=0.0164153985. These are descriptive evaluation-level diagnostics, NOT calibrated experimental significance or independent proof of a universality class. Neighboring spacings are correlated, unfolding is fitted, gaps normalized, and resonance completeness/spin assignments can affect the result. The evaluation explicitly describes inserted unseen p-wave resonances to match Wigner/Porter-Thomas distributions; these p-wave blocks were excluded. The present s-wave selection is still not an unbiased experimental sample. Calibration would require ensembles repeating selection and unfolding, plus window/completeness sensitivity checks. Long-range statistics remain pending. No predictive zero-to-nucleus mapping or energy application follows.