DATA AVAILABILITY — Paper 2
"When does a conservation–divergence spectrum find the specificity code? A prospective, two-dimensional criterion tested across seven protein families"

All data underlying the findings are within the manuscript and its supplementary datasets.

Primary data:
- Seven-family conservation–divergence master tables — per-position within-class conservation, between-class JSD, and anomaly score for serine proteases, protein kinases, GPCR coupling, class I HLA, nuclear-receptor DNA-binding domains, opsin spectral tuning, and class I aminoacyl-tRNA synthetases.
- Crosscutting-index table (dimension 1) — computed from functional labels and NCBI taxonomy only, never from the alignment columns under test.
- Pre-registration files — the frozen predicted verdict, crosscutting basis, and gold-standard residues (each verified by identity against a named reference UniProt sequence) for the two prospective families (aaRS, NR-LBD), committed before scoring.
- Cross-method benchmark table — mutual information, cumulative relative entropy (SDPpred-style), and evolutionary-trace-style scores on the same aligned matrices.
- Supplementary S1 — IPD-IMGT/HLA scale-up (4138 alleles) sample-size and grouping-label robustness.

Source data: sequences aligned to the relevant Pfam profile HMM; gold-standard residues mapped to Pfam match states with named reference structures (aaRS PDB 1FFY / UniProt P56690; NR-LBD PDB 1ERE / UniProt P03372). All source data are publicly available under the accessions listed in the supplementary tables.

Analysis code and pre-registration files (including the frozen, timestamped pre-registration records for the aaRS and nuclear-receptor-DBD prospective tests) are archived on Zenodo: https://doi.org/10.5281/zenodo.21449958 (shared with companion preprints P1 and P3, which use the same conservation-divergence spectrum instrument).

Companion preprints P1 (the instrument) and P3 (protein-language-model cross-examination) (Research Square, 2026; DOIs to be added on posting).