Peritheos 0.15.0

Added

• Make native EOSMAT v4 the human-readable library format: one model name,
  adjacent coefficient values/errors/fixed flags, explicit units, consistent
  thermal blocks and EOS-first ordering. Preserve v2/v3 readers, lossless
  conversion and explicit v3 compatibility exports in Python and Rust.
• Add standalone authoring with create_eosmat() and python -m peritheos create,
  installed validation/formatting commands, a small complete template and
  optional file-local shared metadata. New user files remain unaudited and do
  not require database registration.
• Migrate all 232 bundled material documents to native v4. Add material summaries,
  a scientific-label glossary, scoped reconstruction assessments, published/refit
  coefficient comparisons and direct links to readable reconstruction reports.
  Available coefficient plots are comparisons, not independent residual tests.
• Add direct pressure-scale EOS views with route-specific pressure evaluation,
  volume inversion, thermal limits and provenance. Group 51 published alternatives
  under 27 source/model entries across 19 materials. Preserve every source record
  and expose reference variants through explicit listing/search options.
• Include the staged Brown (1999) NaCl historical-data reconstruction and independent
  2026 refit, with source transcriptions, uncertainty/normalization qualifications,
  diagnostic variants, figures and reproducible scripts. It is not a correction
  endorsed by the original author or a replacement pressure standard.
• Include the staged ferrous-ringwoodite investigation, source observations,
  separate static and thermal records, Brown-calibration hypotheses, diagnostics
  and figures. The unreproduced thermal model remains outside the bundled catalog.
• Extend Sakai (2025) Au, Fei (2007), Yokoo (2009), Dorfman (2012) and palladium
  source investigations with their recovered inputs, conditional reproductions,
  unresolved limitations and documented follow-ups.

Changed

• Retain only the three individually source-checked Brucite 300/473/673 K
  isotherms in executable catalog selection. Preserve the unresolved 873 K,
  continuous thermal models and all diagnostic refits in the audit archive.
  Published thermal-EOS parity has not been established.
• Keep recommendation/default holds explicit for unreproduced Fu CaSiO3, Baty
  and Frost experimental Pd, Sakai Fe-Ni, Shen–Smith MgO/NaCl-B2 and wadsleyite
  mappings. Preserve published coefficients and observations; missing experimental
  Pd defaults do not fall back to a separate 0 K DFT calculation.
• Make B1/B2 explicit in NaCl names and EOS labels, with phase and space-group
  metadata. Separate source verification, numerical parameter parity,
  author-workflow reproduction and recommendation for quantitative use.

Fixed

• Correct scientific audit sample columns, phase/medium selections, thermodynamic
  paths, ruby re-reductions and coupled spin-fit mappings while retaining source
  data. Compare Frost Pd published/refit pressure residuals at the same measured
  volumes and keep adjusted-volume residuals separately labelled.
• Preserve upstream 0.14.1 observation validation and export fixes, archive-local
  test fixtures, UTF-8 portability and calculated-fit replay tolerances.
• Preserve the original Brown source CSV bytes in Git archives so their recorded
  checksums remain verifiable after a fresh checkout.

Compatibility and scientific limits

• Require NumPy >=1.22.0 to avoid native OpenBLAS crashes with the old macOS
  ARM64 NumPy 1.21 wheels. Preserve SciPy >=1.9.3 and all scientific calculations.
  Exercise minimum dependencies on macOS as well as Linux, and include figure
  rendering dependencies in the development/test environment.
• Native v4 is an explicit format version; downstream v3 consumers can request
  compatibility exports. Standalone files need no database audit metadata.
• Research candidates retain their staged/diagnostic status. Passing software
  checks does not establish an original-author fit reproduction or physical
  validation. Published values are not replaced by independent refits.