Peritheos 0.17.3

Added

• Add six published cementite EOS records: Li (2002) and Ono–Mibe (2010)
  experimental BM3, both Takahashi (2019) nonmagnetic BM3–MGD pressure-scale
  variants, Sata (2010) nonmagnetic BM3, and Hu (2019)'s six-shot principal
  Hugoniot through 245 GPa. Preserve original coefficients, errors, measured
  and adopted reference volumes, source observations and calibration lineage.
  Keep calculated shock temperatures, porous precursor normalization, possible
  high-pressure melting and unresolved author workflows explicit.
• Bundle the recovered Bykova (2018) Coesite fit export and Prescher (2015)
  Fe7C3 author inputs separately from earlier tabulated-data diagnostics.
  Coesite's 22 inputs contain 21 measured entries and a derived ambient
  anchor; its literal-DAT and workbook-precision reconstructions remain
  distinct. Fe7C3's 28-row numerical replay reproduces the six saved
  coefficient/error values at printed precision; Table S4 remains preserved
  under its provenance-only use restriction.
• Add a source-by-source domain review with checks against preserved article
  tables, official supplements and author artifacts. Keep unavailable original
  workbook/instrument-file access and unauthenticated correspondence explicit.
  Add an independent scientific-value/source-array fingerprint alongside the
  existing historical release and merge-metadata proofs.

Fixed

• Retain Scott (2001)'s scoped similarity finding and correct current Prescher
  (2012) summaries: the primary comparison uses 19 numbered Table S1 states,
  agrees in central coefficients, and does not reproduce the K0 error. The
  earlier 22-marker comparison and split fits remain historical audit evidence;
  only the full-range published Prescher record is installed.
• Update source/refit ledgers, observation views, reconstruction reports and
  discovery inventories for 234 documents and 685 records: 231 executable
  materials, 630 source choices and 682 records including reference variants.
  Preserve all 679 pre-existing source coefficient sets and original datasets.
• Standardize material display names while retaining historical search aliases.
  Preserve conditional uncertainty findings, source-use restrictions and
  unresolved investigations, including the wadsleyite source-model gap.
• Reload the explicitly written Brown NaCl refit instead of a cached artifact;
  retain checksum-bound CSV bytes and the field order of unchanged assessments.

Development and compatibility

• Write integrated research documents and reports as UTF-8 bytes with stable
  line endings. Restrict replay tolerances to named computed fields; keep
  source readings, fixed normalization and acceptance thresholds exact, with
  mutation tests and write/read checks under a simulated Windows ANSI locale.

• Compare recalculated cementite diagnostics within explicit numerical bounds
  across Python/BLAS builds while keeping source coefficients, fixed values,
  observation counts and scientific decisions exact. Preserve saved reports
  and add mutation checks proving source changes cannot pass as refit roundoff.

• Decode research JSON as UTF-8, serialize report paths with forward slashes,
  and preserve the author Coesite DAT bytes on Windows without changing source
  fingerprints, observations or stored reports.

• Match the minimum-dependency CI plotting parser to the declared development
  floor; retain the NumPy 1.22.0 and SciPy 1.9.3 library dependency checks.

• Use complex-step derivatives for the existing Sata rounding-box diagnostic
  so all three starts converge on the minimum SciPy stack, retaining its
  objective, bounds, convergence/feasibility checks and archived findings.

• Incorporate Roland Mas's explicit sampler f64 annotation and allow 1e-9
  absolute variation when replaying the saved brucite optimizer objective across
  numerical backends. Preserve the stricter residual-consistency check, source
  inputs and scientific findings. Keep Rust 1.83 support and existing dependencies.

• Ship an installed code/data licensing guide, dataset notice index and complete
  Creative Commons texts in both distribution formats. Replace the Python
  distribution's blanket MIT metadata with an explicit component-scope reference;
  library code remains MIT and unspecified third-party source terms stay explicit.

• Separate routine CI from the complete release compatibility matrix and share
  unchanged numerical audit results through isolated test copies. Preserve
  fresh calculations for mutation, solver-patching and repeatability tests.

• Keep model equations, public APIs and published EOS coefficients unchanged. Source
  checks and conditional numerical agreement do not establish physical
  validation or complete reproduction of an author's original workflow.

• The finite-reference EOS, pressure-scale views and native EOSMAT interfaces
  are unchanged from their earlier releases. Locally preserved research backups
  and publisher full texts remain separate from curated package evidence.