peritheos 0.7.0__tar.gz → 0.9.0__tar.gz

peritheos-0.9.0/CHANGELOG.md

@@ -0,0 +1,849 @@
+# Changelog
+
+All notable changes to Peritheos are documented here. The project follows
+[Semantic Versioning](https://semver.org/).
+
+## [Unreleased]
+
+## [0.9.0] - 2026-09-12
+
+This release adds Birch–Murnaghan references to the double-Debye Helmholtz
+models and optional material families for catalog browsing. Existing material
+and EOS identifiers, numerical records, and flat discovery remain unchanged.
+
+### Added
+
+- BM2, BM3, and BM4 references for both double-Debye Helmholtz models in
+  Python, Rust, and `.eosmat` loading, with reference energies consistent
+  with pressure in both absolute and reference-temperature modes.
+- Optional material family metadata and curated bridgmanite, Mg–Fe monoxide,
+  post-perovskite, phase D, NaAlSiO4 calcium-ferrite type, magnesite, olivine,
+  and garnet families. Python provides family lookup, exact family
+  filters, and grouping of catalog/search results with standalone fallback.
+  Existing material identifiers, flat discovery, and EOS records are preserved.
+  Python and Rust validate and round-trip the optional `.eosmat` `family_id`;
+  Rust exposes `Material::family_id()`.
+
+### Compatibility
+
+- Python's existing Vinet double-Debye constructors remain supported, and
+  `.eosmat` stays at format 3. Older readers cannot load the newly supported
+  Birch–Murnaghan/double-Debye model combinations.
+- Rust double-Debye types now accept a reference-EOS type parameter, defaulting
+  to `Vinet`. The two corresponding `ThermalModel` variants hold models with
+  `IsothermalModel` references; callers constructing or destructuring those
+  variants directly must account for that type change.
+
+## [0.8.0] - 2026-09-11
+
+This release expands the model and dataset APIs and revises
+the scientific catalog. It contains 210 material documents, 209 executable
+materials, and 573 primary-source-validated EOS records. Source validation
+checks the published parameterization; it does not imply independent refit
+parity or a recommendation for every quantitative application.
+
+### Added
+
+- Typed observation loading through `Material.get_dataset()`, `Dataset`,
+  `DatasetColumn`, `DatasetResource`, and `PressureVolumeData`. Embedded and
+  packaged CSV data expose read-only arrays, verified resource checksums,
+  column metadata, unit conversion, and pressure-volume views with reported
+  uncertainties. `DatasetError` and `DatasetLookupError` provide typed failures;
+  the existing raw `Material.datasets` mappings remain available.
+- `EulerianFiniteStrainAcoustic`, `fit_acoustic_finite_strain()`, and
+  `AcousticFitResult` for joint density/P-wave/S-wave velocity fitting,
+  including optional density errors, bounds, robust losses, covariance,
+  adjusted observations, and JSON-safe diagnostics.
+- Python and Rust `Vinet3` and the Fratanduono et al. (2020) 298 K Cu reference
+  record, completing the implementation described prematurely in the 0.7.0
+  notes. Same-run Cu volumes now support Shen–Smith pressure reconstruction;
+  reduced isentropic states are not treated as independent 298 K fit data.
+- Thermal model APIs and `.eosmat` registrations for `Dewaele2006`,
+  `DebyeQuadraticThermalPressure`, `HollandPowellThermalPressure`,
+  `AsymptoticPowerLawMieGruneisenDebyeExcess`, and
+  `SoundVelocityDebyeHelmholtz`. These cover single-Debye iron scales with
+  anharmonic/electronic terms, Fei's quadratic Debye pressure, a composable
+  Holland–Powell thermal term, a quadratic excess free energy, and a
+  sound-velocity-derived Debye model. Native implementations support the first
+  four; the sound-velocity model provides Python thermodynamic and shock-state
+  calculations. `Dewaele2006` includes the exact `gamma_inf=0` limit.
+- `MieGruneisenDebye` reference-isentrope subtraction and optional `Cvmax`,
+  with native support for the BM3 shock-static construction. `Tange2009Debye`
+  exposes `constrained_reference_parameters()` for the source's coupled ambient
+  elastic and heat-capacity constraints.
+- Cubic compressibility (`bulk_modulus_law="reciprocal_cubic"`) and a
+  logarithmic temperature correction to `K0_prime` in `ThermalReferenceStateEOS`,
+  including native evaluation and fitting. The existing linear modulus law
+  remains the default; these laws define a mechanical P-V-T surface.
+- Complete caloric Helmholtz observables for `DorogokupetsOganov2007`, and
+  ionic, anharmonic, and total entropy/internal energy for both double-Debye
+  Helmholtz models, consistent with their reference-isotherm subtractions.
+- The executable `ruby_sokolova_2013` calibration in Python and Rust, with
+  explicit provenance for the Sokolova pressure-scale reconstruction.
+- Primary-source Fe, Fe0.9Ni0.1, and NaCl-B2 records from Brown (2000),
+  Dubrovinsky (2000), Dewaele (2006), Yamazaki (2012), and Sakai (2014): 43
+  published records and 348 observations, with qualified regression audits.
+  Fei et al. (2016) adds two 300 K hcp-Fe BM3 alternatives, its thermal pressure
+  surface, and all 96 supplementary observations; remaining source and partial
+  refit discrepancies are documented.
+- Hirose et al. (2008) Au: the 300 K isotherm, both thermal fits, and all 21
+  paired Au/MgO observations. Sakai et al. (2018) adds the hcp-Re Vinet scale
+  on Yokoo Pt, all 58 Figure 10 vector markers, and a 26-point RP01 refit with
+  coefficient and uncertainty parity; alternate selections remain diagnostics.
+- Source-owned MgSiO3 records from Wang (2004), Zhou (2014), and Komabayashi
+  (2008), MgAl2O4 records from Irifune (2002), Ono (2006), and Sueda (2009),
+  and high-/low-spin ferropericlase records from Fei (2007) and Marquardt
+  (2009). Source-fit alternatives, published errors, staged protocols, and
+  independently derived refits are distinguished in their provenance.
+- Separate Ye et al. (2017) Au/Pt/MgO 300 K standards and Zhu et al. (2025)
+  thermal P-V-T records. The corrected Ye supplements reproduce the printed
+  fits and errors; Zhu's released v3 optimizer/property coefficients and
+  bundled thermal inputs support the thermal reconstruction. Differences
+  from the preprint and standalone calculators remain explicit.
+- Fei et al. (2004) Au/Pt pressure standards, a separately identified Fu
+  CaSiO3 literature-data refit, and Kuznetsov (2002) fcc/hcp Pb reference
+  isotherms. The Pb isotherms do not establish a refit of the unpublished
+  full thermal observation set.
+
+### Corrected
+
+- Chen et al. (2018) tetragonal CaSiO3 now uses the source-stated Vinet EOS,
+  replacing `ca_perovskite_tetragonal_chen_2018_bm2` with
+  `ca_perovskite_tetragonal_chen_2018_vinet`. All seven table rows, the Z=4
+  volume normalization, two-sigma errors, and Pt calibration are preserved.
+  Numerical agreement remains qualified because source weights and pressure
+  errors are unavailable; no separate refit record is introduced.
+- Zhao et al. (1997) jadeite now includes its complete 300–1280 K thermal
+  relation and the fitted reference coefficient `V0=403 ų`, replacing the
+  measured `403.32(8) ų` table value. The 31 hydrostatic states and Decker
+  NaCl lineage support a qualified refit with unresolved source weighting.
+- Restored the complete Fischer et al. (2011) B1/B8 FeO thermal relations,
+  Anzellini et al. (2025) Ir BM3 plus Holland–Powell thermal EOS, and
+  Mosenfelder et al. (2009) MgSiO3 post-perovskite shock-static EOS. Ir uses
+  the mean pyrometry temperature for its 122 supplementary observations;
+  Mosenfelder's reconstruction uses 48 static states and six PPv shock states.
+- Holmes et al. (1989) Pt now maps the exact published universal-equation
+  coefficients to Vinet parameters instead of using their rounded elastic
+  interpretations. Its approximate thermal extension still uses the published
+  rounded modulus; the seven shock observations qualify the curve rather than
+  independently reproducing the unavailable theoretical fit.
+- The Au/Pt/MgO `*_zhu_2025_vinet_300k` names now resolve as aliases of the
+  source-correct `*_ye_2017_vinet_300k` identifiers. Zhu's thermal models have
+  separate `*_zhu_2025_pvt` identifiers.
+- Source-only cards are skipped by executable catalog enumeration. Coesite-V
+  remains accessible through the document API, but cannot be constructed as
+  a `Material` without an executable EOS.
+
+### Scientific evidence and limitations
+
+- Consolidated recovered source tables, digitizations, deterministic audits,
+  and record-level refit results. These include Sun (2016/2022) and Noguchi
+  (2013) CaSiO3, Li (2006) acoustic MgO, Noguchi (1999) shock-reduced NiO,
+  Redfern magnesite, Datchi diamond, and MgSiO3 staged fits. Akins (2004)
+  liquid MgSiO3 retains a qualified `similar` source-equation check from three
+  reduced shock states; it does not establish strict fit parity. Published values
+  remain distinct from diagnostic refits. Noguchi's subscription-only table
+  is represented by a row-free reproducibility artifact and local-input script.
+- Dewaele (2019) metal evidence is consolidated without claiming completion of
+  all 30 records. The common ledger classifies 24 records as `similar` using
+  two-ruby-scale refits within the printed 95% intervals, without inferred
+  refit covariance. Recovered Fe EPAPS observations additionally support both
+  scale reductions in the dedicated audit, although the common ledger still
+  reports discrepancies for Fe under its different fitting path.
+- All 41 Lazicki Be helium-run rows are bundled, including the version-of-record
+  `a=2.162 Å` at 24.6 GPa and the superseded manuscript token `2.262 Å` in
+  provenance. The two Be records nevertheless remain `not_refittable` in the
+  current common ledger, and the dedicated audit has not integrated this
+  recovery. The earlier claim that 26 records were linked to completed refits
+  is therefore not the released validation state; neither is the old claim
+  that the Be table is unavailable.
+- hcp-Pb retains 17 tabulated Dewaele rows, 21 digitized Kuznetsov Figure 3
+  markers (seven room-temperature and 14 thermally reduced), and a separate
+  nine-marker selection recovered from Dewaele Figure 2. The 24-point
+  room-temperature combination, 38-point all-marker combination, and 26-point
+  Figure 2 combination are distinct diagnostics. The current common ledger
+  evaluates the 21 Kuznetsov markers alone and reports `parity_not_achieved`;
+  the dedicated metal audit fits the 17 Dewaele rows alone. No combined
+  selection establishes the unpublished original rows, weights, pressure-scale
+  handling, or full-fit parity. Published Pb coefficients are preserved.
+- Anderson Au, Dorogokupets–Oganov Pt, Sokolova's global calibration, and
+  Tange MgO retain qualified component/partial reconstructions rather than
+  claims of complete global-refit parity. The Luo MgO full refit remains
+  `not_refittable`; derived diamond composites use `source_reconstruction`.
+  Funamori (1998) transformation-derived MgAl2O4 curves remain executable
+  source constructions with no statistically independent regression to rerun.
+- Sun (2019) liquid FeSiO3 audits distinguish the 40 liquid and six nonliquid
+  table states and digitized thermal-fit inputs from an exact numerical source
+  regression. Ismailova Fe0.88SiO3, Cohen–Lin FeSiO3, and Dorogokupets (2015)
+  MgSiO3 retain explicit missing-input or fit-protocol limitations. Recovered
+  checkpoints and curve agreement do not establish unavailable source data.
+
+See the [record-level refit ledger](docs/primary-eos-refits.md) and
+[paper investigation ledger](docs/paper-investigation-ledger.md) for the
+qualified outcomes and links to individual reproductions.
+
+### Removed and compatibility
+
+- Removed 92 Delta-project benchmark records (50 code-specific PBE curves and
+  42 experimental-reference comparison curves), their benchmark-only cards,
+  datasets, and import/reproduction tooling from the production distribution.
+- Removed 220 Sun et al. (2010) universal-EOS benchmark parameterizations and
+  49 phase-unresolved cards. The numerical source table, Morse/Sun-Morse
+  implementations, and equation tests remain benchmark fixtures; the
+  independently refitted AIP Handbook Cu isotherm retains its own card.
+- Removed 14 Caracas et al. (2005) exploratory CaSiO3 tilt-branch records
+  without diffraction-ready structures. The full nine-structure coefficient
+  table remains an audit fixture; production retains the Pm-3m and I4/mcm pairs.
+- Removed the Chantel (2012) composite bridgmanite EOS and Schoelmerich (2020)
+  shock-corrected stishovite EOS from production, and demoted the one-anchor
+  coesite-V reconstruction to non-executable provenance. Their available source
+  evidence remains; Schoelmerich's fitted-curve digitization is removed because
+  it cannot independently reproduce the unpublished shock correction.
+- Removed identifiers and benchmark resources are no longer available through
+  catalog lookup or package-resource access. Callers must update the Chen
+  identifier and account for the distinction between material documents and
+  executable materials. Ye/Zhu's former isotherm identifiers remain aliases.
+- `.eosmat` stays at format 3, with additional model discriminators and thermal
+  options; older consumers may not construct those models. A `refit` record
+  still requires `fit_provenance`, but may now represent a source-data refit
+  without `derived_from_record`. Existing default thermal conventions and
+  raw dataset mappings are preserved. Rust callers must account for added
+  model/reference enum variants in exhaustive matches and new fields when
+  constructing affected thermal structs directly. The ruby calibration array
+  now has eight entries, which affects fixed seven-element array types.
+
+## [0.7.0] - 2026-09-07
+
+### Added
+
+- Added the primary-source-validated Fratanduono (2020) 298 K Cu record and
+  independent `Vinet3` equation, schema and exports. Shen–Smith
+  same-run pressure reconstruction now reuses this record; the reduced
+  isentrope is not treated as a 298 K refit dataset.
+
+- Added a 200-record experimental-metal EOS tranche, the Baonza pseudospinodal
+  EOS with native evaluation and fitting support, and the phase-specific hcp-Pb
+  material card. The expanded catalog now contains 286 materials and 813
+  primary-source-validated EOS records.
+- Added a 200-record experimental-metal EOS tranche, the Baonza pseudospinodal
+  EOS with native evaluation and fitting support, and the phase-specific hcp-Pb
+  material card. The expanded catalog now contains 286 materials and 812
+  primary-source-validated EOS records.
+- Added a 200-record experimental-metal EOS tranche, the Baonza pseudospinodal
+  EOS with native evaluation and fitting support, and the phase-specific hcp-Pb
+  material card. The expanded catalog now contains 286 materials and 812
+  primary-source-validated EOS records.
+- Added a 200-record experimental-metal EOS tranche, the Baonza pseudospinodal
+  EOS with native evaluation and fitting support, and the phase-specific hcp-Pb
+  material card. The expanded catalog now contains 286 materials and 812
+  primary-source-validated EOS records.
+- Added a 200-record experimental-metal EOS tranche, the Baonza pseudospinodal
+  EOS with native evaluation and fitting support, and the phase-specific hcp-Pb
+  material card. The expanded catalog now contains 286 materials and 812
+  primary-source-validated EOS records.
+- Added seven primary-source-validated LitCurate EOS records from four papers:
+  two Kubo et al. (2000) Mg0.9Al0.2Si0.9O3 bridgmanite BM3 fits, two Kubo et
+  al. (2006) MgGeO3 post-perovskite sensitivity fits, two Matsui et al. (2012)
+  high-spin ferropericlase BM3-MGD models, and the Koemets et al. (2023) FA50
+  Pnma bridgmanite BM2. Complete source tables, phase-specific fit masks,
+  deterministic reproductions, pressure-scale provenance, and explicit
+  accept/hold/reject decisions for every same-paper LitCurate row are bundled.
+- Added a generated paper-level investigation ledger covering every audited
+  primary source, including coefficient discrepancies, unavailable direct
+  refits, and papers withheld or deferred without a production EOS record.
+
+- Added a primary-source-validated absolute-zero Vinet-MGD EOS for cubic boron
+  nitride from Datchi et al. (2007). The explicit absolute thermal-pressure
+  baseline reproduces all 66 Table IV states at 0.584 GPa RMSE and the paper's
+  independent 300 K zero-pressure volume.
+- Added Suzuki's (2016) epsilon-FeOOH BM3 thermal EOS with an explicit
+  reference-temperature linear-expansivity convention, all 33 Table 1 P-V-T
+  observations, and a refit recovering every free coefficient within 0.40
+  published standard deviations.
+- Added Noguchi et al.'s (2013) non-default 700 K BM2-MGD CaSiO3 perovskite
+  EOS, preserving its cubic-phase domain, Fei et al. Pt calibration, and
+  staged-refit parity without redistributing the subscription-only table. A
+  row-free reproducibility artifact now records normalized source-table hashes,
+  all four published fit branches, explicit broad-peak exclusions, and
+  independent Fei/Holmes Pt-scale diagnostics; users with lawful local access
+  can rerun the complete audit through the reproduction script.
+- Audited Katsura et al.'s (2004) 127-row ringwoodite dataset and documented
+  why its thermal EOS remains excluded: the published coefficients cannot be
+  reproduced with the chemically required seven-atom normalization.
+- Added Fu et al. (2024) BM2 and BM3 records for Fe-Al-bearing bridgmanite,
+  with the complete 22-row primary compression dataset, Au pressure-scale
+  provenance, independent reproduction, and record-specific tests.
+- Added phase-restricted principal Hugoniots for B1 MgO and B1 NiO. The MgO
+  relation preserves the published coefficients and uncertainties; the NiO
+  relation is a documented OLS derivation from all eight published final-state
+  observations with the elastic precursor excluded.
+- Added primary-source-validated Sun et al. (2016, 2022) CaSiO3-perovskite
+  records: a 300 K tetragonal I4/mcm BM3 isotherm with the published Z=1
+  volume converted to the conventional Z=4 cell, and a high-temperature cubic
+  Pm-3m BM3-Mie-Gruneisen-Debye EOS with explicit metastable-reference-state,
+  phase-validity, and Fei et al. Pt pressure-scale metadata.
+- Added the Luo et al. (2023) B1-MgO pressure scale as a 0 K Vinet curve plus
+  the published absolute second-order thermal-pressure polynomial, with all
+  five new shock states and the complete 576-value Tables II--III P-V-T grid.
+- Added the primary-source-audited Dewaele et al. (2000) B1-MgO BM3-MGD EOS
+  and all 61 P-V-T observations from its Table 2. The record preserves the
+  Jamieson platinum pressure basis, distinguishes adopted, fitted, and fixed
+  quantities, reproduces the paper's 145 GPa BM3 benchmark at `V/V0=0.667`,
+  and recovers `q` within combined two-sigma uncertainty from the 41 new heated
+  rows with `gamma0` fixed as in the source's staged procedure. The paper's
+  Murnaghan, Vinet, logarithmic, and alternate
+  BM3 rows are documented as comparison fits rather than duplicate EOS records.
+- Added a primary-source-validated Vinet--Mie--Gruneisen--Debye EOS for cubic
+  CaSiO3 perovskite from Kawai and Tsuchiya (2014), together with all 60
+  published Table 1 isochor benchmarks, explicit computational provenance,
+  source inconsistencies, and a documented raw-data/refit limitation.
+- Added the Wang et al. (2012) stishovite Vinet-Mie--Gruneisen--Debye EOS and
+  all 56 Table 1 P-V-T observations with simultaneous Au calibrant volumes and
+  printed uncertainties. The source's equally successful BM3 alternative,
+  300 K thermal reference correction, Tsuchiya (2003) Au calibration lineage,
+  numerical Table 3 checks, and an errors-in-variables parity refit are fully
+  documented.
+- Added the two distinct Reynard et al. (1996) MgSiO3-akimotoite BM3 fits on
+  the ruby-fluorescence and preferred ice-VII pressure assignments, together
+  with all 16 Table 1 unit-cell observations, paired ice-VII calibrant volumes,
+  pressure-scale provenance, independent high-pressure reproductions, and
+  parity-checked refits with the published 212 GPa bulk modulus held fixed.
+- Added the Schulze et al. (2018) Phase Egg single-crystal BM3 EOS and all 16
+  printed compression rows, with the excluded ambient observation flagged and
+  the 15-row fit independently reproduced. The record distinguishes the
+  analyzed `Al0.98Si0.92H1.39O4` composition from the ideal `AlSiO3OH`
+  structural model, traces the hydrogen coordinate to Schmidt et al. (1998),
+  and explains why the paper's Vanpeteghem-data refit is not a second EOS for
+  the 2018 specimen. The Dewaele et al. (2008) ruby calibration used by the
+  experiment is now executable and linked explicitly.
+- Added one primary-sourced static 0 K Phase Egg LP BM3 record from Mookherjee
+  et al. (2019), including the official 11-row supplementary P-V grid,
+  diffraction-ready Schmidt et al. (1998) structure, published-value
+  reproduction, and diagnostic coefficient-parity refit. The canonical final
+  DOI is `10.2138/am-2019-6694`; the 2018 accepted-manuscript DOI is retained
+  only as source lineage to prevent a duplicate publication record.
+- Added the selected Shi et al. (2022) Rh2O3(II)-type Al2O3 BM3-MGD thermal
+  EOS as a distinct `Pbcn` phase, with all 75 official supplementary P-T-V
+  observations, published uncertainties, Pt pressure-scale provenance,
+  source-state reproduction, and an independent errors-in-variables refit.
+  The audit documents all eight corundum/Rh2O3(II) sensitivity fits and avoids
+  treating LitCurate's flattened BM3 rows as independent isothermal records.
+
+- Added the Campbell and Heinz (1994) RbCl-B2 BM3 material, including all 24
+  Table 1 observations and a parity-checked refit. Corrected the CsCl resource,
+  which had mistakenly contained an incomplete subset of the RbCl table, and
+  added all nine Yagi (1978) CsCl compression ratios needed for the complete
+  22-point parity reproduction.
+
+- Added Birch-Murnaghan Eulerian finite-strain/normalized-stress diagnostics,
+  including first-order P-V error propagation and fitted-model curves. The
+  numerical API is plotting-library independent; an executable notebook shows
+  how to produce the conventional F-f plot with Matplotlib.
+
+- Added phase-specific linear `Us`-`up` shock Hugoniot EOSs in Python and Rust,
+  including Rankine--Hugoniot pressure-volume inversion, velocity, density,
+  energy, tangent-modulus, uncertainty propagation, OLS/WLS/errors-in-variables
+  fitting, JSON-serializable fit results, and typed precursor, mass-basis, and
+  branch-domain APIs. Hugoniots remain in `Material.eos_records`, with filtered
+  `hugoniot_records` and `equilibrium_records` views and category-scoped
+  defaults. Loading history and transformed-branch identity are independent;
+  record evaluation enforces the declared branch domain, while `V0`, `rho0`,
+  formula units, and molar mass are cross-validated. Structured derivation
+  metadata covers coefficients obtained from SESAME or published tables.
+- Added a bundled executable pressure-calibration library for the Mao (1978),
+  Mao--Xu--Bell (1986), Dewaele (2004, 2008), Holzapfel (2005), and
+  Dorogokupets--Oganov (2007), and IPPS-Ruby2020 ruby R1 scales in Python and
+  Rust, together with the Akahama--Kawamura (2006) and Eremets et al. (2023)
+  diamond-anvil Raman-edge scales. All 34
+  identified ruby-calibrated material records now link to an exact calibration
+  identifier. Public APIs convert ruby scales through the R1 wavelength ratio,
+  diamond scales through the Raman wavenumber ratio, transform XRD pressure
+  scales through virtual same-standard volumes, and recursively normalize a
+  material EOS through its recorded calibration graph. An explicit edge
+  registry preserves the sources and transformations for simultaneous and
+  jointly optimized cross-calibrations. New records add the Fratanduono (2021)
+  Au anchor and the Tateno (2019) and Chidester (2021) B2-KCl thermal EOSs; the
+  complete 155-row Chidester KCl/Pt-derived-pressure table is bundled. The
+  complete Dorogokupets--Oganov (2007) Pt four-oscillator thermal EOS is now
+  implemented in Python and Rust and makes the Chidester KCl-to-Pt edge
+  executable, including the published KCl-effective-to-Pt-surface temperature
+  transformation.
+  Observation-level re-reduction from paired measured calibrant volumes remains
+  available as a separate operation. A dedicated pressure-scale normalization
+  guide documents valid same-standard, ruby, and cross-material paths, the
+  qualifying literature, validity and uncertainty rules, and current library
+  coverage. A set-level route helper now finds and ranks XRD-standard targets
+  reachable from every supplied EOS, with an executable notebook covering
+  direct, optical, recursive, and temperature-transforming routes.
+- Added the complete 51-row Somayazulu et al. (2023) B4C P-V-T table as a
+  provenance-bearing `.eosmat` dataset, linked it to new MGD and Berman thermal
+  records, and added a reproducible comparison of published, effective-
+  variance, and latent-coordinate fits. The open-source EosFit engine audit
+  also adds an explicit truncated-quadratic `berman` reference-volume law in
+  Python, Rust, and the interchange schema. An opt-in
+  `b4c_somayazulu_2023_berman_refit` record carries the EosFit7c public-data
+  refit alongside, without replacing, the published parameterization; generic
+  refit lineage and fit-provenance fields make that distinction machine-readable.
+- Added typed `search_materials()` and `search_eos_records()` discovery across
+  identity and aliases, formula and phase, equation family, DOI and reference,
+  thermal/caloric capability, uncertainty, scientific-validation status, and
+  closed calibration ranges with explicit `contains` or `overlaps` semantics.
+- Failed material, record, and raw-document identifier lookups now suggest
+  close stable identifiers.
+- Calibration discovery treats missing bounds as unknown rather than unbounded,
+  and material searches require all record-level criteria to match one record.
+- Added forward DAC-confinement prediction through
+  `volume_with_dac_confinement(P_cold, T, f_dac=...)`, together with an explicit
+  reference-relative `thermal_pressure_increment()` for displaying the full
+  thermal increment, retained confinement pressure, and total hot pressure.
+  Python material records and the Rust scalar, batch, and `.eosmat` APIs expose
+  the same workflow in their public volume conventions.
+- Added `DoubleDebyeLogMomentHelmholtz` in Python and Rust together with the
+  primary-source-validated Correa et al. (2008) diamond record. The complete
+  Helmholtz model implements the Vinet cold curve, logarithmic-moment
+  double-Debye weights, zero-point motion, and the published $T^2$
+  anharmonic contribution, with pressure, volume, energy, entropy, and heat
+  capacity evaluation.
+- Added optional reference-isotherm anchoring to both double-Debye Helmholtz
+  models. A numeric `Tr` subtracts the simulated ionic and anharmonic
+  contribution at that temperature before adding it to `rt_eos`; `Tr=null` in
+  `.eosmat` preserves the literal 0 K cold-curve formulation. The catalog now
+  includes Correa- and Benedict-thermal diamond variants anchored to the
+  experimental Dewaele 298 K Vinet isotherm.
+- Added a shared Python/Rust error contract with domain-specific Python
+  exceptions, stable machine-readable codes and context, Rust error-kind
+  accessors, preserved fitting source chains, and native-extension parity.
+- Added interpreted Rust `EosFitResult` parameters, latent states, profiled
+  covariance, standard errors, correlation, and fit statistics.
+- Added public Rust `.eosmat` validation, serialization, and save APIs with
+  extension-preserving canonical and legacy round trips.
+- Added PEP 561 typing metadata and a repository-wide `mypy` CI gate.
+- Added array-aware pressure and temperature conversions to `peritheos.units`,
+  plus conventional-cell to formula-molar-volume conversion in both directions.
+- Added separate complete-Python and native Rust coverage gates and
+  property-oriented native-versus-compatibility tests.
+- Added workflow-oriented Rust API documentation, six executable core and
+  fitting examples, docs.rs metadata, and warning-free rustdoc/doctest CI.
+
+### Changed
+
+- Corrected the B2-KCl refit reproductions for Walker (2002), Tateno et al.
+  (2019), and Chidester et al. (2021). Walker now follows the source's staged,
+  unweighted preferred
+  fit. Tateno now uses the final published `gamma0=2.3`, `q=0.8`, integrated-
+  Gruneisen Debye law, and the correctly aligned official Supplemental Table S1
+  workbook rather than the split accepted-manuscript table. Chidester now uses
+  the source's full simultaneous fit scope—123 Dewaele room-temperature rows
+  plus 155 new high-temperature rows—and the integrated-Gruneisen Debye law,
+  eliminating the false `q=0` boundary and recovering all five coefficients.
+- Unified the normal Python material API with the bundled `.eosmat` library.
+  `list_materials()` and `list_eos_records()` now return all bundled materials
+  and directly executable records in deterministic identifier order. Historical
+  pressure-scale identifiers and constants remain numerically stable
+  compatibility lookups after a record-by-record audit, while
+  `get_material_document()` remains the advanced raw-document API.
+- Canonical identifiers now take precedence over historical convenience names.
+  In particular, `get_material("diamond")` returns the complete seven-record
+  document-built material instead of the former five-record convenience
+  grouping; all historical diamond records remain available by their record
+  identifiers and compatibility constants.
+- EOS records now evaluate extrapolated states by default. Published ranges are
+  treated as calibration/data coverage, with opt-in enforcement through
+  `check_validity=True`; `within_calibration_range()` is the preferred coverage
+  query and `within_validity()` remains a compatibility alias.
+- Documented the factor-of-two normalization conflict between the Correa (2008)
+  and Benedict (2014) diamond anharmonic terms while preserving each published
+  equation and coefficient literally. For the published diamond parameters the
+  coefficient is volume independent, so this discrepancy affects caloric
+  quantities but not direct $P(V,T)$ or $V(P,T)$ evaluation.
+- Expanded Python fallback, Rust batch, error-category, thermal-domain, and
+  `.eosmat` validation tests; raised the branch-aware Python coverage floor to
+  90% and the Rust line-coverage floor to 85%.
+- Consolidated the public `peritheos-core` and `peritheos-fit` Rust crates into
+  one `peritheos` crate. Fitting and uncertainty APIs now live under
+  `peritheos::fit`; the Python API is unchanged.
+- Deprecated the old `peritheos.utils` pressure and temperature conversion
+  imports in favor of the consolidated `peritheos.units` API.
+- Reworked the getting-started documentation around validated material records,
+  clarified source-build requirements, and grouped contributor-only design
+  documents outside the main user navigation.
+
+## [0.6.0] - 2026-09-01
+
+### Added
+
+- Added the generic `DoubleDebyeHelmholtz` full-free-energy EOS with a Vinet
+  0 K cold curve, volume-dependent double-Debye modes and weights, zero-point
+  motion, a volume-dependent $T^2$ correction, analytic pressure, normal
+  pressure/volume/temperature inversion, fitting support, and documented
+  Benedict et al. diamond parameters as an example rather than model defaults.
+  The public Rust core implements the same Helmholtz, pressure, caloric, and
+  inversion equations and loads the model through native `.eosmat` dispatch.
+- Added the Benedict et al. (2014) diamond coefficients as an audited material
+  record in both the curated catalog and bundled `.eosmat` library, with
+  explicit per-atom/conventional-cell conversion and cold-curve caveats.
+- Added DAC two-volume temperature inversion for absolute Helmholtz models by
+  defining the confined thermal pressure relative to the 300 K isotherm; the
+  material-record wrapper accepts conventional-cell volumes directly.
+- Added native Rust loading for canonical Peritheos format-3 and legacy
+  Dioptas format-2 `.eosmat` files, with executable runtime-dispatched EOS
+  records, preserved JSON extensions, and automatic conventional-cell to
+  molar-volume conversion for energy-based thermal models.
+- Added eight executable notebook tutorials covering pressure calibration,
+  material-library exploration, `.eosmat` round trips, room-temperature EOS
+  and gold-scale comparisons, thermal state surfaces, DAC temperature
+  sensitivity, and fit-to-prediction uncertainty propagation.
+
+### Fixed
+
+- Corrected the scientific provenance of all eleven Sokolova pressure scales:
+  their reference inputs and final coefficients originate in Sokolova et al.
+  (2013), Tables 1 and 4, while the 2016 paper is the spreadsheet
+  implementation/correction source. Each material record now carries structured
+  source lineage and fitting-data caveats. The misleading `_sokolova_2016`
+  record identifiers and public constants were removed in favor of
+  `_sokolova_2013`; `Sokolova2016` remains the corrected calculator formalism.
+- Replaced the migrated InN BM3/experimental-volume hybrid with Muñoz and
+  Kunc's published theoretical wurtzite Murnaghan fit, including a reference
+  volume reconstructed from their Table 1 theoretical lattice constants.
+- Made the Campbell--Heinz B2-KCl entry an explicit two-primary-source
+  composite: its absolute `V0` and propagated error now follow from the
+  published B2/B1 volume ratio and Dewaele et al.'s B1 reference volume.
+- Consolidated duplicate `majorite`/`mgsio3-maj` materials, corrected the
+  Hanfland lithium equation family from BM3 to its combined-phase Vinet form,
+  and removed the unsupported Fei-FeO and Hixson-W standalone BM3 records.
+- Resolved the remaining primary-source blockers for CsCl, magnetite, Li,
+  majorite, MW60 magnesiowuestite, NiS, phase D, cubic SnO2, and SrO B1/B2.
+  Corrected equation orders, cell conventions, phases, fitted-versus-fixed
+  flags, ranges, and all printed parameter errors. Phase D now has distinct
+  AntA and AntB reference-volume records, and the inherited unsupported
+  majorite error and tenfold SnO2 volume-error transcription are removed.
+- Marked the published parameter errors for 14 Al, Cu, W, Ni, Ag, diamond,
+  alpha/omega-Ti, Si-V/Si-VII/Si-X, Re, corundum, and LiF records as 95%
+  confidence half-widths. File-loaded uncertainty propagation now converts
+  these intervals to normal-equivalent standard errors instead of treating
+  them as one-standard-deviation errors.
+- Corrected the imported Dioptas Fei et al. (2007) Au and Ne records to use
+  `MieGruneisenDebye` with `debye_temperature_law="variable_exponent"` rather
+  than the implicit `integrated_gruneisen` default. The `.eosmat` records
+  preserve the original behavior and cite equation 3 in explicit
+  migration-correction metadata.
+- Corrected the Hazen--Finger (1979) zircon record from an inconsistent BM2
+  representation to BM3 with the published assumed `K0' = 6.5`, including the
+  reported `V0` uncertainty.
+- Corrected the Holmes et al. (1989) platinum record from BM3 to its published
+  universal (Vinet) isotherm, restored its model reference volume and
+  0--550 GPa static range, and represented Equation (12) with the published
+  constant thermal-pressure coefficient.
+- Restored the Ross (1997) magnesite fitted `V0` and uncertainty and normalized
+  the Haines et al. (2001) Mo2C reference volume from the primary specimen's
+  measured ambient lattice parameters.
+- Corrected primary-source values and error metadata for CaSiO3, CaO B1/B2,
+  rutile GeO2/SnO2, PbS B1, wadsleyite, jadeite, and B2 KCl; corrected the
+  migrated B2-KCl thermal component to Walker et al.'s additive BE1 form and
+  retained the published uncertainty of its directly fitted `alpha0*K0`
+  product. Published errors are retained even where the associated value was
+  fixed during a fit, as for Shim et al.'s CaSiO3 `V0`.
+- Replaced the migrated Anderson et al. Au `AlphaKT` approximation with the
+  exact Equation (29) logarithmic-volume linear thermal pressure, restored its
+  density-derived reference volume, Table V domain, and partial published
+  `(dKT/dT)V` uncertainty, and promoted the record after primary-source review.
+- Removed the Martinez et al. aragonite global HT-BM3 record: Table 7 omits
+  its fitted `V0(298 K)`, and the remaining coefficients do not reproduce the
+  printed 64-point table under documented pressure- or volume-residual least
+  squares. The independently reproducible staged BM2 result is retained and
+  extended with its published Equation (2) `K0(T)` slope and Equation (3)
+  direct-linear reference-volume law.
+- Promoted the Scott et al. (2001) cementite BM3 record after checking the
+  complete primary article: the ambient `V0`, weighted-fit coefficients, all
+  printed errors, 300 K reference state, and compression interval are now
+  traced to pages 1875--1877. The measured `V0` is explicitly fixed in the
+  fit metadata, and the unreported covariance/confidence convention remains
+  documented rather than inferred.
+- Corrected the Clendenen and Drickamer (1966) CoO record from an inherited
+  BM3 representation to the published Murnaghan Equation 4, restored the
+  Table II ambient cell and Table III 30.8 GPa range, and retained null errors
+  because the primary paper reports no parameter uncertainty or covariance.
+- Promoted the Noguchi et al. (1999) NiO shock-derived 300 K BM3 isotherm after
+  checking the official 1998 primary conference paper for the sample reference
+  lattice, its propagated uncertainty, and the Mie--Gruneisen reduction. The
+  final journal article supplies the 147.6 GPa range and `K0`/`K0'`; their
+  errors remain null because the authors do not report them.
+
+### Added (catalog and native backend)
+
+- An executable documentation notebook using the complete printed Martinez et
+  al. (1996) aragonite Table 3 dataset to demonstrate 298 K and staged-isotherm
+  BM2 fitting, thermal-trend recovery, scaled joint P-V-T fitting, residual
+  visualization, and uncertainty/chi-square interpretation.
+- A reproducible primary-source audit for the curated migrated `.eosmat` EOS
+  records. All 147 bundled records are now directly validated against original
+  publications, official supplements, or stable institutional reports, and no
+  record remains pending or deferred. The
+  bundled machine-readable ledger records source locations, the independently
+  checked Shen--Smith (2026) Vinet fits and errors, the B4C order inconsistency,
+  and the restored Hanfland graphite `V0` uncertainty.
+- A mechanism-oriented `ThermalReferenceStateEOS` implementation for the
+  temperature-dependent `V0(T)`/`K0(T)` formulation used by the validated ice
+  VI/VII records. It now supports a generic `thermal_expansion_law`, including
+  exact analytical integration of `alpha0+alpha1*T`; the constant law remains
+  backward compatible. A separate `reference_volume_law="linear_temperature"`
+  represents a directly linear mean-expansion relation without conflating it
+  with integrated instantaneous expansivity. The Dioptas `AlphaKT` interchange
+  type maps to the canonical `thermal_reference_state` model identifier.
+- Primary-source-validated native material records for the Martinez et al.
+  (1996) staged aragonite BM2 P-V-T parameterization and the Dewaele et al. (2012) B2-KCl
+  P-V-T pressure calibration. KCl uses the paper's Vinet reference isotherm,
+  additive thermal-pressure term, fixed fictive `V0`, and explicit
+  experimental-versus-computational validity provenance. It is the preferred
+  `kcl.eosmat` record. After primary-source corrections and duplicate removal,
+  the catalog now contains 147 records, all validated.
+- Primary-audit corrections restore Sokolova `n`/`Z`, silica Debye `n`, and
+  ice `Tr` inputs omitted by migration; every validated migrated record is
+  constructability-tested.
+- A reproducible BurnMan/Pytheos public-API black-box comparison report,
+  deliberately separated from primary-source validation and test baselines.
+- Executable `Material` conversion through the same canonical `.eosmat`
+  format 3 used for Dioptas exchange. Optional symmetry, lattice, space-group,
+  atom-site, peak, and unknown extension fields survive a Peritheos round
+  trip; cell-to-molar volume conversion is explicit per EOS record. Loading
+  uses a fixed model registry, refuses unaudited records by default, and keeps
+  snapshot-v2 reading only for compatibility.
+- A configurable `debye_temperature_law` on `MieGruneisenDebye`, with
+  `integrated_gruneisen` as the backward-compatible default and
+  `variable_exponent` for sources that directly publish a volume-dependent
+  exponent. Also added the mechanism-named
+  `MultiOscillatorGruneisenThermalEOS` class, which accepts any
+  isothermal `EosBase` and uses a generic numerical $dK/dP$ fallback where
+  needed; the earlier paper-named imports remain compatibility aliases.
+
+- A first-class `Material`/`EOSRecord` catalog API with GPa pressure, conventional-unit-cell
+  volumes, scalar/array pressure and volume inversion, explicit material/phase
+  and unit metadata, DOI-level parameter provenance, published validity
+  envelopes, JSON-safe catalog records, and uncertainty propagation from
+  measured volume/temperature and published parameter errors.
+- A primary-source-validated catalog: Tange et al. (2009) Fit3-Vinet
+  P-V-T MgO B1 and the Dorfman et al. (2012) 300 K Vinet co-compression scales
+  for Au, Pt, Mo, NaCl B2, and Ne; Dewaele 2019 LiF and NaCl B1/B2; Dewaele
+  2012 KCl and KBr B1/B2; Datchi 2007 c-BN; and Dewaele 2008 diamond, Ag,
+  and Ni.
+- All eleven Sokolova thermal pressure markers: MgO, diamond, Al, Cu, Ag, Au,
+  Pt, Nb, Ta, Mo, and W, with the original 2013 fit provenance and the
+  corrected 2016 workbook equations.
+- The Fei et al. (2007) internally consistent Au, Pt, NaCl-B2, and Ne thermal
+  scales and a dedicated Debye-temperature convention that preserves the
+  paper's equation rather than substituting the generic integrated form.
+- The quasi-hydrostatic 300 K hcp Re Vinet scale of Anzellini et al. (2014),
+  with Table III lattice-data regressions and its published 95% fit intervals
+  retained distinctly from one-standard-deviation errors.
+- The Tange et al. volume-dependent Gruneisen Mie-Gruneisen-Debye thermal model,
+  with printed Table 5 regression cases and analytic thermodynamic checks.
+- A reusable linear thermal-pressure EOS for the Dewaele KCl/KBr equation and
+  state-only uncertainty propagation where a source reports no parameter errors.
+- A Peritheos-owned flat `.eosmat` format 3, normative JSON Schema, complete
+  115-material/147-record EOS database migrated from Dioptas 0.10.0 with
+  explicit validation status and provenance, legacy Dioptas format-2 input,
+  and tested Dioptas 0.10.0 read compatibility. A dedicated schema reference
+  documents every field, discriminator pairing, default, unit, validation
+  status, and consumer compatibility rule.
+- A Rust workspace containing native EOS, fitting, uncertainty, and private
+  PyO3 binding crates, with Rust 1.83 as the library MSRV.
+- Shared Python/Rust compatibility fixtures and migration baselines for all
+  isothermal and thermal model families.
+- Multi-platform native-wheel release jobs for supported CPython versions on
+  Linux x86-64/ARM64, macOS Intel/Apple Silicon, and Windows x86-64.
+- Pull-request wheel build and isolated-install smoke tests on Linux, macOS,
+  and Windows, complementing the full tagged-release wheel matrix.
+- Dependency-free public Rust batch traits, typed joint EOS fitting, and
+  model-aware linear and Monte Carlo uncertainty entry points.
+- Package-contained scientific fixtures and a two-crate archive verifier that
+  tests the required core-before-fit crates.io publication sequence.
+- A pinned Rust dependency-source and SPDX-license audit covering the supported
+  Linux, macOS, and Windows target graphs.
+
+### Changed
+
+- Built-in Python EOS classes now preserve their public API while delegating
+  evaluation, inversion, thermoelastic, and caloric calculations to Rust.
+- The material catalog's linear, logarithmic-volume, configurable
+  reference-state, variable-exponent Debye, asymptotic-power-law Debye, and
+  generic multi-oscillator mechanisms now use the same native evaluation and
+  fitting architecture. Thermal fits accept fixed categorical equation choices
+  through `configuration`, and native linear uncertainty supports records with
+  measurement errors but no published parameter covariance.
+- Named bounded robust fitting losses and uncertainty propagation statistics
+  now use native numerical kernels. Custom reference EOS classes, callable
+  fitting losses, and NumPy-seeded Monte Carlo draws retain documented
+  compatibility paths.
+- Native fits now return their profiled global-parameter covariance directly;
+  Python no longer recomputes it through a separate SciPy/NumPy path.
+- Errors-in-variables fits now use colored latent-coordinate Jacobians and a
+  block Schur-complement solve, with stress coverage for large and rank-deficient
+  datasets.
+- Birch-Murnaghan kernels now use an algebraically equivalent cube-root form,
+  and large independent EOS arrays use deterministic thresholded parallel
+  evaluation while the Python interpreter lock is released.
+- Holzapfel bulk-modulus derivatives now execute directly in Rust, obsolete
+  Python natural-strain coefficient formulas are removed, and the historical
+  coefficient-level Holzapfel helper remains available through a Rust-backed
+  compatibility wrapper.
+- Native batch calls now have concurrent large-array stress coverage in
+  addition to deterministic order, shape, stride, and round-trip checks.
+- Native least squares now equilibrates differently scaled Jacobian columns,
+  reports failed steps as failures instead of false `xtol` convergence, and
+  uses a rank-aware Moore-Penrose covariance calculation.
+- Latent-coordinate covariance profiling now preserves observation-local
+  blocks, avoiding the previous dense cubic post-fit calculation.
+- Native EOS dispatch is restricted to exact built-in Python classes so
+  subclass overrides and Debye/Einstein model identity remain authoritative.
+- Linear uncertainty kernels now reject non-positive-semidefinite parameter
+  covariance and negative state variance instead of clamping invalid inputs.
+- Tagged releases now include CPython 3.14 free-threaded wheels alongside the
+  standard CPython wheel matrix.
+
+## [0.5.0] - 2026-08-30
+
+### Added
+
+- A comprehensive equation reference for every public isothermal and thermal
+  EOS, including parameter definitions, model domains, numerical inversion,
+  thermoelastic identities, fitting diagnostics, uncertainty propagation, and
+  explicit public-unit conversion factors.
+- Documentation distinguishing the equations typeset in Sokolova et al. (2016)
+  from the accompanying Excel workbook calculation implemented by
+  `Sokolova2016`, including differences in the bulk-modulus derivative,
+  characteristic-temperature multiplier, oscillator pressure contributions,
+  reference-isotherm subtraction, and working pressure units.
+- Temperature inversion from pressure and volume for all thermal EOS models,
+  including NumPy broadcasting through `temperature()` and
+  `calculate_temperature()`.
+- Coupled temperature inference from ambient and heated volumes with a
+  fractional DAC confinement contribution through `temperature_from_volumes()`.
+  The empirical `f_dac * thermal_pressure` pressure increment is solved in its
+  algebraically reduced form and requires `0 <= f_dac < 1`.
+  Documentation distinguishes this fraction of EOS thermal pressure from a
+  fraction of cold pressure and describes its physical limits, calibration,
+  identifiability, baseline-drift, and uncertainty constraints.
+- Fixed-volume preparation for temperature inversion, avoiding repeated
+  Sokolova volume-integral evaluations during root finding.
+
+### Changed
+
+- Refreshed the locked runtime, development, documentation, and release
+  dependencies to their latest versions compatible with the supported Python
+  and declared package-version ranges.
+- The minimum supported SciPy version is now 1.9.3, ensuring a stable binary
+  installation and numerical fitting path on Python 3.9 across supported
+  platforms.
+- Documentation now uses Material for MkDocs with responsive navigation,
+  improved search and code presentation, and automatic light and dark themes.
+- The model overview now focuses on model selection and reference-EOS
+  compatibility, while advanced two-volume DAC analysis has moved out of the
+  introductory tutorial into a dedicated guide.
+
+## [0.4.0] - 2026-08-09
+
+### Added
+
+- Branch-aware coverage enforcement, Ruff lint and formatting gates, and a
+  minimum-supported-dependency CI job.
+- Literature-tagged numerical cases for all public EOS families and an
+  independently solvable weighted least-squares fitting benchmark.
+- Automatic GitHub Releases with source and wheel artifacts.
+- Citation metadata, API stability, contribution, security, support, conduct,
+  issue, and pull-request policies.
+- Versioned Read the Docs builds using the locked uv documentation environment.
+- Joint reference-isotherm and thermal parameter fitting with complete
+  cross-covariance through `fit_joint_eos`.
+- Per-observation correlated P-V and P-V-T covariance matrices and robust
+  least-squares losses for all fitting entry points.
+- Human-readable fit summaries and versioned, JSON-safe result export including
+  model parameters, covariance, adjusted observations, diagnostics, and solver
+  metadata.
+
+### Changed
+
+- Uncertainty propagation tests now cover Monte Carlo state sampling, invalid
+  covariance and option handling, one-sided numerical derivatives, wrapper
+  behavior, and failed sampling; branch-aware project coverage exceeds 90%.
+- Corrected the fourth-order Birch-Murnaghan documentation to match its
+  implemented Eulerian-strain signs and exponents.
+- Package maturity metadata now identifies the 0.4 development line as beta.
+- The minimum supported NumPy version is now 1.21, matching the public typing
+  APIs used by Peritheos.
+- Distribution package discovery is restricted to `peritheos`, preventing
+  generated documentation directories from entering or blocking builds.
+
+## [0.3.0] - 2026-08-08
+
+### Added
+
+- Murnaghan, modified Tait, and second- through fourth-order natural-strain
+  room-temperature equations of state.
+- Holland-Powell thermal modified Tait equation of state.
+- Bounded P-V and P-V-T errors-in-variables fitting with independently optional
+  pressure, volume, and temperature standard errors.
+- `FitResult` covariance, correlation, parameter-error, residual, chi-square,
+  information-criterion, and adjusted-state diagnostics.
+- `EOSUncertainty` for propagating complete covariance matrices, parameter
+  errors with correlations, or partial independent parameter errors into EOS
+  calculations.
+- Linear covariance propagation and reproducible Monte Carlo propagation for
+  pressure, volume, bulk modulus, and arbitrary public EOS quantities.
+- Optional propagation of pressure, volume, and temperature errors in the
+  requested state.
+- Explicit combination of separately quantified thermal and reference-EOS
+  uncertainty blocks.
+- Isothermal compressibility, thermal expansivity, constant-volume and
+  constant-pressure heat capacities, thermodynamic Gruneisen parameter, and
+  adiabatic bulk modulus APIs.
+- Vibrational entropy, internal energy, Helmholtz energy, enthalpy, and Gibbs
+  energy for the Mie-Gruneisen-Debye and Mie-Gruneisen-Einstein models.
+- Molar-volume and density conversion helpers with explicit unit validation.
+- Literature-tagged reference cases and expanded numerical regression tests.
+- A multi-page MkDocs documentation site covering models, fitting,
+  uncertainty, properties, units, validation, references, and development.
+
+### Changed
+
+- EOS models now expose reconstructable parameter values and safe parameter
+  replacement, including dotted reference-EOS parameters such as
+  `rt_eos.K0`.
+- Thermal EOS calculations consistently document and enforce their molar
+  volume, pressure, and temperature units.
+- CI now builds the documentation strictly, and source distributions include
+  documentation and literature-validation data.
+- The legacy fitting argument `sigma` remains supported as an alias for
+  `pressure_sigma`.
+
+### Statistical assumptions
+
+- Parameters omitted from an uncertainty specification are treated as exact.
+- Individual parameter errors without a correlation matrix are treated as
+  mutually independent.
+- Combining thermal and reference-EOS results from separate fits requires an
+  explicit block-independence assumption.
+
+## [0.2.0] - 2026-08-07
+
+### Added
+
+- Fourth-order Birch-Murnaghan support.
+- Complete Sokolova et al. (2016) pressure terms and parameters.
+- Automated test and trusted PyPI publication workflows.
+
+### Changed
+
+- EOS pressure and volume validation and inversion were hardened for invalid
+  and out-of-domain states.
+- Project naming and release metadata were standardized.
+
+[Unreleased]: https://github.com/CPrescher/peritheos/compare/v0.9.0...HEAD
+[0.9.0]: https://github.com/CPrescher/peritheos/compare/v0.8.0...v0.9.0
+[0.8.0]: https://github.com/CPrescher/peritheos/compare/v0.7.0...v0.8.0
+[0.7.0]: https://github.com/CPrescher/peritheos/compare/v0.6.0...v0.7.0
+[0.6.0]: https://github.com/CPrescher/peritheos/compare/v0.5.0...v0.6.0
+[0.5.0]: https://github.com/CPrescher/peritheos/compare/v0.4.0...v0.5.0
+[0.4.0]: https://github.com/CPrescher/peritheos/compare/v0.3.0...v0.4.0
+[0.3.0]: https://github.com/CPrescher/peritheos/compare/v0.2.0...v0.3.0
+[0.2.0]: https://github.com/CPrescher/peritheos/releases/tag/v0.2.0