xtalforge

Appearance

Changelog ​

All notable changes to this project are documented here. The format follows Keep a Changelog and the project uses Semantic Versioning (pre-1.0: 0.0.x until the first essential release, then 0.x.y).

[1.3.0] - 2026-11-12 ​

Phase 10 makes xtalforge a MOF platform: experimental CIFs can be checked and cleaned, deconstructed into nodes, linkers and a topology, described with the standard geometric and adsorption descriptors, modified, and screened in batches. The public API is additive.

Added ​

  • MOF checking and cleaning: mof_check (overlaps, fragments, valence, disorder, metal coordination, charge), mof_solvent_remove with the embedded solvent table and the CoRE FSR/ASR policy, mof_disorder and mof_charge_balance.
  • Framework understanding: mof_deconstruct (all-node SBU/linker split and contracted quotient net), mof_mofid (canonical unit SMILES, topology, MOFid string and SHA-256 MOFkey), mof_open_metal_sites and mof_descriptors.
  • Native porosity: mof_pores with periodic union-find PLD, LCD, accessible surface area in m2/g and m2/cm3, accessible volume, void fraction, channel dimensionality and the geometric pore-size distribution.
  • Adsorption: mof_guests (TraPPE CO2, N2, CH4, H2, H2O), mof_forcefield (UFF/EQeq host, Lorentz-Berthelot mixing), mof_widom (Henry coefficient and heat of adsorption), mof_gcmc (single-component isotherm), mof_energy_grid and mof_raspa input export.
  • Modification: mof_functionalize (NH2/OH/CH3/F), mof_defects (missing linker and cluster defects with formate/water/hydroxide caps), mof_interpenetrate, mof_stack_layers and mof_mixed (ordered orbit or seeded random mixed-metal/linker frameworks).
  • Screening: mof_read_database for CoRE MOF and QMOF directories, mof_screen for a resumable descriptor/Henry table with MOFkeys and mof_deduplicate (MOFkey first, structure matcher for ties).
  • CLI mof subcommands (check, solvent-remove, disorder-resolve, charge-balance, deconstruct, mofid, open-metal-sites, descriptors, pores, adsorption, modify, database), Python functions and WASM JSON exports. docs/api/mof.md, docs/guide/mof-workflows.md and two new algorithm pages document the phase; the parity matrix gains 23 rows.
  • P1Atom disorder groups, the MOF host-guest data files and the new mof-data feature flag.

Fixed ​

  • The MOF checker does not report a charge imbalance for a metal-free covalent framework; charge balance is assessed only when metals are present.

[1.2.0] - 2026-11-11 ​

Phase 9 adds the core operations of a standard crystal library beyond the Crystal-Builder parity of 1.1.0. Every capability is available in the Rust API, the CLI, Python and WebAssembly, and the public API is additive.

Added ​

  • Structure comparison: match_structures (Niggli reduction, integer lattice mapping, supercell matching, species comparators), periodic_rmsd, kabsch_alignment and group_structures. CLI compare and group.
  • Structure edits: deform (Voigt strain, deformation gradient, target volume and density), seeded rattle, edit_structure (add, remove, replace, move, species maps, partial replacement, sort and wrap) with undoable commands, matched interpolate and optimal_supercell (ASE find_optimal_cell_shape objective).
  • Local environments: periodic voronoi with exact half-space clipping and solid angles, coordination with cutoff/Voronoi/CrystalNN strategies and continuous shape measures, bond_valence_sums, oxidation_states (ICSD MAP search with charge balance), radial_distribution (total and partial RDF/PDF, running coordination), bond-angle distributions and structure_fingerprint.
  • Surfaces: unique_miller_indices under the detected point group, slab_terminations (symmetric slabs, dipole and polar report, minimal slab), the wulff construction and adsorption_sites with place_molecule.
  • Defects and disorder: symmetry-unique vacancies, substitutions and interstitials with minimum-image supercells, orderings ranked by an in-process Ewald sum, special_quasirandom with pair/triplet correlation reports, lattice_matches and stack_slabs heterostructures.
  • Mechanical and thermodynamic basics: elastic_tensor from analytic stress finite differences with Voigt-Reuss-Hill moduli and a stability flag, eos_fit (Birch-Murnaghan, Vinet, Murnaghan with errors) and convex_hull (formation energies and a two-phase simplex hull distance).
  • The CLI gains compare, group, deform and the features command group; Python gains the corresponding functions; WASM gains JSON-returning exports for every operation.
  • Documentation: docs/api/features.md, algorithm pages for comparison, local environments, surfaces, defects and mechanics, and a parity matrix with 29 new rows.
  • Roadmap to 2.0.0: Phases 9-17, focused on MOFs (core features beyond Crystal-Builder parity, the MOF platform, usability audit and 1.x polish, native bindings, ecosystem interoperability, scientific completeness, performance, workflows and the 2.0 redesign), with the 2026-11-11 usability findings.

Fixed ​

  • P1Atom carries the disorder group of its parent site, so orderings and SQS generation can find partial occupancies after the P1 expansion.

[1.1.0] - 2026-11-11 ​

First feature release after 1.0.0, published on 2026-11-11. It adds the 3D playground and the end-to-end audit battery, exposes the scene and text-returning WASM helpers, and folds in the fixes from the functional audit of every surface (Task 8.9). The public API remains backward compatible: the new JSON and object fields are additive.

Added ​

  • The 3D playground in the docs: load, transform and build a structure with the WASM package and inspect the same Rust scene in an orbitable WebGL view.
  • scene_data on the WASM, Python and CLI (mesh --scene) surfaces, and n_atoms in the WASM structure summary, so a viewer no longer reconstructs geometry in JavaScript.
  • Text-returning WASM variants of the symmetry transforms and the carbon/MOF builders (to_p1_text, primitive_text, standardize_text, asymmetrize_text, set_space_group_text, change_setting_text, reduce_text, lower_to_subgroup_text, build_graphene_text, build_nanotube_text, build_c60_text, build_c20_text, build_mof_text, build_ribbon_text, build_twist_text, build_ztc_text, build_schwarzite_text, build_polymer_text, hydrogens_text and ff_optimize_text).
  • The end-to-end audit battery (tests/e2e/): 68 Playwright checks against gemmi, spglib, ASE and NumPy references, plus deterministic 3D screenshot goldens, and scripts/e2e.sh to run it. The playground UI checks cover the tabbed panels, atom selection, the Brillouin zone, the powder plot, UFF energy, exports, fullscreen, porosity, the MOF builder and all four powder refinement methods.
  • A guided Tour tab in the playground: seventeen steps that load a sample or run a builder, set the camera and open the tab where the result is read (CIF reading, space group, Brillouin zone, measures, supercell, slab, PXRD, nanotube, twisted bilayer, C60, MOF, disordered carbon, polymer, porosity, force field, powder refinement and export). The powder plot gains a 2θ axis, hkl labels for the strongest lines, I/I0 in the table and recalculates when the structure changes; the structure panel adds the disordered-carbon builder.
  • build_carbon_text on the WASM surface returns the disordered carbon as CIF for a viewer.

Fixed ​

  • PXRD intensities used the Z-based Gaussian form factor Z exp(-0.5 Z^(2/3) s^2) for every element, so relative intensities were off by up to 50 % (rutile (1 1 1) 28 % of (1 1 0) instead of 19 %). The neutral-atom International Tables (1992) coefficients are now embedded for H to Cf and the Si, NaCl, Fe and rutile patterns match an independent calculation to 0.1 %. Rietveld/Le Bail/Pawley stages that start from the calculated pattern inherit the change.

  • commensurate_twist ignored the requested angle: it rounded every twist to 0 or 60 degrees, and its "60 degree" variant shifted both layers. It now builds the commensurate moire cell ((2, 1) for 21.79 degrees, 14 atoms per layer) and records twist_angle and twist_indices; n x m repeats the moire cell.

  • build_polymer packed the chain into a cube shorter than the chain, so the periodic images overlapped (C-C 0.67 Angstrom, coordination 5-7 for polyethylene). The repeat units are now joined head to tail along a into a periodic chain and the cross-section sets the density.

  • embed_smiles had no bond-angle term, so rings collapsed (benzene with 1.17 Angstrom 1-3 contacts and non-planar). Angle targets from the hybridization (180, 120, 109.47 degrees) are now part of the relaxation and the default iteration count is 1500.

  • build_carbon placed carbons at mesh-patch centres without relaxing them, so atoms could sit 0.15 Angstrom apart, and put edge terminations along the sheet normal instead of the free in-plane valence. The dual graph is now relaxed to 1.42 Angstrom bonds and 120 degree angles, terminations follow the free valence, and a termination that would overlap another atom is dropped and reported as a bare edge. Termination and graph bonds that cross the cell boundary now carry their lattice image; before, 71 of the default dia build's bonds were stored without it and drawn 23-33 Angstrom long across the cell.

  • P1Cell::expand and Lattice::wrap turned a tiny negative coordinate (such as -1e-32 from a Cartesian round trip of 0) into exactly 1.0. The UFF view then held a duplicated atom, so ff vibrations on silicon reported modes of +-14 000 cm^-1 (energy +1028 eV after the Hessian displacements). Coordinates now stay in [0, 1); silicon gives three zero acoustic modes and degenerate optical modes.

  • le_bail and pawley subtract a SNIP background before extracting intensities and le_bail uses the profile-weighted partition; on rutile with a 50-count baseline R drops from 0.69 to 0.27.

  • MOF nodes are oriented with a Kabsch fit over every connection-to-edge assignment instead of the greedy mean-direction rule, so planar nodes in srs/hcb no longer force linkers to be stretched 0.55-1.26 times.

  • build_ztc and build_schwarzite return three-coordinate sp2 carbon (the relaxed dual of the remeshed gyroid/Schwarz P surface, with explicit bonds) instead of mesh vertices with coordination 5-8; their sampled grid now includes the closing plane, so ztc_surface_mesh and schwarzite_surface_mesh are closed across the cell faces.

  • Wyckoff letters are matched on the input coordinates when the input basis is conventional, so they follow the input origin like spglib and the CIF (silicon in origin choice 1 is 8a, not moyo's 8b).

  • eeq_charges is now EEQ2019 (multicharge parameters, coordination-dependent electronegativity, erf-damped Coulomb) instead of a copy of QEq; isolated water gives O -0.598 like the sci-form/dftd4 reference.

  • refine_phase with refine_cell starts from a Pawley cell (new first stage pawley_cell), so a cell 1 % off no longer traps the refinement.

  • build_mof refuses two metal-free carboxylate blocks, whose oxygens would overlap, instead of returning the overlapping framework.

  • build_ztc refuses cells below 12 Angstrom and build_schwarzite below 8 Angstrom, where the surfaces cannot hold three-coordinate carbon.

  • eeq_charges uses an Ewald sum for the periodic Coulomb term (MOF-5 Zn was +4.4 with the real-space cutoff).

  • Python ff_charges and WASM ff_charges_info accept eeq; the CLI info JSON and text carry n_atoms; WASM errors that were bare strings are now Error objects with code = "invalid_input"; PXRD reflections with equal two-theta are ordered identically on native and WebAssembly builds.

  • A refinement panel in the playground (Powder tab): load the bundled synthetic rutile pattern or a file and run Le Bail, Pawley (with size and strain), Rietveld or the automatic run, with an observed/calculated/ difference plot; the tour has a matching step (seventeen steps).

  • The automatic powder result exposed only the quick cubic-index edge as its cell, so the refinement panel and the CLI table showed the indexing candidate as the "best" cell. CLI, Python and WASM now also return best_cell: the Pawley-refined cell of the leading candidate, or its indexed cell when the Pawley stage was skipped (rutile automatics report a = 4.5993 Angstrom instead of the indexed 4.5902 Angstrom).

  • Running a builder or transform in the playground left the sample selector on a stale sample id; it now shows "Built or transformed", and the nanotube tour step looks from +x so the tube lies across the view.

  • Python powder_lebail, WASM powder_lebail_info and powder_pawley_width_info; WASM Pawley results carry size, strain and the fitted calculated profile. scripts/check-surfaces.sh and the release checklist compare 97 fields across CLI, Python and WASM.

  • fit_phase refines the constant background with the scale, as documented (the automatic run's phase Rwp on rutile drops from 1.05 to 0.33).

  • The CLI Rietveld uses the library's default iteration count, like Python and WASM, and exits quietly (status 141) when its output pipe closes instead of panicking.

  • The playground radiation selector was empty: seven static <option> elements were hoisted into one static block that Vue cannot insert into a <select>. The options are now rendered from a list.

  • The playground twisted-bilayer form called the builder with m = 0, which is rejected; it now defaults to the (2, 1) cell at 21.79 degrees.

  • Structure::supercell copied only the asymmetric sites, so a supercell of a symmetric structure lost its symmetry expansion (a 2 x 2 x 2 silicon supercell had 8 atoms instead of 64, in the CLI and WASM). It now expands the P1 cell first through Structure::as_p1; the CLI size estimate and the supercell tests use the full-cell count.

  • Structure::is_p1 trusted only the Hall symbol, so a CIF with an explicit operation loop and no space-group name was treated as P1 and reported the wrong formula, density and charge. It now also requires every operation to be the identity.

  • The CIF-inspector demo's inline silicon CIF lacked the centring translations and reported one atom and 0.29 g/cm3; it now uses the Hall symbol and shows Si8 and 2.329 g/cm3.

  • The WASM structure summary now includes the P1 n_atoms, matching Python and the whole-cell formula work released in 1.0.0.

  • brillouin_zone_for_structure detects the primitive cell, so silicon and rock salt (fcc), bcc iron and other centred cells no longer return the cube of the conventional reciprocal cell. The clipping now starts from a cube large enough for anisotropic cells, so graphene returns its hexagonal prism instead of a box; the cap ordering also works for non-unit plane normals.

Changed ​

  • The WASM budget moves from 3904 to 3968 KiB for the audit fixes (IT92 table, commensurate twist, periodic polymer, embedder angles, carbon relaxation, ZTC/Schwarz P dual, Le Bail/Pawley background, MOF quaternion fit and build_carbon_text; 3943 KiB measured).
  • The WASM release budget moves from 3840 to 3904 KiB for scene_data, n_atoms and the text-returning builders (3845 KiB measured).

[1.0.0] - 2026-11-10 ​

First stable release, published on 2026-11-10 on crates.io (13 crates), PyPI (7 platform wheels plus an sdist), npm, the GitHub release and GitHub Pages. The public API is frozen and changes follow the SemVer policy in docs/api/stability.md. This release incorporates the fixes accepted during the 1.0.0-rc.1 stabilization period.

Fixed ​

  • Structure::formula, cell_contents, cell_mass, density, z and total_charge (and therefore info, structure_info, the agent info verb, Python and WASM summaries) counted the asymmetric unit, so every structure with symmetry operations reported the wrong formula, density and charge: NaCl read from a CIF showed Na Cl and 0.55 g/cm3 instead of Na4 Cl4 and 2.19 g/cm3. They now use the P1 expansion; z is the gcd of the element counts.
  • The CIF reader records coincident_atoms_warning when atoms of different sites coincide in the P1 cell with total occupancy above one (a file that lists the whole cell under a non-P1 group), and the CLI info prints it and symmetry_warning on standard error.
  • Writing to a device such as /dev/null works: atomic writes rename the temporary file only for regular files, so devices and FIFOs are written in place.
  • graphene: the atoms were stored with Cartesian coordinates as fractional ones and the b vector used nx/2 instead of ny/2, so the sheet was distorted (bonds of 1.07 Angstrom) and not a supercell when nx != ny.
  • nanotube: the strip was not restricted to one circumference, the translation vector used gcd(n, m) instead of d_R = gcd(2n + m, 2m + n) and the atoms were not rolled onto a cylinder. (5,5) had 166 atoms and 537 bonds; it now has the 20 atoms and 30 bonds of the chirality formula. The nanotube cell is now 2R + vacuum on a and c with the axis along b.
  • nanoribbon: the vacuum stretched the strip instead of adding empty space; the bonds are 1.42 Angstrom again.
  • structure_info (and the CLI info, Python and WASM equivalents) rejected the versioned {"schema_version", "structure"} envelope that --json commands write, so xtalforge --json supercell ... | xtalforge info - failed; both forms are accepted now.
  • powder index (cubic search): a face-centred pattern indexed on a larger multiple of the true cell (NaCl returned 7.94 Angstrom); the smallest cell among equally good candidates is now chosen.
  • mof build: the topology cell was fixed, so a long linker was squashed (0.45x for stilbene in pcu) and thousands of atoms overlapped; the builder now fits a uniform cell scale to the linker (cell_scale metadata). A node with the wrong connection count, a linker with more than two arms, a linker scale outside 0.8-1.25 and atom pairs closer than 0.9 Angstrom are reported in warnings, and the CLI prints them on standard error.
  • powder rietveld --refine always modelled the Cu K alpha doublet, so a single-wavelength pattern drove the ADP multiplier to its bound; the new --no-doublet flag matches Python's doublet=False.
  • powder automatic / run_automatic: the peak search did not subtract the background, so on a noisy pattern with a raised baseline every noise maximum was a peak, the indexing returned a wrong cell (16.5 Angstrom for NaCl) and the run did not finish in five minutes in a release build. It now subtracts a SNIP background and applies a noise-based threshold; the same pattern runs in 2.4 s and returns Fm-3m first. PeakOptions::noise_limited exposes the same search and powder index --background uses it (a = 5.614 instead of 16.5).

Tests ​

  • Phonons of fcc argon with UFF (phonon_argon.rs) and the CLI --no-doublet flag have regression tests.
  • Regression tests cover the whole-cell properties and the coincident-atoms warning, the versioned structure_info envelope, atomic writes to devices, the graphene/nanotube/ribbon geometry, cubic indexing, the MOF cell fit, background-subtracted peak search and the CLI warning output.
  • xtalforge-wasm has a browser suite (tests/browser.rs) that runs with wasm-pack test --headless --chrome; the Node suites are unchanged.

Documentation ​

  • Added 29 diagrams with one shared style under docs/public/diagrams/ (animated SVG; every procedure page: lattice, neighbours, supercell, slab, bonding, rings, P1, CIF, formats, prepare, UFF, optimisation, Ewald, PXRD, indexing, Le Bail, Pawley, Rietveld, extinction, symmetry, nets, MOF, carbon builders, vibrations, limits, validation, agent, measures) and embedded them in the matching algorithm pages.
  • Audited the algorithm and API pages against the code. Corrected the powder instrument description (no wavelength or Caglioti terms), the distance-grid sampling and surface area, the optimiser cell freedom, the ellipsoid scale factors, the error mapping of the CLI and Python bindings, the quick-start version pin and the MOF entry point; recorded the resulting known subsets in the roadmap.
  • Corrected the quick-start (structure_info takes structure JSON; use Structure.from_cif / read_cif for CIF), the WASM size figures, the migration table (no modules CLI command, no separate powder crate) and the stale Phase 0 status and AGENTS.md gap list; documented the Pawley profile-mismatch caveat.
  • Corrected the API examples so every Rust snippet compiles (missing use lines and the Calculator trait, ParamSet::with_param, f64 supercell matrices), the CommandStack::push signature, the error-code tables (limit_exceeded, symmetry), the isosurface algorithm name (marching tetrahedra) and the external feature wording.
  • Corrected the QEq kernel description (hardness only, truncated 12 Angstrom sum) and stated that EEQ equals QEq at the default dielectric, the duplicated external-validation bullet and the tutorial commands (powder lebail/pawley, build graphene/nanotube take positional arguments).

[1.0.0-rc.1] - 2026-11-10 ​

The first release candidate for 1.0.0, published on 2026-11-10 on crates.io (13 crates), PyPI (7 platform wheels plus sdist), npm, the GitHub release and GitHub Pages. The public API is frozen; the stabilization period accepts fixes only, then 1.0.0 follows.

Added ​

  • Phase 8 Task 8.5: distribution metadata. Every published crate has an include allow-list (src, its own data bundle, the agent skill), docs.rs metadata, and scripts/check-package.sh checks the tarballs and the 12 MB ceiling. The release workflow builds abi3 wheels for manylinux/musllinux (x86_64, aarch64), macOS (x86_64, arm64) and Windows; xtalforge gains completions <shell> and man; crates.io publication includes the agent crate.
  • Phase 8 Task 8.6: regression benchmarks for symmetry, UFF, PXRD and the distance grid with scripts/check-bench-budget.sh, determinism tests across threads, a deterministic mutation harness for every text parser and the project archive, and a limits audit at and beyond every budget.
  • Phase 8 Task 8.7: tutorials for eight end-to-end workflows, the Crystal-Builder migration guide, and consolidated algorithm pages for analysis, builders, the MOF builder, the powder pipeline and modules with a coverage index.
  • Phase 8 Task 8.8: the API stability and SemVer page, a compile-time freeze test for representative signatures, and a closed parity matrix with every exception recorded.
  • Phase 8 Task 8.4: the machine-learning bridge and model cache. Python MlCalculator wraps any ASE calculator (MACE, ORB-v3, MatterSim) and maps energy, forces and stress to eV, eV/Angstrom and GPa; ml_backends() probes the modules. The CLI models list|fetch|verify|remove and xtalforge_ff::models record URL, license and SHA-256 in a user cache and never bundle weights. MACE-MP-0 (small) golden data covers H2, water and methane within 1e-6 eV, 1e-5 eV/Angstrom and 1e-6 GPa, with timings.
  • Phase 8 Task 8.3: open extension registries and a Python plugin surface. FormatRegistry, ScalarRegistry and StyleRegistry join the existing calculator and module registries; examples/plugin is a buildable crate that registers a format, a calculator, a module, a scalar and a style. Python Plugins registers callables as modules, calculators and formats, with JSON in/out and typed errors. The draw-style enum stays closed; the open part is the settings preset, recorded in docs/api/plugins.md.
  • Task 0.9: bundle a 14-structure subset of the QMOF Database (CC BY 4.0, Figshare 13147324 v18) and validate the reader against the tabulated atom count, volume, density and formula; four P1 files recover their QMOF space group at symprec 0.01 A, and a CIF round-trip preserves the cell.
  • Task 0.9: bundle an 18-complex subset of tmQM 2024 (MIT, one complex per metal from Sc to Ir) and validate the XYZ reader against the header stoichiometry, the perceived metal coordination and an XYZ round-trip. UFF builds, types and evaluates every metal with finite forces.
  • Task 0.9: bundle an 18-structure subset of the CoRE MOF 2019 public archive (CC BY 4.0, Zenodo 14184621) and validate against it. Every file parses, expands to P1 and bonds; four frameworks are checked against their own _chemical_formula_sum, every declared _cell_volume is reproduced within 0.05 A^3, CIF round-trips preserve sites and volume, and HKUST-1 and ZIF-8 give bounded PXRD and porosity numbers.
  • Phase 8 Task 8.2: xtalforge mcp serves the verb catalogue as MCP tools over stdio (JSON-RPC 2.0, protocol 2024-11-05, no external MCP crate), with generated input schemas and an instructions payload for chat clients. Fourteen skill documents ship with the CLI (xtalforge skill path|list|show|install) and are published as the automation guide. Calls run under the standard size limits; cancellation is process-level and the window proxy is out of scope, both recorded in docs/algorithms/agent.md section 11.
  • Phase 8 Task 8.1: the xtalforge-agent crate: one verb catalogue shared by the Rust, CLI, Python and WASM surfaces (22 verbs), a stateful session whose mutations use Origin::Agent, inspection and diagnostics (overlap, occupancy, charge, dangling and under-coordinated atoms, unusual coordination, symmetry), command-log change bursts, typed JSON answers and a pure-Rust orthographic renderer to SVG and PNG. CLI capabilities, render and agent; Python AgentSession, agent_run, agent_capabilities, agent_verbs; WASM AgentSession, agent_run, agent_capabilities, agent_verbs. The WASM size budget moves from 3648 to 3840 KiB to carry the new exported surface (3775 KiB).
  • Phase 0 Task 0.9: external validation suite. EQeq is compared with the authors' program on the COD MOF-5 cell (424 charges within 1e-3 e), UFF with sci-form on methane and stretched water (within 1e-4 kcal/mol), and the DFTB+ parser with a recorded Si SCC output. docs/algorithms/external-validation.md records the bundled sources, their licences and the data sets still to add.
  • Phase 5 Task 5.6: bundle the full RCSR 2019-06-01 snapshot in data/rcsr/ and identify nets by coordination sequence and point symbol against all 2,931 entries. RcsrIndex::from_embedded() returns RcsrIdentification with the unique name, an ambiguous set, or the nearest candidates; the 2,909 keyed entries are validated against their canonical keys at index build time. The CGD reader now accepts both RCSR edge dialects.
  • Phase 6 Task 6.8: the disordered-carbon recipe builder: convolution surface around a periodic net, split/collapse/flip remesh with tangential relaxation and projection, ribbon paths over the triangle graph, a three-coordinate dual and H/F/O edge termination. CarbonBuild reports density, ratios, ring census, mean C-C, pieces and periodicity; CLI build carbon, Python build_carbon, WASM build_carbon_recipe_info.
  • Phase 7 Task 7.10.1: every leading-cell row now carries space_groups, the candidates whose general-position systematic absences every indexed reflection obeys. The screen enumerates the cell's crystal system, evaluates the structure factor of two generic points per reflection and ranks by group order; CLI, Python and WASM expose the candidates.
  • Phase 7 Task 7.10.1: the automatic table now ranks extinction classes by a Le Bail BIC and Pawley-fits every leading (cell, class), as the reference does. A row carries cell_rank, class_rank, class_symbol, delta_bic and the groups that share the class; the 1 % Rwp tie rule orders the table.
  • Phase 4 Task 4.9: validate the DFTB+ energy, force, stress and population parser against the recorded si_scc_detailed.out fixture. The xTB GFN methods remain available in-process through sci-form (Task 4.10).
  • Phase 1 Task 1.12: rings_explicit and ring_counts_explicit census a builder's stored bonds without distance perception.

Added ​

  • Phase 4 Task 4.10: link the published MIT crate sci-form 2.0.0 behind the optional sci-form feature and implement Calculator for GFN0, GFN1 and GFN2-xTB. The methods are molecular only (a space-group structure is refused), GFN2 returns central-difference forces, and GFN0/GFN1 refuse energy_forces. The registry lists gfn0, gfn1 and gfn2.

Fixed ​

  • Phase 5 Task 5.4: order the four crossings of a marching-tetrahedra 2-2 split around their quadrilateral. The fan taken in the generation order crossed itself, leaving non-manifold edges and a surface that was not watertight.
  • Phase 7 Task 7.6.3: Le Bail extraction returns intensity esds from the profile normal matrix, plus Rwp and the reduced chi-squared; the CLI report includes them.
  • Phase 7 Task 7.10.1: the automatic run fits a Pawley model of every leading indexing candidate and returns a table sorted by Pawley Rwp, with the cell esds; CLI, Python and WASM expose the table.
  • Phase 7 Task 7.8: run the Rietveld refinement in McCusker order (analytic scale/background, zero shift and free cell numbers, width, then Wyckoff coordinates, isotropic ADPs, occupancies and March-Dollase texture) with a bounded direct search per stage, covariance-based parameter esds and the stages list. Coordinates move only along the Wyckoff null space (free_directions), the metric-family cell ties are restored after every move, and the copper K alpha doublet can be placed. A rutile golden pattern from Crystal-Builder validates the recovered position, cell and zero shift.
  • Phase 7 Task 7.6.2: add the Pawley fit with one intensity per reflection, exact linear covariance and intensity esds, free cell numbers, zero shift and size/strain width model, R-factors and chi-squared; a Python-generated golden pattern validates the recovery.
  • Phase 7 Task 7.10: the automatic run indexes the five metric families and adds a Pawley step of the indexed cell; Python powder_pawley and WASM powder_pawley_info expose the fit.
  • Phase 7 Task 7.5.1/7.5.2: add general metric-family indexing (cubic, tetragonal, hexagonal, orthorhombic, monoclinic) with cell refinement, de Wolff M20, Smith-Snyder F_N and a coverage-ranked table; Python powder_index, WASM powder_index_info and powder index --general.
  • Phase 7 Task 7.3.1: add the SNIP background envelope, per-line Gaussian plus linear background fitting and fit_peaks, with --background and --refine-peaks on the CLI.
  • Phase 1 Task 1.15.4: map a non-P1 _geom_bond row from its site labels, operation code and image to P1 contacts (MIL53 matches the 36 golden contacts), collapsing symmetry-equivalent images by their minimum-image vector. Unmatched records annotate nothing instead of inventing an edge.
  • Phase 1 Task 1.15.4: warn when a CIF operation loop's count disagrees with the named space group.
  • Phase 3 Task 3.6.3: add LAMMPS text dump (element column or type map, orthogonal/triclinic, Cartesian/scaled coordinates) and DFTB+ geo_end readers to the trajectory dispatcher.
  • Phase 1 Task 1.15.4: write non-P1 explicit bonds back as a validated P1-index CIF loop and compare the full symmetry-operation set with the named space group, not only its count.
  • Phase 4 Task 4.7: choose overlapping disorder components from the declared CIF formula and cell charge, make the planner graph-aware, add hydrogen filling, the close-contact list, the prepared-structure undoable command path and their tests.
  • Phase 4 Task 4.9.2/4.9.3: wire DftbCalculator and XtbCalculator to the runner, parse detailed.out/xTB stdout, validate force counts and support timeout, cancellation and scratch inspection.
  • Phase 4 Task 4.9.1: add the scratch-folder process runner with wall-clock timeout, cancellation, capped stdout/stderr logs, relative input files and simulated-executable tests.
  • Phase 4 Task 4.7: add duplicate merging, disorder-component choice, solvent removal, charge-balance reporting and the prepared-sample golden sanity pass; PrepareReport now returns the prepared structure.
  • Phase 4 Task 4.8.3: add supercell force constants, shortest-image folding, the acoustic sum rule, a Monkhorst-Pack q-mesh, Gaussian phonon DOS and a phonopy spring-model golden comparison. An explicit q_path samples a band path.

Changed ​

  • The workspace version is 1.0.0-rc.1; the facade now forwards sci-form to xtalforge-ff and xtalforge-agent instead of declaring a no-op.
  • The agent crate joins the published set and the crates.io publication order (after the domain crates, before the facade).
  • The WASM release budget is 3840 KiB (3767 KiB measured).
  • Python wheels are built for manylinux/musllinux (x86_64, aarch64), macOS (x86_64, arm64) and Windows (x86_64), plus an sdist.

[0.0.3] - 2026-11-09 ​

Changed ​

  • Phase 1 Task 1.11.1/1.11.5: perceive bonds with the reference defaults (scale = 1.15, metal-metal pairs forbidden) and validate the full golden corpus: every available bond set and coordination number matches Crystal-Builder after P1 atom mapping.
  • Phase 1 Task 1.6/1.15: expand each site on its own duplicate grid, so disorder components at the same position both survive, and resolve a CIF Hall/H-M symbol into operations when the file omits the operation loop.
  • Phase 1 Task 1.12: replace the chordless-cycle scan with the periodic Franzblau walk, expose Ring::path/ring_paths with lattice images on all four surfaces, and validate counts and paths against the golden corpus.
  • Phase 1 Task 1.15.4: round-trip Uij (matched by label), disorder group/assembly, the oxidation-number charge fallback and the explicit-bond loop; split the CIF writer and structure extraction under cif/.
  • Phase 3 Task 3.6.2/3.6.3: add lazy XyzIndex frame access and the VASP XDATCAR reader with XYZ/XDATCAR detection in the CLI.

[0.0.2] - 2026-11-09 ​

Changed ​

  • Phase 1 Task 1.10.1/1.10.2: replace the direct periodic pair scan with a linked-cell index in fractional space and exact image enumeration, add the 10^3/10^4/10^5-atom benchmark target and compare it with Crystal-Builder.
  • Phase 1 Task 1.10.2: P1 expansion now deduplicates through a fractional grid, removing the quadratic scan that dominated large structures.
  • Phase 1 Task 1.7.1: supercell boundary duplicates use the same fractional grid, cutting the 2x2x2 MOF-5 build from 37.5 ms to 5.5 ms.

Fixed ​

  • Rename the Linux desktop entry to the reverse-DNS component ID so appimagetool finds the AppStream metadata; the AppImage build no longer reports the missing-metadata warning, and the .deb/.rpm payloads carry the matching desktop ID.

[0.0.1] - 2026-11-08 ​

Changed ​

  • Refresh the documentation against the implemented Phase 1-7 surface: the README status, guide, API pages, algorithm pages, validation table, benchmark notes, third-party attribution and roadmap current-status/gap list.
  • Record the remaining documented subsets (neighbour acceleration, phonon q-mesh, preparation steps, general indexing and figures of merit, powder profile simplifications) in docs/roadmap/index.md.

Fixed ​

  • Upgrade zip to 9.4.2, clearing RUSTSEC-2026-0367, RUSTSEC-2026-0368, RUSTSEC-2026-0369 and RUSTSEC-2026-0429.
  • Fix the rustdoc intra-doc links that failed cargo doc -D warnings.
  • Expose C20, twists, ribbons, ZTC/Schwarz P surfaces and polymers in the CLI, Python and WASM, closing the Phase 6 surface-parity gaps; add regression tests for each surface.

Added ​

  • Phase 6 Task 6.1/6.5/6.7/6.8/6.9: SMILES 3D embedding, metal coordination, the sketch model and SMILES writer, commensurate twists, C20/C60 face spirals, gyroid/Schwarz P carbon surfaces and polymer chain packing.

  • Phase 7 Task 7.1-7.11: the rietx inventory and acceptance metrics, the instrument model, whole-range and structural least-squares refinement, free cell parameters and extinction symbols, the combined objective, the automatic powder run, and the CLI/Python/WASM surfaces.

  • Phase 6 Task 6.11-6.16: the xtalforge-mof crate with six topologies, eight blocks, framework assembly, rods, layers, merged nets, decoration, invented linkers and pore-space partition.

  • Phase 6 Task 6.17: mof CLI commands, Python/WASM catalogues and builder, and the MOF API page.

  • Phase 4 Task 4.3.3: native EEQ-style charges.

  • Phase 4 Task 4.6: hydrogen placement.

  • Phase 4 Task 4.7: the preparation pipeline and report.

  • Phase 4 Task 4.9: engine probes plus DFTB+, xTB and ORCA input writers and output parsers.

  • Phase 4 Task 4.10: the sci-form adapter contract and capability limits.

  • Phase 4 Task 4.12: ff CLI commands, Python/WASM force-field surfaces and the algorithm pages.

  • Phase 3 Task 3.7: the .xtalproj ZIP container with independent format_version, readable parts and a newer-version gate.

  • Phase 3 Task 3.8.2: symmetry, pxrd and uff parameter presets shared by the CLI, Python and WASM schemas.

  • Phase 3 Task 3.10.7: SetCharges, Interpenetrate and MarkConnectionPoints commands.

  • Phase 3 Task 3.12.3: in-memory workspace archives for WASM and on-disk workspace binding for Python.

  • Phase 3 Task 3.16: formats, params, project, trajectory and script CLI commands; Python Document/project/workspace bindings; WASM WasmDocument plus project and workspace exports; the editor demo.

  • Phase 4 Task 4.8: Gamma-point Hessian, vibrational modes, IR spectrum, Gamma-point phonons, harmonic/ideal-gas thermochemistry and the quasi-harmonic free-energy scan.

  • Phase 4 Task 4.3-4.5: QEq/EQeq charges, per-site charges, Ewald energy and forces, FIRE/L-BFGS optimization (P1/asymmetric, optional cell), equations of state and relaxed scans.

  • Phase 4 Task 4.1/4.2: Calculator API, registry, run records and the native UFF/UFF4MOF engine with analytic gradients.

  • Phase 5 Task 5.1-5.3: calculated PXRD with radiation, reflections, structure factors, LP and profiles; periodic distance grids and porosity.

  • Phase 6 Task 6.5/6.6: graphene sheets/stacks and nanotube roll-up builders.

  • Phase 3: format registry with capabilities/loss warnings, atomic writes, pymatgen JSON/CSSR/GEN/LAMMPS/SHELX/PDB, CGD nets, Grid3/cube/XY, trajectories, typed parameters, commands/document with undo, selection and tracking, workspace/modules, view data, colormaps and ASCII STL.

  • Phase 3 Task 3.1-3.3: the format registry with capabilities and loss warnings, atomic writes, and pymatgen JSON, CSSR and DFTB+ GEN readers and writers.

  • Phase 2 Task 2.8: layer-group detection for slabs, experimental collinear magnetic groups and molecular point groups.

  • Phase 2 Task 2.10.2-2.10.4: Bravais classification, ASE-derived special points and band paths in the user's reciprocal basis.

  • Phase 2 Task 2.3.5: change of setting by Hall number.

  • Project bootstrap: AGENTS.md rules and versioning policy, roadmap (phases 0 to 8, verified against the Crystal-Builder, sci-form and moyo sources), rule-check script, and VitePress documentation scaffold.

  • Cargo workspace (Task 0.1.1) with xtalforge-core, xtalforge-symmetry, xtalforge-io, the facade xtalforge, xtalforge-cli, xtalforge-python and xtalforge-wasm; shared versions for nalgebra, serde, thiserror, rayon, moyo and flate2.

  • Toolchain, formatting and lint configuration (Task 0.1.2): rust-toolchain.toml, rustfmt.toml and the workspace lints for unsafe code, missing docs, warnings and panic-prone patterns.

  • Repository hygiene (Task 0.1.3): .editorconfig and deny.toml with the license allow list, advisory checks, duplicate-version ban and native-build bans.

  • Facade crate (Task 0.1.4) with stable core, symmetry and io module names, a prelude, and the parallel, sci-form, ml, external and mof-data features documented in the crate and in the guide.

  • Contributor guide (Task 0.1.5) and the Cargo features guide.

  • Rule-check self-test (Task 0.2.1): positive and negative fixtures and a runner covering the 500-line boundary and the ban on warning-suppression attributes.

  • Optional Git hook (Task 0.2.2): scripts/install-hooks.sh and a pre-commit hook running the rules, formatting and clippy on touched crates.

  • CI workflow for the repository rules and their self-test (Task 0.2.4), plus blocking formatting and clippy jobs with warnings denied.

  • Data bundle loader (Task 0.2.3): xtalforge_core::data with embedded gzip bundles, SHA-256 verification of the decompressed payload, once-only decompression and JSON decoding through OnceLock, and lazy external bundles for WASM and large databases.

  • Data policy page and the data/ directory convention, including the gzip fixture and checksum tests for the loader.

  • moyo audit (Task 0.3): requirements matrix, gap decisions, eight-structure spglib comparison at four tolerances, wasm/pyo3 build check, and the report in docs/algorithms/symmetry-backend.md.

  • sci-form reuse audit (Task 0.4): verified inventory, port quality gate, overlap decisions and the sci-form feature policy in docs/algorithms/sci-form-reuse.md.

  • Hello-world slice (Task 0.5): version() and cell_volume() in the Rust API, CLI, Python wheel and WASM package; shared vectors, API pages, parity check, binding architecture and error mapping.

  • Version and release tooling (Task 0.6): scripts/sync-version.sh with a self-test, agent-version check, pull request template and release workflow.

  • Documentation site (Task 0.7): edit links, search, repository link, API and algorithm pages, four-surface code groups, KaTeX, and an offline cell-volume WASM demo.

  • CI workflows (Task 0.8): Rust matrix, Python, WASM, docs, caching and path filters, plus the existing rules job.

  • Golden-data oracle (Task 0.9): curated tests/corpus/ with provenance, Phase 1 generator and data, Crystal-Builder test-data inventory, xtalforge-testkit and the regeneration policy.

  • Dependency replacement map (Task 0.10): import survey, replacement decisions and license status for every Python dependency.

  • Mandatory test layout (AGENTS.md section 2.2): Rust tests live in crates/<crate>/tests/, independent of the source files.

  • Phase 1 Task 1.1: periodic table, deuterium and dummy species, Pyykko radii, symbol parsing, element_info in Rust, CLI, Python and WASM, the generated API table and the shared element vectors.

  • Phase 1 Task 1.2: Lattice with matrix rows, standard orientation, parameters, volume, metric tensor, reciprocal cells, d-spacing and plane normals, coordinate conversions, wrapping, exact minimum-image search, transforms, strain, volume scaling and validation; cell --params in the CLI and lattice_info/lattice_minimum_image in Python and WASM.

  • Phase 1 Task 1.3: the limits model with Standard/Generous/WarnOnly profiles, soft and hard verdicts, drawn/supercell/grid/phonon/Hessian estimates, Error::LimitExceeded, and limits_bounds/limits_check in Rust, Python and WASM.

  • Phase 1 Task 1.4: Site (label, canonical element, occupancy, Uiso/Uij, charge, disorder group, props), Structure/StructureBuilder, molecule in a padded box, formula/mass/density/charge properties, the info CLI bridge and structure_info in Python and WASM.

  • Phase 1 Task 1.5: StructureDocument with SCHEMA_VERSION, strict top-level fields, version checks and a versioned fixture.

  • Phase 1 Task 1.6: SymOp parsing/printing/composition/inversion and P1Cell/P1Atom expansion with site/operation/lattice provenance and special-position deduplication.

  • Phase 1 Task 1.7: diagonal and general supercells as P1, origin shifts, metadata/site-property preservation and the supercell limit check.

  • Phase 1 Task 1.8: surface bases for arbitrary Miller planes, slab construction with layers, vacuum and P1 output, and plane angles.

  • Phase 1 Task 1.9: finite-cell images, Cartesian point transforms, plane fitting/alignment and displacement ellipsoids.

  • Phase 1 Task 1.10: periodic neighbor search with images, distances and deterministic ordering; neighbors and neighbors_of.

  • Phase 1 Task 1.11.1/1.11.4: distance-based bond perception with covalent radii, dummy exclusion, bond images and the connectivity graph.

Changed ​

  • License is now MIT OR Apache-2.0 (confirmed by the maintainer, Task 0.1.3), replacing the initial MIT-only LICENSE with LICENSE-MIT and LICENSE-APACHE.
  • The English-only rule is retired (maintainer decision, 2026-11-08). The language checker, allow list, Spanish fixtures and commit-msg hook were removed; no language check runs in CI.
  • Workspace lints deny warnings and missing docs. scripts/check-rules.sh rejects allow/expect attributes for dead code, unused lints and warnings.
  • Commit discipline adopted (AGENTS.md section 2.1): one directory per commit, feat(<scope>): or ref(<scope>): subjects, and the completed tasks listed from hardest to easiest in the body. The optional pre-commit and commit-msg hooks enforce the directory scope and the message format.
  • Workspace unsafe_code changed from forbid to deny so the pyo3 binding crate can relax it locally with a comment, as the roadmap allows.
  • GitHub Actions workflows are committed but disabled at repository level by maintainer instruction; they have not run yet.

Last updated: