Changelog

All notable changes to Nucleide are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Releases are cut with scripts/bump-version.sh X.Y.Z, which stamps the [Unreleased] section below and updates [workspace.package] version and workspace crate dependency versions in the root Cargo.toml. Git tags (vX.Y.Z) are the release source of truth; CI publishes Python wheels and workspace crates from tags.

[Unreleased]

[0.16.0] - 2026-09-18

Added

• Per-species ion temperatures on single-fuel parametric configs in nucleide-plasma-source (exposed through the species_temperatures dict without a fuel dict): single-fuel D-T reacts at the landed mass-weighted T_DT = T_D + (2/5)·(T_T − T_D), single-fuel D-D at T_D, with the Ballabio lines following per branch (D-T at T_DT, D-D at T_D). Equal pair temperatures (T, T) reproduce the shared-temperature single-fuel kernel bit-for-bit (strengths, sampled stream, cards). Anchors: (T_D, T_T) = (20, 30) keV gives T_DT = 24 keV with ⟨σv⟩_DT = 5.414667327922193e-22 m³/s and the D-T line at mean 14.077934372230146 MeV, sigma 0.3700390508551786 MeV. Mixture paths are untouched; the None-pair path keeps the landed expression trees verbatim (no new RNG draws); the deuterium tail still requires a fuel mixture.

• Sublet S4 ingestion and S5 inhalation hazard kernels in nucleide-alara-io (one shared HAZARDS shape): each folds per-nuclide activities with the caller-supplied 50-year committed dose coefficients (e_ing / e_inh, Sv/Bq, never vendored) into the TOTAL ... HAZARD FOR ALL MATERIALS dose (Sv) plus the excluding-tritium companion, with no split key. Pure arithmetic over caller inventories; negative / non-finite inputs and overflow are loud errors. Exposed as nucleide.alara.alara_ingestion_hazard / alara_inhalation_hazard (entry keys nuclide, activity_bq, e_ing_sv_per_bq / e_inh_sv_per_bq), with hand-vector gates in tests/test_alara.py and C9 rows in validation/clearance_vs_pypact.py.

• Sublet S3 gamma dose-rate kernel in nucleide-alara-io (new dose module): the slab dose C·B/2·Σ μa/μm·Sγ (B = 2) and the point dose C·Σ μa/(4πr²)·e^(−μr)·mₛ·Sγ (Sv/h, C = 3.6e9·|e|, Sγ = I·A) over caller gamma groups — specific activity, group yields, and air/mixture attenuation all caller-supplied, never vendored (mixture_mu folds elemental values with mass fractions). Point distances below 0.3 m clamp to 0.3 m and report it loudly via the clamped flag, never silently. Exposed as nucleide.alara.alara_dose_slab / alara_dose_point / alara_dose_mixture_mu, with hand-vector gates in tests/test_alara.py and C8 rows in validation/clearance_vs_pypact.py.

• Sublet S6 transport-ratio and S7 IAEA clearance-index kernels in nucleide-alara-io: S6 folds per-nuclide activities with the caller-supplied A2 limits (TBq, never vendored) into the dimensionless Total Bq/A2 ratio plus the effective A2 it defines (nucleide.alara.alara_transport_ratio, entry keys nuclide, activity_bq, a2_tbq); S7 folds activities with the caller-supplied IAEA levels (Bq/kg, never vendored) and the total mass into the dimensionless clearance index with its <= 1 screening class (boundary included) and dominant contributor (nucleide.alara.alara_iaea_clearance_index, entry keys nuclide, activity_bq, limit_bq_per_kg). Pure arithmetic; bad inputs and overflow are loud errors. Hand-vector gates in tests/test_alara.py and C10 rows in validation/clearance_vs_pypact.py.

• He/dpa ratio uncertainty propagation in nucleide-damage (he_dpa_ratio_uq, exposed as nucleide.damage.he_dpa_ratio_uq): per-draw ratios over the seeded joint [flux, he, dpa] MVN block (the landed linalg engine, no new sampling machinery), reporting the draw mean ± draw standard deviation with k-SE gates against the second-order bias-corrected expectation and the first-order delta-propagated standard deviation, plus the distribution-free 68% interval (draw 16th/50th/84th percentiles) gated against the Fieller-construction quantiles — the all-regime verdict that holds at honestly-sized blocks where the symmetric-spread gate cannot. Any draw at non-positive dpa fails loudly with the landed ZeroDpa vocabulary, never inf/NaN with a spread. The single-response fold_uq("he_dpa_ratio") spelling stays a loud error naming the two-response function. Gates in tests/test_damage.py and G6–G7 rows in validation/damage_vs_specter.py.

• Runnable workflow notebooks in notebooks/ (first-wall damage + UQ, activation screening + radiological totals, tokamak source sampling + card emission): synthetic inputs and committed fixtures only, fixed seeds, no outputs vendored in the files, executed end to end in CI (tests/test_notebooks.py, needs nbclient/nbformat/ipykernel), with Colab badges for one-click runs once the PyPI release is live.

• Tutorial code-execution gate (tests/test_tutorial_code.py): every runnable ```python block in docs/tutorials/python/*.md executes in CI in document order, so the tutorials double as verified examples. Placeholder-path snippets now point at real fixtures (or carry an explicit skip marker for user-side/h5py/scratch-file demos), and the fetch_compendium tutorial snippet reads the committed compendium file instead of downloading.

• Interactive demos for the new surfaces (new tabbed radiological-totals demo covering S3–S7, He/dpa ratio-UQ block in the damage demo, single-fuel species-pair inputs in the fusion-source demo, all backed by new thin WASM endpoints), plus theory-page coverage of the new equation sets, tutorial prose, and citations — including verified references for the previously uncited equilibrium (VMEC, DESC, simsopt) and blanket (Stellaris concept paper) theory pages.

• Tutorial sidebar reorganization: one workflow-grouped reading order shared by the sidebar, the index pages, and the docs index (duplicate and missing position values fixed).

Fixed

• Target-less <decay> chain entries now parse as decay out of the modeled chain (pure diagonal loss, no gain term) — the decay analogue of the long-standing target-less <reaction> convention, including target="nothing". Previously the committed fixtures/depletion/chain_ni.xml was rejected outright even though its target-less decays (Fe55→Mn55 and friends) are physically just truncated chains. An explicit target naming a nuclide absent from the chain stays a loud UnknownNuclide error; CRAM and Bateman agree exactly on the pure-loss exponential.

[0.15.0] - 2026-09-18

Added

• OpenMC statepoint tally bridge (nucleide.openmc, pure Python — no Rust changes). Statepoint files are HDF5 and are never read in Rust; the facade drives the caller-side OpenMC Python API (call sequence pinned against OpenMC 0.16.0, the validation-container build: StatePoint(path, autolink=False) with an explicit close(), then tally.get_reshaped_data(value=..., expand_dims=True) for "mean" and "std_dev") and returns plain nested lists the landed folds already accept: tally_arrays / read_statepoint yield [voxel][group] flux plus matching rel_err (std_dev / mean, 0.0 where the mean is 0) and the G+1 MeV energy edges ([] unbinned — a single total group), with cell_flux / voxel_flux selecting the group vector the damage folds take alongside the edges as bounds. tally_to_csv / tally_from_csv move the same dict through a commented CSV with stock stdlib only, so an OpenMC machine exports and any machine reads back bit-identically. Mesh tallies are RegularMesh-only (C-order voxels, centimetre bounds); a missing file, missing OpenMC, missing tally/score, multi-nuclide tallies, multi-bin non-energy filters, a second mesh/energy filter, and non-regular meshes are all loud errors naming the remedy — never silent empties. OpenMC stays a caller-side lazy import (the base package never depends on it). Gates are synthetic-statepoint unit tests in tests/test_openmc_bridge.py (pinned fetch sequence, mesh C-order and unit conversion, singleton cell-filter transparency, CSV round-trips, loud-error vectors, and the cell_flux → nrt_dpa feed check).

• Wall-load mapping in flux coordinates in nucleide-equilib-io (exposed through nucleide.equilib as wall_load): closed-form per-cell accumulation q[j][k] = Σ_i S·J·Δs of a caller birth-rate density field onto the caller wall surface sharing the same flux coordinates (radial-major voxels, one-field-period toroidal convention, Jacobian voxels from the landed fourier_jacobian kernel) plus the angle-weighted one-field-period total, which conserves the discrete births to roundoff. Pure glue — no transport, no shadowing, no FEM/thermal, no CAD, no HDF5; shape mismatches and non-finite or negative densities/Jacobians are loud named errors reusing the crate vocabulary. Gates are hand-computed toy geometries at 1e-12 relative (axisymmetric uniform field vs the analytic S·J0 wall flux with the 12π² total, a cosine-Jacobian vector, radial weighting, and discrete conservation) in wall.rs unit tests and tests/test_equilib.py, with always-run E9–E10 rows in the same equilib validation report.

• Closed-form ECRH accessibility kernel in nucleide-plasma-source (exposed through nucleide.plasma_source as ecrh_scalars / ecrh_accessibility): cold electron-cyclotron resonance (B_res = 2π·m_e·f/(n·e)), the weakly-relativistic Maxwell–Jüttner shift (B_res(T_e) = γ·B_res, γ = 1 + T_e/511 keV), and the O1/X1 cut-off densities (n_c,O1 = ε₀·m_e·(2π·f)²/e², n_c,X1 = n_c,O1·(f − f_ce)/f, evanescent None where f ≤ f_ce) from the published ECRH/dispersion facts (Bornatici et al., Nucl. Fusion 23 (1983) 1153; Stix, Waves in Plasmas (1992) Ch. 1), located by linear interpolation along a caller beamline (s [m] strictly increasing, B [T], n_e [m⁻³]) into plain-float resonance/cut-off positions — the per-port penalty inputs for blanket bookkeeping. Pure functions with named loud errors (empty/ragged/non-monotonic beamlines, negative densities, bad frequency/harmonic/temperature); no ray tracing, no launcher design, no HDF5. Gates are hand-computed vectors at 1e-9 relative (170 GHz ITER-class resonance/shift/cut-offs plus beamline crossings) in ecrh.rs unit tests and tests/test_ecrh_access.py.

• Sublet S1+S2 radiological totals in nucleide-alara-io (exposed through nucleide.alara as alara_total_activity / alara_decay_heat). S1 sums per-nuclide activities Ai = Ni λi (Bq) with the FISPACT-II IRT α/β/γ split (IRT 4 → alpha; IRT 1, 2, 11, 14, 16, 17, 19, 20 → beta; IRT 3 → gamma; IRT 12, 13 split α/β and IRT 15 splits α/γ by caller-supplied branch fractions); S2 folds caller-supplied average decay energies into per-class decay heat Ai·E·C1 (kW, C1 = eV→kJ). Both carry the excluding-tritium companion and conserve parts-to-total exactly; unmapped IRTs and bad values are loud named errors. Pinned to the open CCFE-PR(16)53 preprint (Table X, Tables VI–VII) and the open FISPACT-II output_interpretation reference — never the paywalled journal pages; decay energies stay caller-supplied, never vendored. S3–S7, ICRP/IAEA coefficients, bremsstrahlung, DPA/KERMA/gas, and fission counts stay out. validation/clearance_vs_pypact.py gains always-run C7 gates (hand-computed vectors at exact equality) in the same clearance report.

• TBR and blanket power bookkeeping (nucleide-blanket, nucleide.blanket): raw TBR from caller (bred, source) tallies, multiplicative per-port coverage haircuts, blanket energy multiplication, the tritium burn rate from pinned 17.6 MeV / 3.0160492 u constants, and the breeding-margin / net-surplus fuel-cycle metrics. Pure arithmetic over caller transport tallies — no transport solving, no geometry optimization, and no coupling into the tritium permeation kernel (which keeps its no-breeding-coupling scope line). Stellaris Point-A values gate the kernel as regression vectors: raw TBR 1.1070, 1.074 after the 3% ECRH-port haircut, multiplication 1.20, burn 416.6 g/day.

• Coil fast-fluence / lifetime analytics in nucleide-damage (exposed through nucleide.damage as coil_fast_flux / coil_fast_fluence / coil_accumulate / coil_lifetime / coil_remaining plus FPY_SECONDS): magnet lifetime bookkeeping over caller spectra — the fast-flux sum above a caller threshold (groups count iff their upper edge clears the threshold; a threshold cutting through a group includes the whole group), fluence as flux × time, piecewise-constant history accumulation, and weakest-link life (min of limit/rate, or limit-minus-accumulated over rate clamped at zero). dpa spectral weighting stays in the landed folds (the NRT/arc fold at one second is the rate the life kernel consumes). Zero rates never breach and are skipped (all zero is infinite life with a None limiting channel); all shapes, signs, and orderings are loud named errors reusing the fold vocabulary. Limit tables are always caller-supplied — published design numbers (3e22 / 1.5e23 m⁻² fast-fluence limits at a 99th-percentile coil fast flux of 9.5e13 m⁻²s⁻¹ → ~10 FPY, limited by the first channel) are gates at pinned tolerance, never defaults. Magnetics, quench, structural analysis, and transport solving stay out.

• Equilibrium data readers in nucleide-equilib-io (exposed through nucleide.equilib as probe_variant / read_wout / jacobian / jacobian_grid / parse_indata plus typed indata_* helpers). A pure classic-netCDF wout reader accepts CDF-1 (43 44 46 01) and CDF-2 (43 44 46 02) per magic probe (ncdump -k must report classic / 64-bit offset) and rejects HDF5-backed netCDF-4 and CDF-5 loudly for facade-side conversion (no HDF5 in Rust); it reads the documented variable lists (nfp, ns, xm, xn, rmnc, zmns, lmns, gmnc minimum with dims radius/mn_mode/mn_mode_nyq, later-use bmnc/bsubumnc/ bsubvmnc/bsubsmns/currumnc/currvmnc, optional mpol/ntor/phiedge/volume_p). The &INDATA namelist grammar covers scalar switches, power-series profiles, and boundary Fourier tables (sliced indices like RBC(0:4,2) are a loud error; NCURR = 1 reads as the I'(s) profile; absent LFREEB reads fixed boundary). Flux-surface Jacobian helpers evaluate the gmnc Fourier sum at points and over one-field-period grids for volume weighting. Reads data, never solves equilibria; synthetic fixtures are built from the published variable lists only (never solver output). Gates are magic-byte accept/reject vectors, Jacobian hand vectors, and INDATA grammar vectors in-crate plus tests/test_equilib.py and the validation/equilib_vs_desc.py oracle (DESC/simsopt legs SKIP loudly outside the container).

• Arbitrary-3D / stellarator birth-rate lattice source in nucleide-plasma-source (exposed through nucleide.plasma_source with kind="lattice"): the caller supplies the full 3D birth distribution as a point list — points of {"position": [x, y, z] [cm], "rate": w >= 0, "ion_temperature_kev": Ti} (relative birth-rate weights, arbitrary global scale; Ti = 0 is the monoenergetic nominal line) — with the landed machinery applied per node (D/T mixture rates, Ballabio spectra, seeded sampling, card emission with drift reports; MCPL projection stays caller-side). The optional field_periods/base_angle pair declares field-period symmetry (both keys or neither): the points are the base-sector cloud, replicated uniformly around the machine axis, with the symmetric total scaled by the period count. A two-point hand vector pins the contract (total 3, mean birth position (100, 0, 25), mono 14.021 MeV line); a single node reproduces the landed point source bit-for-bit; a uniform ring cloud converges to the analytic ring moments; the folded base-sector stream reproduces the expanded full-cloud stream bit-for-bit (including the single-node case: one base node fans out over all copies); cards carry the cloud’s cylindrical-R/vertical/energy marginals with the same drift rows plus a lattice discretization row and still round-trip byte-identically through the typed SDEF reader. Non-finite positions, negative rates, empty or zero-total clouds, and non-symmetric fold inputs are loud named errors (InvalidLattice / InvalidSymmetry). The Stellaris-class precedent (30 radial × 50 poloidal × 100 toroidal point sources, 14.06 MeV monoenergetic) is the named design point; equilibrium solving, transport solving, CAD/DAGMC, tabulated plasma data, and HDF5 stay out. validation/ plasma_source_vs_openmc.py adds always-run P12 gates (hand vectors, single-point recovery, ring-limit closure, fold replication and totals, card round trip, mixture branch fire) and a container-only O12 probe on a dense monoenergetic ring cloud vs the upstream ring geometry and Ballabio helpers.

• Deuterium hot-tail fraction for the D/T fuel mixture in nucleide-plasma-source (exposed through nucleide.plasma_source as the deuterium_tail={"fraction": eta, "temperature_kev": T_tail} dict [keV] on a parametric fuel blend): the one pinned single-tail-temperature non-Maxwellian shape within the Eriksson et al., Comput. Phys. Commun. 199 (2016) 40 arbitrary-distribution framework — the bulk (1 − eta) deuterium at T_D plus a hot tail eta at T_tail, each sub-pair reacting at its own mass-weighted relative temperature (D-T at T_DT/T_DTt, D-D at T_D/T_mix/ T_tail) with the Ballabio lines following per sub-branch through the strength rule, the sampler’s branch roulette, the card energy marginals, and a deuterium tail drift row carrying the tail neutron share. The pinned 70/30 hand vector (T_D = 20, T_T = 30 keV, eta = 0.05 at T_tail = 60 keV) gates the rate at 1e-12 relative; eta = 0 reproduces the no-tail mixture kernel bit-for-bit (strengths, sampled stream, cards — regression gates); the tail requires a fuel mixture and composes with species temperatures and toroidal sectors. Non-finite, out-of-range, or negative tail parameters are loud named errors (NonFinite / InvalidTail / NegativeIonTemperature). The effective-temperature line convention carries the same error stance as the species pair (the residual against the full Eriksson numerical integration is out of scope); the rest of the full Eriksson generalization stays a loud boundary. validation/ plasma_source_vs_openmc.py adds always-run P13 gates (sub-rate-rule moments and branch share, the drift tail-neutron share, zero-fraction recovery, loud errors) and a container-only O13 probe against the upstream quadrature with the NeSST reactivities evaluated at the five sub-pair temperatures.

• Interactive demos for the new surfaces: the fusion-source demo gains a parametric + hot-tail tab, the activation demo a Sublet S1+S2 tab, plus new ECRH, blanket, equilibrium, coil (damage), and wall-load demos. Fixed the WASM sampleFusionSource parametric path, which rejected the kind/n/seed sampling envelope its own dispatcher requires (FusionParametricSpecJson denies unknown fields to keep stray fuel_mixture/species_temperatures/sector keys loud — only the envelope keys are stripped before parsing).

[0.14.0] - 2026-09-17

Added

• Vented-sink recombination internal interfaces for multi-layer tritium stacks in nucleide-tritium (exposed through nucleide.tritium): LayerStack::new now accepts Interface::recombination(rate) alongside the linear Sieverts/Henry laws. A vented gap carries a single face concentration x ≥ 0 with half-cell fluxes J_L = g_L·(c_L − x), J_R = g_R·(x − c_R) (g = 2D/dx per side, no solubility involved) and the desorption sink K_r·x² venting from the face, so J_L − J_R = K_r·x². The bulk matrix is CUT at the gap (block-diagonal); a lone gap closes in closed form x = 2P/(√(Q²+4K_rP)+Q), coupled faces (several gaps, or recombination outer ends alongside) by Newton iteration with the analytic Jacobian — same FV θ-stepper and linalg::tridiag solve, with the per-step close fused into the trap Picard loop exactly like the G6 outer ends. The K_r → 0 limit is a continuous-concentration joint (not a Sieverts law); a non-positive permeation drive is a loud error in the steady state (the transient follows the G6 clamping stance per step). The flux-continuous product law has no spelling by construction (spurious insulated root, symmetry breaking — permanently rejected). Gates G10a–G10g/G11a–G11d (in-crate unit tests, synthetic stacks with hand-derived closed forms, no fixtures/tritium/ changes): 2-layer closed form at 1e-12 with gap residual and desorption-carrying outer balance, K_r limit pins, a 4-layer mixed-law stack (one interface of each law) with flux continuity across both linear gaps, coupled-Newton residuals at the Newton contract, traps on the owning layer’s isotherm, transient θ-balance with the desorption sink to roundoff, dt-halving order bands, late-time asymptote, and the trap-coupled Picard-fusion balance. The Python steady_layers/transient_layers facades accept a {"kind": "recombination", "rate": Kr} dict per gap (bare "recombination" strings fail loudly — the rate has nowhere to go) and report interface_faces (face value per gap, None at linear gaps; [time][gap] in transients) for closing the discrete balance with the desorption term; single-slab and default-stack signatures are unchanged.

• IRDFF-II dosimetry-response registry expansion (nucleide-nuclei, nucleide._internal.parse_irdff_g725): the V2_REACTIONS extension adds the 26 further named MF=3 dosimetry sections of the same pinned IRDFF-II_g725.zip (same URL + SHA-256 pins, same cache, same loud offline/mismatch errors) — the remaining threshold (n,p)/(n,a)/(n,2n) monitors, capture foils, fission chambers, and high-threshold bismuth monitors, for 34 reactions total. MF=10 isomer sections stay out, the default pack stays v1, and the unfold iterators are unchanged. Parser review: the pinned file carries L2=99 (not 0) in three MF=3 head cards (notably v1’s 56Fe(n,p)), now accepted as 0-or-99 and loudly pinned. Gates: synthetic mocked-fetch unit tests plus a container live-fetch gate (loud SKIP when offline) with hand-checked first-response groups and unfold round-trips at pinned tolerance. Nothing IAEA-copyright is vendored.

• MAXED maximum-entropy unfolding (nucleide-unfold, nucleide.unfold.maxed): the fourth and last adjustment family in the crate — Shannon relative entropy against the guess-as-prior over the exponential Lagrange family, driven by an equilibrated Levenberg-Marquardt dual step with a chi-square target (target_chi2, default: the detector count) and a per-iteration trust-region cap. Caller sigmas are the 1/σ² weights, zero measurements carry zero weight (the GRAVEL house rule), and a relatively converged run whose chi-square still exceeds the target is a hard NotConverged (never a partial spectrum). Validation adds always-run U19-U23 gates (determined recovery, counting-statistics underdetermined round-trip, bracketed IRDFF-II analytical shape probes, SAND-II cross fixed points, contract errors) to the same unfold report.

• D(d,p)T proton bookkeeping and the pinned T-T branch normalization for the parametric tokamak plasma source (nucleide-plasma-source, nucleide.plasma_source): ParametricPlasmaConfig::proton_strength_density / total_proton_strength (plus FuelMixture::tt_neutron_coefficient) report the D(d,p)T proton production rate alongside the neutron source under the pinned 50/50 convention — the proton density IS the D-D neutron-branch density bit-for-bit (pure D-D gives exactly one proton per neutron; D-T-only gives exactly none) — while the sampler draws no proton energies, the cards carry no proton distributions, and proton transport stays out of scope. The T-T neutron term is pinned as the third mixture term S = n²·[f_D·f_T·⟨σv⟩_DT + (f_D²/2)·⟨σv⟩_DD + f_T²·⟨σv⟩_TT] (neutron multiplicity 2 folded in, reacting at T_T, matching the upstream openmc-plasma-source rule) but is NOT transported: no Bosch–Hale T-T fit exists (Bosch & Hale 1992 covers only D(d,n), D(d,p), D-T, D-³He), no Ballabio-class T-T line exists (three-body continuum), and the oracle’s vendored tables stay out of scope — a pure-tritium mixture stays a loud error. A mixture with f_T = 0 reproduces the landed two-branch kernel bit-for-bit (regression gates). New proton_accounting(spec) facade (per-neutron ratio plus sector-aware totals for parametric specs; None totals for ring/point specs, which carry no density model). validation/plasma_source_vs_openmc.py adds always-run P9 gates (exact pure-fuel ratios, the 70/30 share vs quadrature, the no-T recovery anchor, the loud pure-T error) and a container-only O9 probe on a tritium-rich 10/90 blend showing the sampled stream sits on the two-branch upstream quadrature while the three-branch reference pulls away by exactly the T-T-predicted scale (protons have no oracle — upstream models neutrons only).

• Toroidal sectors for the parametric tokamak plasma source (nucleide-plasma-source, nucleide.plasma_source): the additive start_angle/rotation_angle pair [rad] (ToroidalSector) restricts births to [start_angle, start_angle + rotation_angle) with uniform birth angles and emission totals scaled by rotation_angle/2π (half torus ×0.5 exactly, quarter torus ×0.25 exactly); the exact full-rotation spelling reproduces the landed full-torus kernel bit-for-bit (same seeded stream, same cards — regression gates). A partial sector adds a uniform angle-bin marginal to the cards (PHI=D4 on SDEF, phi d4 on Serpent) plus a toroidal sector drift row; the SDEF card still round-trips byte-identically through the typed reader (the mcnp-io SDEF subset gains the PHI keyword). Non-finite or out-of-range angles are loud errors (NonFinite/InvalidSector; both keys or neither in the Python spec). validation/plasma_source_vs_openmc.py adds always-run P10 gates (in-sector uniform births, full-rotation stream/card recovery, the PHI marginal round-trip with its drift row) and a container-only O10 probe showing a partial sector leaves the upstream-quadrature (r, z, E) moments untouched.

• Per-species ion temperatures for the parametric tokamak plasma source (nucleide-plasma-source, nucleide.plasma_source): the additive species_temperatures={"D": T_D, "T": T_T} dict [keV] (SpeciesIonTemperatures) reacts a D/T fuel mixture at distinct Maxwellian species temperatures — the D-T branch at the mass-weighted relative temperature T_DT = T_D + (2/5)·(T_T − T_D), the D-D branch at T_D (Eriksson et al. 2016, the full distinct-Maxwellian generalization) — through the emission strength, the sampler’s branch roulette, the card energy marginals, and the axis spectrum summary. The D-T Ballabio line at T_DT is an effective-temperature convention (Table III was fitted to single-temperature Maxwellians; the residual against the full Eriksson numerical integration is out of scope, and non-Maxwellian reactants stay a loud boundary). A uniform T_D = T_T pair reproduces the shared-temperature mixture kernel bit-for-bit (strengths, sampled stream, cards — regression gates). The pair requires a fuel mixture (a single-fuel config with the pair set is a loud error); the profile ion temperature is then unused for rate and spectrum (still validated), while the density profile keeps shaping the emission. Non-finite or negative temperatures are loud errors (NonFinite/NegativeIonTemperature). validation/ plasma_source_vs_openmc.py adds always-run P11 gates (pair-temperature sampled moments and branch share vs quadrature, equal-T stream/card recovery) and a container-only O11 probe against the upstream quadrature with the NeSST reactivities evaluated at the pair temperatures.

[0.13.0] - 2026-09-16

Added

• D/T fuel mixtures for the parametric tokamak plasma source (nucleide-plasma-source, nucleide.plasma_source): a fuel dict {"D": f_D, "T": f_T} on the parametric source spec (the openmc-plasma-source spelling; then reaction is optional and unused) switches the emission model to the Eriksson et al., Comput. Phys. Commun. 199 (2016) 40 mixture rate at the shared profile ion temperature, S = n²·[f_D·f_T·⟨σv⟩_DT + (f_D²/2)·⟨σv⟩_DD] — the standard n_a·n_b·⟨σv⟩ rate with the 1/(1+δ_ab) same-species guard, the DRESS rate construction upstream openmc-plasma-source implements. Birth energies run a per-particle branch roulette between the two Ballabio lines; emitted cards carry the same product-form marginals, and the magnetic-axis spectrum summary reports the two-branch Gaussian-mixture moments. The equimolar (f_D = f_T = 1/2) and pure-D-D (f_D = 1) limits reproduce the landed single-fuel kernels bit-for-bit (regression gates), and non-finite, negative, or non-summing fractions are loud named errors. Toroidal sectors stay out of scope. validation/plasma_source_vs_openmc.py adds always-run analytic gates P8 (pinned 70/30 blend: sampled moments vs in-script mixture quadrature, D-D branch share) and a container-only O8 cross-check against the upstream map/profiles with the NeSST D-T and D-D reactivities.

• Spanish CSN conditional-clearance tables for NORM landfill disposal in nucleide-alara-io (exposed through nucleide.alara as alara_clearance_es_table). The Consejo de Seguridad Nuclear draft technical opinion CSN/PDT/AICD/TGE/2503/02 (TGE/VAR/2025/1, hosted on csn.es, accessed 2026-09-16 — RD 1029/2022 and RD 1217/2024 (RINR), both transposing Directive 2013/59/Euratom) publishes Tables 1-3, per-landfill type (inert / non-hazardous / hazardous) clearance levels in Bq/g for each NORM material nature (rocks / ashes / sands / slags / oil & gas), keyed by the Table 4 natural decay-chain keys. All three tables are vendored as attributed embedded TSVs in the EU-table pattern (try_ + cache split, row-count-pinned tests); the Table 4 chain keys expand to per-member entries at the parent value through the nuclei dialect machinery (EU Part-2 precedent), with the last claiming key supplying a shared member’s level, as the source lists each secular-equilibrium chain before its subchains. The caller selects the table and material column explicitly — no cross-table logic, no “most permissive wins”; the EU Annex VII table stays the vendored default. Screening arithmetic only, never a compliance decision. validation/clearance_vs_pypact.py gains always-run C6 gates (transcription spots, chain-expansion members, hand-computed screening vector at exact equality).

• Henry internal interfaces for multi-layer tritium stacks in nucleide-tritium (exposed through nucleide.tritium): LayerStack::new now accepts Interface::Henry alongside Interface::Sieverts. The linear Henry law (c/K_H continuous, flux continuous) shares the Sieverts resistance form — face flux J = (c_i/K_i − c_{i+1}/K_{i+1})/R_f with R_f the two half-cell resistances dx/(2·D·K) and the layer solubility playing K_H — so it folds into the same tridiagonal θ-step matrix with no new solver machinery; recombination internal interfaces stay loud UnsupportedInterface errors (recorded limitation, deferred to its own spike cycle). Gates G9a/G9b (in-crate unit tests, synthetic stacks with hand-derived series-resistance closed forms, no fixtures/tritium/ changes): a 2-layer all-Henry stack against the series-resistance closed form at 1e-12 relative on the fluxes (1e-18 absolute floor) with interface flux continuity and half-cell-corrected interface potentials to roundoff, and a 3-layer mixed Sieverts/Henry stack holding flux continuity across both interface laws against the same closed form; the G7c one-layer recovery anchor is unchanged. The Python steady_layers/transient_layers facades gain an additive interfaces keyword ("sieverts" per gap by default, "henry" supported, "recombination" fails with a clear error); single-slab and default-stack signatures are unchanged.

• Vendored SPECTER Table VII fallback in nucleide-damage (exposed through nucleide.damage as specter_table / specter_damage_energy / specter_ed / specter_spectra): spectrum-averaged damage-energy cross sections in keV-b for 24 elements (Be through Pb) across the report’s 7 spectra (thermal / fission / 14 MeV / HFIR / EBR-II / FFTF / fusion), transcribed from Greenwood & Smither, ANL/FPP/TM-197 (January 1985, US-government public domain — OSTI purl/6022143, the same hash-pinned PDF the validation oracle downloads) by reading the scan pages directly, plus the Table II E_d column (eV) per element. Displacement cross sections in barns follow the report’s own 0.8/2E_d rule with the vendored E_d (or a caller-chosen value). Scope is displacement XS only (no gas columns); the group-wise Appendix A printouts are image-only pages in the scan and stay untranscribed, so the table covers the 24 Table VII elements rather than the 41 Table I entries. The table is an opt-in caller input only — the fold kernels are unchanged and never consult it implicitly; the try_ + cache constructor split, row-count pin (24), and spot tests follow the EU-table pattern. The HFIR column reproduces the oracle’s Table VI Fe/Ti/Cu spots at Table VII print precision (3 significant figures, 5e-3 relative); the oracle script itself is untouched and keeps passing.

• STAYSL-class damped least-squares spectral adjustment in nucleide-unfold (exposed as nucleide.unfold.staysl, second method in the one-method-per-cycle order after SAND-II): the Perey ORNL/TM-6062 (1977) anchored-damped objective φ† = argmin_φ Σ_i (c_i(φ) − N_i)²/σ_i² + λ·Σ_j (φ_j − φ_guess,j)², re-solved each cycle as one weighted least-squares system through the shared linalg::lstsq kernel (no second least-squares implementation) on the current active set, with non-positive groups clipped to zero and pinned NNLS-style. Sigmas are caller-supplied, one finite positive value per detector (w_i = 1/σ_i²); zero, non-finite, or missing sigmas are loud named errors. Damping λ is absolute Tikhonov damping on the distance to the guess; convergence is the shared contract (per-group relative-change tolerance with an explicit iteration cap; exhausting the cap is a hard NotConverged, reusing the landed variant). Gates are synthetic forward-fold-then-recover round-trips at pinned tolerances (determined recovery, sigma-weighted precise-reading dominance, underdetermined rate reproduction, IRDFF-II analytical benchmark-field shapes) plus the cross-method fixed-point check (a SAND-II fixed point is a least-squares fixed point) and the named-contract errors; the always-run validation/unfold_vs_analytic.py oracle gains gates U9-U13 in the same unfold report (no SECTION_ORDER change).

• GRAVEL chi-square-weighted spectral adjustment in nucleide-unfold (exposed as nucleide.unfold.gravel, third method in the one-method-per-cycle order after SAND-II and STAYSL-class): the Matzke PTB-N-19 (1994) multiplicative iteration φ_j ← φ_j·exp(Σ_i W_ij·ln(N_i/c_i) / Σ_i W_ij) with W_ij = (R_ij·φ_j / c_i)·(N_i²/σ_i²) — the SAND-II base rate share times the caller-sigma measurement-weight factor — as an iterator with the same diagnostics shape (spectrum/rates/rate_factors/ iterations/tolerance/max_rel_change). Sigmas are caller-supplied, one finite positive value per detector; zero, non-finite, or missing sigmas are loud named errors, never silent uniform weighting. Convergence is the shared contract (per-group relative-change tolerance with an explicit iteration cap; exhausting the cap is a hard NotConverged, reusing the landed variant). Pinned divergences from SAND-II: zero measurements carry zero weight and are simply not fitted (they pin nothing to zero); at a fixed point every rate ratio is 1, so the weights are irrelevant and SAND-II, STAYSL-class, and GRAVEL fixed points coincide where the formulations agree (cross-checked both directions). Gates are synthetic forward-fold-then-recover round-trips at pinned tolerances (determined recovery, counting-statistics-sigma underdetermined rate reproduction plus error reduction) plus the IRDFF-II analytical benchmark-field shapes (Trkov et al., Nucl. Data Sheets 163 (2020) 1 — published facts used with citation, not vendored data); the always-run validation/unfold_vs_analytic.py oracle gains gates U14-U18 in the same unfold report (no SECTION_ORDER change). MAXED stays recorded for a later cycle.

• IRDFF-II runtime-download response pack in nucleide-nuclei (nucleide.data.fetch_irdff + nucleide._internal.parse_irdff_g725): the official IAEA IRDFF-II_g725.zip distribution is fetched at runtime and hash-pinned (SHA-256 recorded in nucleide.data, FGR-15 precedent — nothing IAEA-copyright is vendored), and the IRDFF-II.g725 member text is parsed into caller-ready response rows over the SAND-II 725-group structure (groups 726 eV bounds plus one cross-section vector per named reaction) for the v1 foil-activation subset (au197_ng, in115_ng, u235_nf, u238_nf, fe56_np, ni58_np, al27_na, na23_n2n; MF=3 sections only — the pointwise MF=10 isomer reactions stay out). The rows feed unfold.sandii/staysl/gravel unchanged; structural problems, missing/duplicate sections, off-structure energies, and row-shape violations are loud errors. Offline or hash-mismatch fetches are loud RuntimeErrors naming the URL or both digests.

[0.12.0] - 2026-09-16

Added

• Clearance / waste-classification analytics in nucleide-alara-io and nucleide-fispact-io (exposed through nucleide.alara and nucleide.fispact). The clearance index CI = Σ Aᵢ/CLᵢ and the sum-of-fractions screening rule (RS-G-1.7 §5, referenced by designation) run as pure arithmetic over parsed (nuclide, activity) inventories — Sublet et al., NDS 139 (2017) 77 is the methodology reference; further indices from that paper stay recorded, not implemented. Limit tables are caller-supplied, with the EU 2013/59/Euratom Annex VII Table A vendored as the attributed default (official legal text transcribed from EUR-Lex CELEX:32013L0059, OJ L 13, 17.1.2014, accessed 2026-09-15; reusable per Decision (EU) 2011/833; Bq/g == kBq/kg; IAEA tables never vendored). Nuclides key through the nuclei dialect machinery; missing limits and bad values are loud named errors; the boundary CI == 1 classifies as satisfied (“does not exceed”). Screening arithmetic only — never a compliance decision. fispact-io gains the clearance-block reader for the real FISPACT-II wide inventory table printed with the HAZARDS + CLEAR keywords (grammar cross-checked against the Apache-2.0 fispact/workshops reference outputs): per-step (interval, time_s, cooling) rows with activity, per-nuclide clearance index, flags, and half-life (Stable → -1). Validation adds validation/clearance_vs_pypact.py: always-run analytic gates C1-C5 (exact CI vectors, boundary probes both sides, EU table spots, fixture parse, end-to-end screening) plus a pypact (Apache-2.0) cross-check of the overlapping inventory columns with loud SKIP outside the oracle environment.

• GDML (Geant4) emission direction for nucleide-csg-xlate: scoped MCNP CSG decks now translate to GDML documents (deck_csg_to_gdml in Rust, nucleide.mcnp.parse_csg_to_gdml / read_csg_to_gdml on the Python side) alongside the OpenMC/Serpent/PHITS directions, over the same v3 scope (surfaces, cells, nested universes, rectangular LAT=1 lattices, material stub). Cells become named boolean solids inside a world volume, universes become assemblies, lattices expand to one placement per element, and infinite half-spaces are bounded by a per-deck cutoff recorded in the drift report (new actions halfspace-bounded, lattice-expanded, material-stub). Emitted documents pin the GDML schema version 3.1.7 (Geant4 v11.4.2 tag); the validation harness validates them against the runtime-fetched, SHA-256-pinned XSD wherever lxml is available. Reflecting and periodic boundaries stay loud errors in this direction (Geant4 expresses boundaries through wrapper code, not geometry markup). New fixtures/mcnp/inp/deck_csg_cylinders.txt extends the deck_csg_* family.

• Parametric tokamak plasma source in nucleide-plasma-source (exposed through nucleide.plasma_source with kind="parametric"): Miller-geometry flux surfaces (Fausser et al., Fus. Eng. Des. 87 (2012) 787 — major_radius/minor_radius/elongation/ triangularity/shafranov_factor) carrying caller-supplied L/H/A-mode density and temperature profiles (mode, pedestal_radius, and the ion_density_*/ion_temperature_* parameter sets; profiles are inputs, never computed), with reactivity-weighted emission (Bosch & Hale, Nucl. Fusion 32 (1992) 611, Atzeni–Meyer-ter-Vehn parametrization; new nucleide.plasma_source.reactivity exposes ⟨σv⟩ in m³/s). Birth positions are drawn ∝ S·R·|J| — closed-form Miller Jacobian with the toroidal R factor — and energies from the local-ion-temperature Ballabio Gaussian, all through the seeded sampler. Source cards for parametric sources carry the radial/vertical/energy marginals as discrete histograms (RAD/EXT/ERG distributions on SDEF, rad/ ext/erg on Serpent src) and the drift report quantifies the tabulation truncation plus the joint-correlation distance a product-form card cannot carry; SDEF cards still round-trip byte-identically through the typed reader. Analytic gates: J == r torus limit, J == κ·r·(1 − 2·esh·r·cosθ/a²), area π·κ·a², volume 2π²·κ·a²·R₀ (any Shafranov shift), finite-difference Jacobian agreement at general triangularity, profile pedestal continuity, reactivity transcription goldens, and sampled-moment gates vs fine quadrature. The validation oracle adds always-run P5–P7 and container-only O4–O7 legs against openmc-plasma-source’s Miller map, Fausser profile functions, and NeSST reactivities (loud SKIP outside the container). Loud boundary: fuel mixtures (Eriksson-weighted reactant distributions), toroidal sectors, and the D(d,p)T proton branch stay named NotYetSupported errors.

• Tokamak fusion neutron sources in the new nucleide-plasma-source crate (exposed as nucleide.plasma_source): ring and point sources over the D-D (2.45 MeV) and D-T (14.1 MeV) reactions with ion-temperature- broadened Gaussian spectra (Brysk, Plasma Phys. 15 (1973) 611; Ballabio et al., Nucl. Fusion 38 (1998) 1723 Table III coefficients), a seeded sampler to particle vectors (position/direction/energy/weight arrays, deterministic under a pinned seed), and MCNP SDEF + Serpent src card emission with a drift report (rel_drift of the tabulated spectrum, reparsed flags). Emitted SDEF cards round-trip byte-identically through the typed nucleide-mcnp-io reader, whose accepted subset now also carries the ring keywords AXS/RAD/EXT; Serpent rows are analytic by design (no Serpent source reader). MCPL projection stays caller-side (vr-tools KDE layering rule): the crate has no mcpl-io dependency. Parametric Miller-geometry plasma profiles, mixed-fuel spectra, toroidal sectors, and the D(d,p)T proton branch are loud named NotYetSupported errors — the second landing’s scope. New-crate checklist applied (workspace/release wiring, gen-reference entries, crate-responsibilities section, README feature table); the two-part validation/plasma_source_vs_openmc.py oracle adds always-run analytic gates (ring/spectrum closed-form moments, determinism, card round trips) plus container-only moment cross-checks against the upstream MIT openmc-plasma-source package (loud SKIP outside the container; the Containerfile gains the pinned oracle layer).

• OpenMC/Serpent weight-window emission in nucleide-vr-tools (exposed as nucleide.vr.emit_openmc_weight_windows / emit_serpent_wwin, taking the same MeshTally + MagicOutput pair as nucleide.vr.magic): MAGIC lower bounds are formatted into an OpenMC settings.xml fragment (<mesh> + <weight_windows> — spelling pinned against the public OpenMC docs §3.29/§3.66 with the sub-element layout the C++ reader parses; energies converted MeV → eV; flat bounds emitted x-fastest per energy group) or into the MCNP WWINP text spelling that Serpent reads via wwin <name> wf "<file>" 2 (user guide §2.2.8.3 — the only wwin FMT whose layout is publicly pinned; the Serpent-native FMT=1 layout is never emitted). OpenMC upper bounds are synthesized as 5 × lower and drift notes report synthesized or null-cell structure; ill-formed inputs (negative/non-finite windows, unsorted energy or mesh bounds, OpenMC tuning parameters outside the reader’s ranges) are loud named errors. Golden-text fixtures and re-parse assertions cover both emitters — the Serpent file round-trips through the existing nucleide-mcnp-io WWINP reader — and always-run structural probes extend the existing magic validation report (container OpenMC load cross-check; Serpent load is a loud SKIP, proprietary and not installed).

• Damage and gas-production metrics in the new nucleide-damage crate (exposed as nucleide.damage): NRT-dpa (Norgett, Robinson & Torrens, Nucl. Eng. Des. 33 (1975) 50–54 — modified Kinchin–Pease with the Lindhard damage-energy partition via the Robinson fit) and arc-dpa (Nordlund et al., Nat. Commun. 9 (2018) 1084, CC BY 4.0 — piecewise efficiency ξ(T_d)) as closed forms over caller-supplied material constants (E_d, b_arc, c_arc; no fitted table is vendored); He/H production in appm and He/dpa ratios by piecewise-constant-per-group spectral folding of caller (flux, response, bounds) slices (nrt_dpa / arc_dpa / gas_appm / he_dpa_ratio). Zero-flux groups contribute exactly 0; negative inputs, non-increasing bounds, and the He/dpa ratio at zero dpa are loud named errors — never inf (the ZeroMaxFlux precedent). fold_uq propagates caller-block uncertainty through the landed linalg MVN machinery (pinned seed, k-standard- error gates against the exact bilinear expectation and the first-order propagated standard deviation; UQ on the ratio is a named-open). SPECTER (Greenwood & Smither, ANL/FPP/TM-197, US-gov PD) is the validation oracle only — validation/damage_vs_specter.py folds report-transcribed HFIR-CTR32 spots (Fe/Ti/Cu dpa; C12/Li7/B10/N14 gas) against the runtime hash-pinned PDF at report print precision, loud SKIP outside the container; no SPECTER table is vendored and ASTM E693/E521 stay designation-only. PKA-spectra solving stays out (the fispact-org PKA evaluator is GPL-3.0, never read). New-crate checklist applied (workspace/release wiring, gen-reference entries, crate-responsibilities section, README feature table); the Containerfile gains poppler-utils for the oracle’s PDF text extraction.

• MCPL particle-list utilities in nucleide-mcpl-io (exposed as nucleide.mcpl.merge_mcpl / extract_mcpl / mcpl_stats / repair_mcpl, all file-based and .gz-transparent): merge_mcpl concatenates compatible files with the first-file header plus a provenance comment (stat:sum comments ride along verbatim; sums are never synthesized or updated; mixed single/double input promotes to double, the lossless direction; any other header-option disagreement is a loud error); extract_mcpl writes an index-range (start/stop) or predicate subset with the source header preserved verbatim; mcpl_stats returns record counts, energy moments, total weight, and a per-PDG-code histogram; repair_mcpl implements the paper-pinned mcpl_repair semantics, recomputing the particle count of a file whose writer never patched it and ignoring a partially written trailing record. Synthetic golden fixtures in fixtures/mcpl/utils/ pin merge concat order, precision promotion, and extract subsets; new always-run U1-U9 gates in validation/mcpl_vs_refs.py cover the same contracts.

• Neutron spectrum unfolding in the new nucleide-unfold crate (exposed as nucleide.unfold): the SAND-II iterative spectral adjustment (McElroy et al., AFWL-TR-67-41, 1967 — US government work, clean-room from the report) as an iterator over a caller-supplied response matrix, measured activation rates, and guess spectrum, with per-group relative-change convergence diagnostics (iteration count, final max change, per-detector measured/folded rate factors). The guess is adjusted by a positivity- preserving weighted geometric mean of per-detector correction factors until the fold reproduces the rates; non-convergence past the explicit iteration cap is a named hard NotConverged error — never a silent partial spectrum (the tritium face-Newton precedent) — and unreachable measurements (zero response rows or rates wiped by pinned-to-zero groups) are a named RatesUnreachable. Every response value and energy-group bound is caller-supplied: IRDFF and other IAEA-copyright libraries are never vendored (a runtime-download pack stays a later decision under the FGR-15 precedent). STAYSL-class least-squares adjustment (on the shared linalg::lstsq kernel), GRAVEL, and MAXED are recorded for later one-method-per-cycle landings. Gates are synthetic forward-fold-then-recover round-trips at pinned tolerances plus the IRDFF-II analytical benchmark-field shapes (Trkov et al., Nucl. Data Sheets 163 (2020) 1 — published facts used with citation, not vendored data); the new always-run validation/unfold_vs_analytic.py oracle replays them (no external unfolding oracle exists, so no container dependency is added). New-crate checklist applied (workspace/release wiring, gen-reference entries, crate-responsibilities section, README feature table).

• Multi-layer tritium permeation in nucleide-tritium (exposed through nucleide.tritium as steady_layers/transient_layers): series stacks of caller-specified layers (thickness, cells, Arrhenius D, solubility K_S, per-layer traps/temperature/source — e.g. W/Cu/CuCrZr first walls) with Sieverts internal interface conditions (c/K_S continuous, flux continuous). The interface fluxes are linear in the cell values and fold directly into the tridiagonal θ-step matrix (same finite-volume stepper and linalg::tridiag Thomas solve), so no interface iteration is needed; outer ends reuse the landed surface taxonomy including recombination ends through the same face-response construction. Henry/recombination internal interface laws are loud UnsupportedInterface errors in v1 (recorded limitation). A one-layer stack dispatches to the landed single-slab kernel and reproduces it exactly. Gates are analytic and synthetic, in unit tests with recorded provenance (no fixtures/tritium/ changes): 2/3-layer series-resistance closed forms at 1e-12 with interface flux continuity to roundoff, property-continuous 2-layer ≡ landed kernel, layered recombination-outlet closed form, transient dt-halving order bands and late-time asymptote (Dirichlet and recombination outlets), and the discrete mass balance to roundoff.

Fixed

• Robustness hardening across the 0.12.0 surface (all loud errors, never silent partials or panics): GDML #n complement of an empty region is ComplementTooComplex instead of a panic; the FISPACT-II clearance reader rejects non-finite/negative inventory values and step times and total_clearance_index reports overflow as an error (now returning Result<f64>); alara-io fraction sums report overflow instead of Ok(inf) and the EU Annex VII loader gains a fallible try_eu_annex_vii constructor with a row-count-pinned test; mcpl-io decoding rejects non-finite particle fields and universal weights on read (mirroring the write path) so corrupt files fail loudly instead of poisoning mcpl_stats; parametric sampling uses a panic-free total-order CDF search plus a fallible try_sample; SAND-II replaces its last expect with a loud error; layered-tritium unreachable! sites become named errors.
• Efficiency: merge_mcpl pre-sizes from declared counts; layered-tritium per-cell layer lookup is a cached-offset binary search; Criterion coverage added for every 0.12.0 kernel and reworked emission path (new unfold, tritium, mcpl, plasma benches; windows_emit and wwinp_to_text cases in the existing vr-tools/mcnp-io benches).

[0.11.0] - 2026-09-14

Added

• EPA FGR 15 external-dosimetry coefficients (EPA 402-R-25-001, July 2025) as a runtime-download feature — nothing from the EPA file is vendored. nucleide.data.fetch_fgr15 downloads the official coefficient zip into the per-user platform cache (%LOCALAPPDATA%\nucleide\Cache on Windows, ~/Library/Caches/nucleide on macOS, $XDG_CACHE_HOME/nucleide or ~/.cache/nucleide/ elsewhere), hash-pinned to SHA-256 71314b3f1d73c197da8b589e74c450f61befce48c66ace6ac1f6e0597560bd91 (mismatch fails loudly — EPA may have revised the file). The new nucleide-nuclei fgr15 module parses the seven Table_4_*.DAT scenario tables (1,252 nuclides × 6 age columns; units string recorded from the header), and nucleide.nuclei gains fgr15_dose_rate / load_fgr15_table / parse_fgr15_table / fgr15_age_index (FGR 15 name spellings H-3, Ba-137m, Sb-124n). FGR 15 is external exposure only and complements — never silently overrides — dose_per_g (HNF-5636/PyNE). validation/nuclear_data_vs_refs.py gains an FGR 15 section (row-count/column gates plus ten hand-transcribed spots at exact equality). Screening-level only — not for safety decisions.
• Fission-yield perturbation consumer: nucleide-linalg perturb_fission_yields now perturbs caller-supplied yield blocks (raw = base * (1 + rel), negatives clamped to zero, rescaled to the incoming block sum — independent blocks sum to 2, cumulative blocks higher; zero-base rows stay zero), closing the last named-open hook in the UQ kernel (the FissionYieldsOpen error variant is removed). The thin Python nucleide.uq.perturb_fission_yields wrapper starts working (signature unchanged); the caller supplies the block and builds rel (tape dY/Y sigmas or correlated draws forwarded from the sampling kernel). New U7 validation gate in validation/uq_lite_vs_sandy.py (theory U8) over a caller-selected U235-thermal subset: perturbed sums exact at the pinned seed, draw moments within the IID k-SE bound.
• Scoped MCNP→OpenMC CSG translation v1 (new nucleide-csg-xlate crate): deck_csg_to_openmc_xml maps deck surfaces, cells, and a material stub (material="void" or the MCNP number) to OpenMC geometry.xml with a per-cell/per-surface drift report (macrobody expansions, complement inlines, reflecting/periodic links, dropped params/cards, each with reasons). Axis planes, spheres, on-axis cylinders, SPH, RPP, and axis-aligned RCC map; cones, quadrics, tori, general planes, other macrobodies, non-flat complements, transforms, lattices, matrix or transformed fills, tallies, and sources are loud named errors, while simple nested universes (U=k, single-universe FILL n) map to universe=/fill=. Thin Python nucleide.mcnp.parse_csg_to_openmc / read_csg_to_openmc facades plus synthetic fixtures/mcnp/inp/deck_csg_*.txt decks and a CSG section in validation/parsers_vs_refs.py (structural probes always run; the OpenMC Region.from_expression cross-check runs in the container).
• Scoped MCNP→Serpent CSG translation (same nucleide-csg-xlate crate): deck_csg_to_serpent_input emits surf/cell cards over the same v2 scope with Serpent-native simplifications (RPP to cuboid, axis-aligned RCC to truncated cylinders, #n passed through as the native cell complement, empty regions via a synthesized inf surface; materials as m<n>/void names for caller-supplied mat cards). Reflecting/periodic boundaries are a loud SerpentBoundaryOutOfScope (Serpent set bc is global; per-surface mapping unverified). Thin Python nucleide.mcnp.parse_csg_to_serpent / read_csg_to_serpent facades plus structural serpent structure probe rows in validation/parsers_vs_refs.py.
• Scoped MCNP→PHITS CSG translation (same nucleide-csg-xlate crate): deck_csg_to_phits_input emits [Surface]/[Cell] sections over the same v2 scope with manual-verified identical symbols (planes, spheres, on-axis cylinders, SPH/RPP/RCC-any-orientation/axis-aligned BOX, coefficients verbatim), native # complements, U=/FILL= params, * reflective surfaces, verbatim densities, and an outer-void -1 heuristic for void union/complement cells (drift-noted). Periodic pointers are a loud PhitsBoundaryOutOfScope. Thin Python nucleide.mcnp.parse_csg_to_phits / read_csg_to_phits facades plus structural phits structure probe rows in validation/parsers_vs_refs.py.
• Rectangular-lattice (LAT=1) CSG translation in all three directions (same nucleide-csg-xlate crate): a LAT=1 cell with a full matrix FILL (every element filled, no transform) now emits a rectangular lattice — OpenMC <lattice type="rectangular"> (dimension, lower_left, pitch, universes; MCNP k, j, i order maps to z-ascending / y-descending / x-ascending row-major), a Serpent cuboidal lat card (type 11) filled via fill <lattice id>, and PHITS LAT=1 with a matrix FILL (ranges plus the universe list in MCNP order verbatim) — with pitch and lower-left derived only from the lattice cell’s RPP or axis-plane-box bounds and lattice ids allocated outside both cell and universe id spaces (each noted as lattice-emitted drift). Hexagonal LAT=2 lattices, 0-holes, non-RPP/axis-plane-bounded lattice cells, single-universe fills of lattice type, and fill transforms stay loud named errors. New synthetic fixtures/mcnp/inp/deck_csg_lattice_rect.txt deck, per-direction universe-order tests, and lattice cross-reference probe rows in validation/parsers_vs_refs.py.
• Tritium-transport analytic-gate spec (no kernel yet): docs/theory/tritium.mdx pins the T1–T2 equation set and the G1–G5 gate contract (steady linear, permeation time-lag, single-trap limits, Sieverts steady, recombination steady; recombination transient stays named-open) plus fixtures/tritium/ oracles for the analytic-gate replay at kernel landing.
• 1D tritium-transport kernel v1 (new nucleide-tritium crate): T1 mobile diffusion with N extrinsic McNabb–Foster trap species (T2), caller supplied Arrhenius data and steady temperature profile (no tables, no heat solve), Dirichlet/Sieverts/Henry/zero-flux/recombination surface taxonomy (recombination steady state closed by the exact face-response construction with G5a/G5b gates; recombination transient stays a named-open G5 error), and a cell-centred finite-volume theta-stepper (Crank–Nicolson default, backward Euler on request) solving through the new shared nucleide-linalg tridiag Thomas-solver module. Analytic-gate replay only (Rust fixture tests plus tests/test_tritium.py on G1–G5 steady at the pinned tolerances; no validation/*_vs_*.py, no FESTIM oracle). Thin Python nucleide.tritium facade, WASM tritiumBreakthrough facade with a breakthrough-curve interactive tutorial.
• Gaussian KDE source sampling (KDSource-class, clean-room) in nucleide-vr-tools: KdeSampler fits an axis-aligned Gaussian KDE over caller particle vectors (Silverman or fixed bandwidths; zero-variance dims are loud) with deterministic draw/pdf (caller randoms, no RNG inside), gated by Gaussian recovery, normalization, and determinism tests. Thin Python nucleide.vr.KdeSampler facade.
• WATTS-class sweep expansion in nucleide-r2s: SweepAxis validation, deterministic cartesian expand_sweep to named case bundles, and assemble_results requiring exact per-case coverage, plus thin Python nucleide.r2s.r2s_expand_sweep facade.
• Facility-flow accounting in nucleide-r2s: snapshot_inventory totals atoms per ARMI bare name (N × V × 1e-24 over snapshot zones) for differencing facility snapshots, plus thin Python nucleide.r2s.r2s_snapshot_inventory facade.
• JADE-class library assessment oracle (validation/library_vs_refs.py, auto-discovered): data hygiene, independent-block sums to 2.0, and cumulative coverage over the committed fission-yield pack (synthetic thresholds, no tapes).
• Legacy SDEF fixed-source reader in nucleide-mcnp-io (new sdef module): typed SdefCard/SdefDist/SdefProblem model over the POS, CELL, SURF, VEC, DIR, ERG, NRM, PAR, WGT, TME keywords (inline literals or Dn references) plus the discrete SIn L/SPn D/ SBn D distribution forms; every other keyword or option letter is a loud drift note or a clean SdefError, never silently misread. Validation covers duplicate cards/keywords, dangling Dn references, orphan SPn/SBn cards, entry-count mismatches, and empty tables; canonical re-emission round-trips the spectroscopy sdef_decay_source card dialect byte-identically (80-column wrapping, C++ defaultfloat precision-6). DeckProblem.sdef typed view (enforced in validate) and thin Python nucleide.mcnp.parse_sdef facade; E9 reader round-trip rows in validation/spectroscopy_vs_pyne.py plus synthetic tests/test_sdef.py round-trip and error-case coverage.
• Tritium recombination transient closure in nucleide-tritium (the last named-open gate in the kernel): solve now accepts recombination ends (J = K_r c_m²) in the transient, closing each implicit θ-step by the G5 affine face-response construction reused per step — one base Thomas solve plus one sensitivity column per recombination end per dt, closed-form face for one end, analytic-Jacobian Newton for two, fused into the trap-Picard loop and sharing rtol/atol. The RecombinationOpen error variant is removed. Since no closed form exists for the recombination transient, asymptotic + self-convergence gates replace the algebraic style: G6a late-time asymptote to the G5a steady flux (1e-6 relative at 60 permeation lags), G6b K_r→∞/K_r→0 recovery of the Dirichlet/zero-flux transients, G6c discrete mass balance with the recombination leak to roundoff, G6d dt-halving θ-method order (Crank–Nicolson ≈ 2, backward Euler ≈ 1), G6e independent method-of-lines cross-check (node-centred central FD + explicit RK4, ghost-node recombination face) of the mid-transient trajectory at 0.5/1/2 permeation lags against the production transient, trap-free and trap-coupled (1.5e-4 relative on the mobile-profile max-norm, ≈5x measured headroom), plus positivity under the face clamp. Python nucleide.tritium.transient facade unchanged (dict shape and signatures stay); the theory page gains the G6 section.

[0.10.0] - 2026-09-14

Added

• Fission product yields from the ENDF/B-VIII.0 neutron-induced and spontaneous fission-yield sublibraries: committed crates/nuclei/src/data/fission_yields.tsv (151,490 rows over 36 parents and 122 incident-energy sets; MF8/MT454 independent + MF8/MT459 cumulative, values verbatim — both sublibraries are unchanged carries of ENDF/B-VII.1 evaluations per the tapes’ README.txt, with the Mattera–Sonzogni cumulative-yield correction not applied), generated by the new --endf-nfpy/--endf-sfpy flags of scripts/gen-nuclear-data.py. nucleide-nuclei gains FissionYieldOrigin/FissionYieldKind enums plus fission_yield_table/fission_yields/default_fission_yields/ fission_yield lookups (lowest-energy independent set = the OpenMC get_default_fission_yields depletion convention); thin Python nucleide.nuclei.fission_yields/fission_yield facades. The depletion chain XML parser falls back to this library (lowest-energy independent set, products filtered to chain members) when a fissionable parent has no <neutron_fission_yields>, raising the existing BadStructure error when the library lacks the parent. Spot gates in validation/nuclear_data_vs_refs.py pin the table to tape values, with a loud-skip OpenMC cross-check when the nfy tapes are absent.
• Seeded Latin-hypercube sampling (nucleide-linalg sample_lhs over caller-supplied blocks: per-dimension one jittered draw per stratum via Fisher–Yates permutation, hand-rolled inverse-normal CDF, shared Cholesky/eigen-clip factor path with x = μ + Bz; thin Python nucleide.uq.sample_lhs + WASM sampleLhs facades). Ships with its separate validation gate U6 (theory U7) (validation/uq_lite_vs_sandy.py: G1 stratification-exact plus G2 moments within the IID bound as an upper bound only) over synthetic fixtures/uq/lhs_2x2.json. A draw mode, not a perturbation convention (perturb_energies(..., "lhs") stays an error).
• Interactive UQ demo: MVN/LHS draw-mode toggle calling the WASM sampleLhs facade with the same (mean, cov, n, seed) inputs and result shape, with E2E coverage.
• Interactive deterministic demo: the RTFLUX section now renders a chart-only per-point flux profile (one grouped-bar trace per energy group over the capped parseRtflux values, following the ISOTXS grouped-bar precedent) with E2E coverage.
• Interactive deterministic demo: PARTISN writer section rendering and validating structured deck dicts in the browser via new WASM partisnRender/partisnValidate facades (exact title/dim/zones (id/material/isotxs_labels/density)/source keys mirroring the Python partisn_render/partisn_validate shape, no camelCase aliases; zone labels validated against a pasted ISOTXS library via parseIsotxs) with render preview, inline validation-error display, and E2E coverage.
• Interactive activation demo: the ORIGEN TAPE6 per-step comparison now takes a dynamic list of N pasted snapshots (add/remove, each re-parsed with the same parseOrigenTape6 reader) into a per-nuclide series chart over all steps in the existing grouped log-bar bundle, with E2E coverage. Multi-snapshot file grammar stays out — pasted snapshots only.
• Interactive activation demo: the R2S snapshot tab hosts an R2S voxels section over a drift-accepted VoxelTags copy-port in nucleide-wasm (copy/split zone-total tagging plus .photonSrc group select-and-sum over nucleide-alara-io photon types only — no nucleide-r2s dependency, so the Rayon chain stays untouched). Zone totals and voxel maps tag in the browser (200-voxel demo cap) with a scatter/bar/heatmap/histogram chart bundle, with E2E coverage.
• Crate-root error re-exports and Result aliases (API-stability pass, additive half): nucleide-linalg re-exports DecayError / LstsqError / SampleError, nucleide-nuclei re-exports DialectError / ParticlesError / RxnameError, nucleide-depletion re-exports CramError, nucleide-mcpl-io re-exports SswError, and nucleide-fluka-io re-exports MaterialError / UsrbinError (nucleide-mcnp-io stays module-path-only by design); the nine crates lacking one gain pub type Result<T> (crate-root in nucleide-depletion, nucleide-kinetics, nucleide-linalg, nucleide-mcpl-io, nucleide-nuclei, nucleide-spectroscopy, nucleide-vr-tools, per-module in nucleide-fluka-io and all nine nucleide-mcnp-io error modules). #![warn(missing_docs)] now covers 17 of 18 crates — nucleide-mcnp-io (~132 missing docs) is the recorded deferral, owned by a future doc-volume cycle.

Changed

• #[non_exhaustive] on all 38 public error enums across the 18 workspace crates (nucleide-mcnp-io’s nine per-module errors included): the API-stability sweep breaks exhaustive downstream matches on these enums — a wildcard arm is now required — and new variants become non-breaking additions. No data struct or non-error enum was touched, and 154 internal Result<T, Error> respellings to the new aliases change no public signature.

Fixed

• Sampling kernel draw-count cap (nucleide-linalg sample::MAX_SAMPLES = 10M enforced in validate_block as SampleError::TooManySamples → ValueError, covering sample_mvn/sample_lhs/sample_lognormal).
• Corrected inv_normal_cdf docs: inputs at/beyond [0, 1] yield NaN, not ∓inf (unreachable — sample_lhs clamps into (0, 1)).
• WASM UQ facades (uqSample/sampleLhs) reject n < 2 with a moments-need-n >= 2 message instead of the confusing sample_cov DimensionMismatch, and reject seed >= 2^53 (f64 integer precision — the saturating as u64 cast would collide streams).
• WASM voxelTagsFromTotals rejects empty zone_of_voxel (previously a vacuous success); voxelPhotonSums rejects non-finite timeS and empty nuclides; partisnValidate rejects empty zones (previously a vacuous pass) while partisnRender keeps rendering zoneless decks by design (callers validate separately).
• UQ demo clears stale results when switching MVN/LHS and rejects seed >= 2^53 client-side; activation demo treats a blank voxel cooling-time field as an error instead of silently querying 0.0.
• U6 validation gate reports FAIL instead of crashing with ZeroDivisionError on a hypothetical n < 2 fixture.

[0.9.0] - 2026-09-13

Added

• MCTAL mesh tallies (nucleide-mcnp-io mctal headline): parses mesh-tally bodies (detector_type <= -1) — the 4-int mesh f line (unknown ni nj nk, bare-f or f<tally> spellings), (ni+1)+(nj+1)+(nk+1) cora/corb/corc bounds, the shared d/u/s/m/c/e/t cards, and vals pairs over the ni*nj*nk mesh cells (writer-loop order, i fastest). Synthetic fixtures/mcnp/mctal/synthetic_mesh.mctal oracle with closed-form val/err pairing, Rust tests, and Python Mctal.mesh_tallies / mesh_tally_vals_array views. Radiograph (detector_type >= 3), point-detector (tally names ending in 5), and perturbation bodies stay named-open errors.

• MCTAL S-tails: optional tfc blocks after standard tallies (jtf line plus 3–4-float data rows, exposed as tallies[].tfc), total/cumulative card variants (ut/uc/…, stored verbatim in tallies[] card variant), and third-token flags on c/e/t cards (stored verbatim as card flag), with synthetic fixtures/mcnp/mctal/synthetic_tfc_variants.mctal coverage. WWINP cylindrical (nr=16), Serpent UnsupportedExpr, SDEF versions beyond 5/6, and SPE gating stay as-is (loud errors or unchanged surfaces — no second field-order source on disk).

• validation/parsers_vs_refs.py: MCTAL section comparing header scalars plus per-cycle keff/lifetime series against PyNE Mctal on the kcode-only fixtures; body-bearing files (standard, mesh, tfc/variant) are loud SKIPs — PyNE skips tally bodies without advancing past them, so no oracle exists there.

• Interactive completion bundle (WASM thin facades only — no new math): MCPL readMcpl / writeMcpl / ssw2mcpl / mcpl2ssw (bytes-based, gzip magic-sniffed, 200-row cap) with a file-upload mcpl-io tutorial staging the golden ssw2mcpl_expected.mcpl + reference.w pair; material materialSeparate / materialBlend / cusumDetect; spectroscopy parseLinesTsv / energyBins / detectorEfficiency / parseDollarSpe / parsePlainSpe; kinetics inhourRho / stablePeriod / promptJump; deterministic parseRtflux (kind-switch mirrors Python); ORIGEN per-step comparison over parseOrigenTape6 plus chart-only ISOTXS totals, TAPE5 flux-vs-step, and TAPE6 per-nuclide views. Named-open and recorded in AD-13: R2S VoxelTags port (rayon feature unification), PARTISN writer, RTFLUX profile / ORIGEN per-step full views, multi-snapshot TAPE6 grammar.

• Shared dense-real weighted least-squares kernel (nucleide-linalg lstsq module, zero dependency change — faer 0.20 column-pivoted QR solve_lstsq route): weighted_lstsq(X, y, w) over caller Vec<Vec<f64>> inputs (non-negative caller weights, sqrt(w) row scaling, named errors for shape/finiteness/weight/underdetermined inputs) with closed-form synthetic gates. One module, two consumers — the UQ-lite sampler keeps its Cholesky/eigen factor path.

• Efficiency-coefficient fit (nucleide-spectroscopy, E7-fit extension, not a new equation number): fit_efficiency solves the log-space linear least squares (y = ln eff, X[i][j] = (ln E_i)^j for fit == 1 or (1/E_i)^j for fit == 2, caller-supplied weights) through the shared kernel, returning order + 1 coefficients for detector_efficiency. Peak search/fit and FWHM stay out; every other coefficient vector stays a caller input. Synthetic fixtures/spectroscopy/efficiency_fit.json oracle (recovery + round-trip at 1e-9).

• Python API: nucleide.spectroscopy.fit_efficiency thin wrapper over the new fit.

• validation/spectroscopy_vs_pyne.py: E7-fit synthetic recovery + round-trip gates (always run) plus a loud oracle-check SKIP — the upstream module ships no efficiency-fitting routine, so there is no container cross-check.

• Python facade bundle over existing Rust (thin wrappers only — no new math or data): nucleide.nuclei gains rxname_label / rxname_doc / rxname_reaction / rxname_id_from_nucdelta / rxname_child / rxname_parent, particle_is_valid / particle_is_valid_pdc / particle_is_hydrogen / particle_is_heavy_ion, and dose_f1 / dose_lung_model; nucleide.material gains mix_by_mass / mix_by_volume / specific_activity / materials_doc_to_xml / expand_elements / collapse_elements (bare element symbols map to natural-element placeholders); nucleide.fluka gains fluka_material_str / fluka_compound_str / fluka_builtin_set; nucleide.alara gains alara_validate_deck / alara_check_block / alara_flux_total / alara_flux_len / alara_output_totals / alara_output_total_activity / alara_photon_total_strength / alara_schedule_total_time; nucleide.origen gains origen_tape6_find / origen_tape6_total_activity / origen_tape9_find; nucleide.cccc gains cccc_rtflux_npoints / cccc_rtflux_point / cccc_rtflux_total / cccc_isotxs_find / cccc_isotxs_len; nucleide.fispact gains fispact_is_output; nucleide.enrichment gains the six assay mass ratios (prod_per_feed, tail_per_feed, tail_per_prod, feed_per_prod, feed_per_tail, prod_per_tail) plus alphastar_i; nucleide.kinetics gains from_ifp; nucleide.vr gains magic_with plus MeshSourceSampler mode / num_voxels / table_len accessors; nucleide.mcnp is unchanged (dict surface complete); nucleide.mcpl gains mcpl_statsum_validate / mcpl_statsum_comment. depletion/serpent stay internals by design (no wrappers). Covered by tests/test_facade_bundle.py on synthetic inputs and committed fixtures.

• Python tutorials: new docs/tutorials/python/ vr-magic.md (MAGIC + alias-table sampling) and docs/tutorials/python/rxname.md (registry, graph ops, particle helpers), a standalone-SWU section in enrichment-cascade.md, and worked mixing/specific-activity/MaterialsDoc/element-expansion sections in build-materials.md (replacing the “not yet exposed” note); all registered in docs/tutorials/python/index.md and docs/README.md.

• UQ-lite log-normal sampling (log-normal scope only; LHS stays out): nucleide-linalg sample gains a LogNormal perturbation convention (parse("lognormal"/"log_normal"/"log-normal"), apply as nominal * exp(delta)) plus sample_lognormal (shared Cholesky/eigen-clip factor path and ChaCha RNG as sample_mvn, then a y = exp(x) post-transform with log-space mean/covariance semantics) and the closed-form (U6) lognormal_mean / lognormal_cov helpers. Synthetic fixtures/uq/lognormal_2x2.json oracle with pinned seed and separate U5 validation gate (validation/uq_lite_vs_sandy.py: ln(y) MVN recovery plus closed-form mean; existing k-SE gates untouched). Python API: nucleide.uq.sample_lognormal / lognormal_mean / lognormal_cov thin wrappers (perturb_energies(..., "lognormal") now parses).

• SSW↔MCPL fidelity tail (nucleide-mcpl-io ssw; bitarray words stay out — no isurf/rawtype decode, no type-word reverse-engineering without a licensed type-table source; synthetic fixtures only, no transport semantics): mcpl2ssw gains force_cs_to_one (default false keeps the true cosine; opt-in forces the stored cs slot to 1.0 with u/v verbatim, removing the stored-cosine delta vs the 2.2.8 binary), niss_override (None = reference passthrough, the upstream-2.2.8 spelling verified over the synthetic pair; Some(v >= 0) stamps the tally-convention value, negative is NissOutOfRange), and allow_polarisation (default rejects non-zero input vectors with PolarisationPresent; opt-in drops them since SSW has no slot); ssw2mcpl gains opt-in polarisation (uniform vector + has_polarisation, non-finite is InvalidPolarisation), universal_pdg (single-kind inputs collapse file-wide, mixed is MixedPdgForUniversal), and universal_weight (equal weights collapse file-wide, mixed is MixedWeightForUniversal); the SSW-PDG table widens beyond neutron/gamma to electron 11 / positron −11 / proton 2212 with identical verbatim geometry (anything else stays loud UnsupportedPdg, no particle framework beyond the table). Each leg has named errors, synthetic both-direction Rust tests, Python facade options (ssw2mcpl polarisation / universal_pdg / universal_weight; mcpl2ssw force_cs_to_one / niss / allow_polarisation; five accepted kind spellings), committed fixtures/mcpl/ssw_conversion/ goldens (extended, polarised, universal), and validation/mcpl_vs_refs.py gates S5–S8 in the same mcpl report (no SECTION_ORDER change). WASM/demo untouched (no cheap surface).

• Python API: nucleide.mcpl.mcpl2ssw keyword args force_cs_to_one / niss / allow_polarisation and ssw2mcpl options polarisation / universal_pdg / universal_weight over the widened kind table.

[0.8.0] - 2026-09-13

Added

• SSW↔MCPL conversion, neutron/gamma-only v1 (nucleide-mcpl-io ssw module, new nucleide-mcnp-io dependency for the SSW header/track types): ssw2mcpl maps kinetic energy verbatim (MeV), time shakes→ms (×1e-5), surface id→userflags (opt-out), and kind→PDG 2112/22, with single-precision + gzip-compressed-bytes output, an optional ≤100 MiB deck-embed blob, and named errors for other particle kinds; mcpl2ssw clones a reference SSW header (code/version/deck passthrough, nrss/np1/orignp1 patched to the converted tally with the reference’s table-2 sign) with surface id from userflags or an explicit [1, 999999] override, time ms→shakes (×1e5), and direction cosines propagated verbatim. Surface id and particle kind ride in as explicit caller parameters — the local SSW reader has no isurf/rawtype decode, so v1 never guesses them from bitarray. Deferred, all named-open: non-neutron/gamma particles (UnsupportedPdg), polarisation and universal codes, upstream bitarray type-word round-trips, and any transport semantics.
• Python API: nucleide.mcpl.ssw2mcpl / mcpl2ssw thin file-based wrappers over the new module (per-track surf+kind pairing, options dict, .gz transparent), plus a synthetic fixtures/mcpl/ssw_conversion/ oracle pair (hand-framed two-track SSW reference with provenance notes + golden ssw2mcpl output).
• validation/mcpl_vs_refs.py: SSW round-trip gates S1–S4 on the synthetic pair plus a live upstream ssw2mcpl/mcpl2ssw (mcpl-extra 2.2.8) converter cross-check T3–T5 — count and closed-form energies agree both directions; the scripts SKIP loudly when absent, never fail without the oracle.
• UQ-lite sampling kernel, decay-only sub-scope (nucleide-linalg sample + decay modules, new rand 0.9 dependency with OS-entropy features off so the crate still builds for wasm): seeded multivariate-normal sampling over caller-supplied covariance blocks (sample_mvn — Cholesky primary with named cholesky/eigen_clip reporting, eigen-clipping fallback at 1e-12 relative floor, relative / absolute perturbation conventions, sample mean/unbiased-covariance convergence diagnostics with caller-supplied tolerances), plus a decay consumer (perturb_branches preserving the 1-BR(SF) deficit by renormalisation, perturb_energies, finiteness-checked passthrough). Fission-yield perturbation stays a named-open hook (FissionYieldsOpen, waits on ENDF fission-yield tapes); ERRORR/NJOY, MF32/40, transport coupling, and vendored covariance stores stay out. Closed-form covariance-recovery gates on synthetic blocks only (pinned seeds, k-SE statistical tolerances, fixtures/uq/).
• Python API: nucleide.uq thin wrappers (sample_mvn / sample_mean / sample_cov / check_convergence / perturb_branches / perturb_energies / passthrough / perturb_fission_yields) over the new kernel.
• validation/uq_lite_vs_sandy.py: synthetic covariance-recovery gates U1–U4 (always run) plus a live SANDY Samples.get_mean/get_cov moment cross-check over the nucleide draws (tape-free, no NJOY) with a loud SKIP when SANDY/pandas are absent; tape-driven sandy.sampling/ERRORR comparisons stay NJOY-gated skips.
• UQ-lite tutorial + interactive demo: docs/tutorials/python/uq-sampling.md (seeded sample_mvn over the synthetic fixtures/uq/ blocks, convergence diagnostics, decay perturbation) and a thin uqSample WASM facade with an interactive tutorials/interactive/uq sampling demo.

[0.7.0] - 2026-09-12

Added

• Fuel-cycle micro-adds (nucleide-material, nucleide-enrichment): Material::separate splits a composition into product/tails streams by per-nuclide efficiency (finite values in [0, 1], unlisted nuclides go wholly to tails, per-nuclide mass conserved), Material::blend mixes streams at fixed ratios with explicit normalization (errors on empty, all-zero, or negative recipes — never a silent uniform split), and a dependency-free one-sided Page CUSUM change detector (Cusum with Welford running mean/variance, new(ref_shift_k=0.5, alarm_h=4.0, startup=10) tuning plus update/status/statistic/reset).
• Python API: nucleide.material.separate_material / blend_material / Cusum and nucleide.enrichment.value_func / swu_per_feed / swu_per_prod / swu_per_tail thin wrappers over the new core.
• validation/enrichment_swu_vs_cyclus.py: two-tier SWU oracle — hand-recomputed closed-form gates on the cyclus assay ladder (feed 0.0072, product 0.05, tails 0.002, 10 units of product) plus a live cyclus.toolkit.enrichment cross-check with a loud SKIP when cyclus is absent.
• Full MCTAL tally bodies (nucleide-mcnp-io): the standard (non-mesh, non-radiograph) per-tally layout — f/d/u/s/m/c/e/t bin cards plus vals (value, rel_error) pairs — with a synthetic fixtures/mcnp/mctal/synthetic_tally_bodies.mctal oracle (closed-form val/err pairing). Mesh tallies, radiograph/point-detector specials, tfc blocks, total/cumulative variants, and perturbation bodies stay named-open errors, never silent skips.
• Python API: Mctal.tallies / tally_nums / npert views plus Mctal.tally_vals_array NumPy bridge ((n_pairs, 2), float64 C-order) over the new bodies.
• MCPL particle-interchange reader/writer (new nucleide-mcpl-io crate): format versions 2 (read-only octahedral directions) and 3 (read/write, adaptive-projection directions), single/double precision, polarisation, user flags, universal PDG/weight, header comments/blobs with upstream stat:sum syntax validation, gzip-transparent paths, and byte-exact synthetic round-trips. Record-level interop is cross-checked both directions against the upstream 2.2.8 implementation. SSW conversion is deferred.
• Python API: nucleide.mcpl.read_mcpl / write_mcpl / McplFile thin wrappers over the new crate, plus a docs/tutorials/python/ mcpl-interchange.md tutorial.
• validation/mcpl_vs_refs.py: synthetic round-trip gates plus an upstream mcpl-package record-level cross-check (count, energies, PDG codes) with a loud SKIP when absent.
• R2S per-voxel photon-source tags (nucleide-r2s, no MOAB/HDF5): VoxelTags mapping zone totals onto native-mesh voxels (tag_zone_totals copies, split_zone_totals conserves), plus .photonSrc group selection (photon_groups_at) and element-wise sums (sum_group_strengths) feeding real group spectra where the uniform-split placeholder stood.
• Python API: nucleide.r2s.tag_zone_strength / photon_group_sums thin wrappers over the new tags.
• SDEF runtime line tables (nucleide-spectroscopy, E9 input path): the documented decay-lines TSV interchange (parse_lines_tsv / read_decay_lines over energy_MeV intensity rows) feeding the existing normalizer — no vendored lines, no legacy reader, synthetic fixtures only.
• Python API: nucleide.spectroscopy.parse_lines_tsv / read_decay_lines thin wrappers over the new reader.
• validation/spectroscopy_vs_pyne.py: runtime TSV interchange oracle — Cs-137 ENSDF lines through pyne.data round-trip the reader (energy column within 1e-9, E9 normalization to 1.0), loud SKIP when PyNE is absent or the energy/intensity pairing stays open.
• Interactive MCTAL demo: parseMctal WASM function over the new tally bodies (header plus per-tally pair counts and totals) with an mctal mode in the MCNP parser demo (inline sample plus the staged synthetic_tally_bodies.mctal fixture) and E2E coverage.

Changed

• Release workflow (release.yml): the crates.io publish list now covers every publishable workspace member (nucleide-kinetics and nucleide-spectroscopy were never listed, which failed the 0.6.0 release at nucleide-bindings with no matching package named nucleide-kinetics), plus the new nucleide-mcpl-io, and a fail-fast guard rejects future omissions instead of retrying lag-waits against a deterministic failure.

Fixed

• Hostile-input hardening in the MCTAL reader (nucleide-mcnp-io): allocations now follow the bytes actually present instead of reserving from declared bin/cycle counts (same class as the 0.5.0 FILL-matrix cap), the eight-card bin product saturates instead of wrapping, and fractional/negative bin counts are malformed-input errors rather than silent truncations. Valid files parse identically.

[0.6.0] - 2026-09-12

Added

• Plotly charts in the interactive Serpent and activation tutorials: the Serpent _res.m demo plots the IMP_KEFF mean ± uncertainty across burnup blocks (new keff_history field on parseSerpentRes), the _det.m demo plots per-detector tally spectra against energy-bin midpoints (new spectra field on parseSerpentDet, with Serpent 1 vs 2 column layouts), and the ALARA/FISPACT output tabs plot the reported total specific activity against cooling time. The activation demo stages the ALARA output/sample2.out and FISPACT inventory.fis fixtures behind new Load sample buttons. Website E2E smoke tests assert each chart renders.
• Serpent and FLUKA output-parsing Python tutorials (docs/tutorials/python/parse-serpent-output.md, docs/tutorials/python/parse-fluka-output.md): read _res.m/_dep.m/ _det.m files via nucleide.serpent.read_serpent and USRBIN .lis tallies via nucleide.fluka.read_usrbin, mirroring the MCNP parsing tutorial. Both are listed in the Python tutorial index and the docs index, and cross-linked from the tutorials “Finding more examples” section.
• SVG figures on all six theory pages (docs/theory/figures/): burnup-matrix column anatomy (depletion), MARC cascade schematic (enrichment), Walker alias-table construction (variance reduction), nucid digit layout (nuclear data), step-reactivity prompt-jump transient (kinetics), and peak/background anatomy (spectroscopy). Figures are staged into the website by sync-data.mjs and embedded with the kit’s theme-aware <SvgFigure> shortcode.
• Theme-adaptive inline-SVG capability map on the documentation home page (docs/README.md): currentColor follows the site theme and var(--primary) follows the accent picker. The home intro was also rewritten as a reader-facing welcome with a “Where to start” section.
• Decay-line SDEF source emission (nucleide-spectroscopy, equation E9): nucleide.spectroscopy.sdef_decay_source takes caller-supplied decay lines (energy, intensity), merges duplicates, sorts, and normalizes them to probability bins rendered as an MCNP point-source card — PyNE field order (POS= always; VEC=/DIR=1 only when directed; WGT=; PAR= via the nuclei particle dialects, MCNP versions 5/6), inline ERG=<E> for a single line, and discrete-distribution ERG=D1 + SI1 L/SP1 D cards wrapped at 80 columns for several. Distribution-card syntax is parser-verified surface only (MCNP sampling semantics stay caller-side); no decay data is vendored. Validation gains E9 normalization/card gates plus container-gated byte-exact diffs against PyNE PointSource.mcnp.
• Interactive legacy-reader tutorials: Serpent output (_res/_dep/_det) and FLUKA USRBIN browser demos, and ORIGEN TAPE5/6/9 tabs in the activation demo — serpent-io, fluka-io, and origen-io are now WASM-exposed with live demos and E2E coverage.
• “Common workflows” section in the Python API reference (five task-oriented snippets in the submodule idiom), an “Optional runtime dependencies” note in getting started (when numpy/h5py are needed), and “See also” links from the depletion, enrichment, kinetics, and spectroscopy tutorials to the committed cross-code validation results.

Changed

• Binding error messages for unsupported CRAM orders and unknown dose pathways/sources now name the accepted values (Python and WASM wordings aligned).

Fixed

• Theory-page SVG figures no longer render as white boxes in the site’s dark theme: the six figures in docs/theory/figures/ are theme-aware (no opaque background, near-black strokes/text follow currentColor, muted grays follow var(--muted-foreground), panel fills follow var(--muted)), and the pages inline them with the kit’s <SvgFigure> shortcode instead of <ImageFigure>. Labels that sit on hardcoded pastel fills keep hardcoded dark text (the nucid Z/A digits no longer turn near-white in dark theme), and the peak-anatomy sideband/background labels gained pastel pill backdrops for contrast in both themes.
• nucleide.serpent.read_serpent no longer wraps matrix variables in a spurious one-element outer list: matrices now map directly to 2-D lists of row lists (r["ABS_KEFF"][cycle] instead of r["ABS_KEFF"][0][cycle], same for _dep.m/_det.m matrices), matching the scalar/vector mapping. A matrix holding non-numeric values now raises ValueError instead of silently reading as an empty list. The Serpent tutorial and the parser cross-validation script were updated to the faithful shapes.
• Interactive demos: the Serpent parser demo no longer crashes on a one-column IMP_KEFF row, and the FLUKA/MCNP mesh views no longer overflow the browser argument limit on meshes above ~65k bins (min/max now computed with loops instead of spread).

[0.5.0] - 2026-09-11

Added

• Legacy I/O completion (nucleide-mcnp-io + nucleide.mcnp): ENDL EEDL/EPDL table reader (endl module: END_OF_TABLE framing, header slices, NFIELDS_RPROP body widths, get_rx selector lookup, and an endftod matching the PyNE C++ semantics) exposed as nucleide.mcnp.read_endl / EndlLibrary.get_rx / endl_endftod, with a synthetic fixtures/endl/ oracle; shared Fortran unformatted-record framework (fortran module: typed BadRecordMarker instead of the upstream AttributeError path, little-endian only) now backing the SSW reader/writer; SSW combining (combine_files port of scripts/ssw_combine.py: signed-orignp1 sum, plain-nrss sum, sign-preserving nps shift, typed incompatibility errors) exposed as nucleide.mcnp.combine_ssw_files (+ ssw_combine_main CLI); Python-side PTRAC event-row stream (ptrac_event_rows over read_ptrac, pinned against the 19-column PtracEvent schema) with an h5py table writer (write_ptrac_hdf5, ptrac_to_hdf5_main CLI; HDF5 bytes never asserted) and a MOAB-free meshtal mesh-data extractor (meshtal_mesh_data, MOAB tagging stays caller-side). Validation gains an ENDL-vs-pyne.endl oracle section (loud skips preserved). Known parity quirks stay documented, not fixed: SSW SF_00001 split and negative-np1⇒table-2, ncrd sign⇒abs width, PTRAC echo-repack to 8-byte mode, meshtal single-group total mirror and Rel_ column map, WWINP {0:13.5E} fields. Big-endian real-bytes input stays unsupported (no fixture oracle).
• Prescribed-reactivity point kinetics (new nucleide-kinetics crate): PKE solver for 1+ delayed-neutron groups under constant/step/impulse/ ramp/polyline reactivity schedules in Δk, equilibrium initials, the inhour relation with stable-period solve, and the prompt-jump approximation. The integrator is an adaptive implicit θ-method (trapezoidal default, backward Euler on request) with exact stepping to schedule knots — no explicit-solver parity. All delayed data are caller-supplied (from_ifp documents the OpenMC provenance note); thermal feedback and flux coupling are out of scope (the WASM kineticsTransient tutorial facade is covered in its own entry below).
• Python API: nucleide.kinetics.solve / equilibrium / initial_rate / inhour_rho / stable_period / prompt_jump thin wrappers over the new core, plus tests/test_kinetics.py and the validation/kinetics_vs_pyrk.py two-tier oracle (synthetic-fixture analytic gates O1–O4 plus a container PyRK ramp cross-check at 1e-3).
• Gamma-ray spectroscopy and measurement (new nucleide-spectroscopy crate): E1 rectangular / E2 five-point smoothing (edges copied), E3 m == 1 background / E4 half-open gross / E5 net counts with positional indexing, E6 quadratic energy bins, E7 log-polynomial efficiency evaluation in MeV (caller coefficients only, no fitting), E8 X-ray line algebra over caller-supplied atomic constants (no vendored tables), and the dollar/plain .spe readers with the upstream quirks pinned (duplicate-tag first-wins, missing-tag errors, live-real $MEAS_TIM:, last-plus-one $DATA:, keV-suffixed triplets, ignored $ROI:/ $PRESETS:/$ENER_FIT:, positional 0-based dollar labels). FWHM fits are parsed, never evaluated; peak search/fit, activities, decay spectra, and plotting stay out of scope.
• Python API: nucleide.spectroscopy.rect_smooth / five_point_smooth / calc_bg / gross_count / net_counts / energy_bins / detector_efficiency / xray_lines / parse_dollar_spe / parse_spe / read_dollar_spe / read_spe thin wrappers over the new core, plus tests/test_spectroscopy.py and the validation/spectroscopy_vs_pyne.py two-tier oracle (synthetic E1–E8 gates plus a container PyNE cross-check at 1e-12).
• Browser-interactive tutorials for point kinetics and spectroscopy: kineticsTransient (step-reactivity PKE solve returning the n(t) series plus the E4 prompt-jump value) and spectroscopySmooth (E1 rectangular / E2 five-point smoothing plus E3–E5 gross/background/net counting) thin WASM facades over the verified crate APIs, with KineticsTransient / SpectroscopyDemo demos and tutorials/interactive/kinetics.mdx / spectroscopy.mdx pages. Demo presets stay synthetic (fixtures/kinetics/ + fixtures/spectroscopy/ values, hand-picked small numbers otherwise).
• NumPy bridges for MCTAL/WWINP/PTRAC (bindings-only, same result_array pattern as the meshtal bridge: flatten → Vec::into_pyarray → reshape view, float64 C-order, ValueError on ragged/out-of-range): Wwinp.ww_row_array / ww_column_array / ww_particle_array ((nft,), (n_groups,), (n_groups, nft) with nft = nf[0] * nf[1] * nf[2]), Mctal.k_arrays (five (n_cycles,) series) / averages_array ((n_cycles, 14), (0, 14) when empty), PtracFile.events_array ((n_events, 19) in ptrac_event_columns order, absent variables as 0.0) / event_field_array ((n_events,)). All arrays are owned, writable, and decoupled from the file data; the wheel stays NumPy-free (NumPy required only at call time, pytest.importorskip in tests/test_mcnp_io.py). Plain-copy helpers (ww_row, ww_column, the k_col/averages getters, events) are unchanged.

Fixed

• Website build: the phillips-1978 entry in docs/theory/spectroscopy.mdx carried no URL, which the docs content schema requires (references.0.url), so npm run build failed. The entry is now plain prose in the References section (citation text only, no journal-verification claim, no invented link).
• Stable-as-zero analytics (nucleide-material): a known nuclide with mass data but no decay constant is now stable (λ = 0) and contributes exactly 0.0 to activity / decay_heat / dose_per_g instead of raising MissingDecay. Stable nuclides skip the dose-factor lookup entirely, so fully stable compositions such as H2O return 0 on every pathway/source. Nuclides with no mass data still raise MissingMass, and radioactive nuclides without a dose row (including -1 air sentinels) still raise MissingDose. Python/WASM signatures are unchanged.

[0.4.0] - 2026-09-10

Added

• Python API: MeshTally.result_array() / totals_array() zero-copy NumPy bridge over the meshtal result/rel_error tables (supersedes the 0.3.0 deferral): result_array() returns owned writable C-order float64 (ve, group) arrays with ve = (i * ny + j) * nz + k, moved out of one row-major flatten via Vec::into_pyarray + a reshape view (no second copy); totals_array() returns (num_ves,) totals pairs directly. Requires NumPy installed at runtime (numpy>=1.26 in the test extra; base wheel stays dependency-free). to_list() / totals_list() stay as the NumPy-free plain-copy path.

• License-free decay-data pack (nucleide-nuclei, AD-7): per-branch daughters from ENDF/B-VIII.0 decay tapes (crates/nuclei/src/data/decay_branches.tsv, 5 068 rows over 3 541 parents; MF8/MT457 NDK RTYP/RFS/BR with SF/fission branches dropped and zero-half-life tapes absent), exposed as decay_branches / branching_fraction lookups plus the DecayData facade and the DecayBranch / DecayBranchMode types. Isomer masses extend ame2020.tsv (738 rows: m_ground + ELIS/931.49410242 u from each isomer tape’s File-1 MT451 record; tapeless isomers fall back to the ground-state mass, Q-values stay ground-state-only). New free-form normalize_nuclide_name in crates/nuclei/src/dialects.rs (symbol-first, then mass-first so N15 stays nitrogen and 92235 stays a ZAID; bare symbols rejected; Ir-192n resolves to state 2). scripts/gen-nuclear-data.py --endf-decay8 regenerates both tables with hard ENDF spot-checks (K-40 beta-/EC pair summing to 1, Es-254 members); decay_energy.tsv stays on ENDF/B-VII.1 with its basis in its header.

• Python API: nucleide.nuclei.decay_branches / nucleide.nuclei.decay_branch_fraction thin wrappers over the new tables. WASM decay_branches / decay_branch_fraction table-level mirrors.

• validation/nuclear_data_vs_refs.py: DECAY_BRANCH_SPOTS oracle (K-40 pins 3.93839e16 s plus its beta-/EC pair; Es-254 members per the repo tables) following the decay-energy spot pattern; committed results untouched, container rerun deferred.

• nucleide-depletion analytic Bateman decay fast path (crates/depletion/src/bateman.rs): cached C/C⁻¹ eigendecomposition closed form (Bateman 1910; Amaku–Pascholati–Vanin CPC 181 (2010)) for decay-only lower-triangular chains, generated at runtime from the crate’s own chain data, plus an f64-careful BatemanHp variant (magnitude-sorted terms, Neumaier compensation, exp_m1 for small λt; no new dependencies). New Method::{Cram(Order), Bateman, BatemanHp} selector ("cram16"/"cram48"/"bateman"/"bateman_hp") threading deplete_with_method / integrate_with_method / DecayInventory::decay_with_method (existing Order-based entry points stay as CRAM shims; dt == 0.0 returns the input exactly). Non-decay systems fall back to CRAM-48 inside the dispatcher: near-degenerate half-lives (1e-12 relative gap with a live coupling path), cyclic or out-of-order topology, and reactions/fission on; stable nuclides use limit forms inline.

• Python API: method= on nucleide.depletion.deplete / deplete_series / DepletionSystem.solve / solve_vec / Inventory.decay (default "cram48"; an explicitly non-default method overrides order). WASM deplete / depleteSeries gain an optional trailing method.

• validation/depletion_vs_openmc.py: in-memory Bateman-vs-CRAM-48 cross-check column (single-step vs CRAM-48 within 1e-6, vs analytic within 1e-8, series band within 1e-6); committed results untouched, container rerun deferred.

• nucleide-mcnp-io L3 semantic objects (crates/mcnp-io/src/semantic.rs): typed MODE (37 particle shorthands, default {N}), TRn cards (degrees flag, 3-entry displacement, 5–9 entry rotation, main-to-aux flag, hidden inline FILL transforms), auto-created universes (U/-U, data-block lists with J/nJ/nR/nM expansion), LAT (1|2 only), cell FILL (single universe or 3-D matrix with transform reference or hidden transform; data-block lists are simple per-cell only), per-cell IMP/VOL, periodic surface pointers, and a minimal typed tally model (Fn number + particle classifier + entries with grouped FMn/En). DeckProblem::validate() centralizes duplicate-number conflicts, dangling material/surface/complement/transform/periodic/fill links, redundant cell+data definitions, write-time state checks, and nucleide-defined lattice/fill cross-checks (LAT-without-FILL and FILL-matrix-without-LAT are errors); particle/mode mismatches are notes via validation_notes(). Synthetic fixture fixtures/mcnp/inp/deck_l3.txt (no license needed).

• Python API: DeckProblem mode/transforms/universes/lattices/ fills/importances/volumes/tallies getters, validate() / validation_notes(), and set_mode/set_cell_universe/ set_cell_lattice/set_cell_fill setters.

• nucleide-emit single-material emission (crates/emit/): one Material renders through five code dialects — MCNP m cards (mass fractions, configurable xs suffix, 128-column packing), Serpent mat cards ({zaid}.{lib} ids), FLUKA COMPOUND cards, ALARA mixture blocks, single-zone PARTISN decks — plus a mass-drift report (emit_drift) with per-nuclide drop reasons and re-parse verification where readers exist (MCNP, ALARA).

• Python API: nucleide.emit.emit_cards / emit_drift_table thin wrappers over the new crate.

• nucleide-emit ARMI blueprint bridge (crates/emit/src/armi.rs): from_armi_mass_fracs builds a Material from ARMI-side post-expansion mass-fraction keys (database names, bare names, ZAIDs-as-strings, AAAZZZS ids, unambiguous MC2-3 labels via the nuclei bridge) for unchanged MCNP/Serpent/FLUKA/ALARA/PARTISN emission. One-way, dict-in only: elemental keys (caller passes expanded massFrac), number fractions, enrichment shorthands, balance, and temperatures stay caller-resolved; bare AM242 is rejected (pass AM242M/AM242G explicitly).

• Python API: nucleide.emit.emit_armi_cards / emit_armi_drift_table thin wrappers over the bridge.

• nucleide-nuclei dose factors (crates/nuclei/src/data/dose_factors.tsv, 1,116 rows: 93 folded nuclides × 4 pathways × 3 sources from the BSD-3 PyNE dbgen/dosefactors*.csv tables, HNF-SD-WM-TI-707 Rev.1 / HNF-5636 App. O; +D folds into the parent, GENII/DOE air are -1 sentinels): DosePathway/DoseSource/DoseEntry lookups plus a DoseData facade, generated by the committed stdlib-only scripts/gen-nuclear-data.py (--dose-*.csv flags over local copies).

• nucleide-material dose per gram: DoseProvider trait (DoseFactors, NoDoses, plus impl for nucleide_nuclei::data::DoseData), Material::dose_per_g / total_dose_per_g following PyNE’s Material::dose_per_g equations (per-gram map semantics; screening-level only, not for safety decisions).

• Python API: nucleide.nuclei.dose_factor and nucleide.material.dose_per_g thin wrappers over the vendored tables + material analytics.

• WASM WasmDeckProblem (bindings/wasm/src/lib.rs) + deck-editor demo (website/src/components/interactive/DeckEditor.tsx, docs/tutorials/interactive/deck-editor.mdx): fromText deck parsing (text only, no filesystem in the browser), typed cells/surfs/mode/ transforms/universes/lattices/fills/importances/volumes/ tallies views, materialNumbers/dataNames, cellInventory (cell→material map), validate/validationNotes, and setCellDensity/ setCellMaterial/setMode/setCellUniverse/setCellLattice/ setCellFill setters over the existing Tier 1 DeckProblem API. WasmInventory over the existing DecayInventory API (units default "atoms", decay honoring rates + solver method via the predictor single-step, activities/masses/moles, fractions, halfLivesReadable, add/sub/mul/div, toCsv/fromCsv) plus cumulativeDecays/ progeny/branchingFraction/decayMode/chainEdges free functions. The demo parses the inline synthetic deck_minimal.txt, fetches the synthetic deck_l3.txt sample staged by sync-data.mjs, edits cells through the setters, and shows validation plus byte-identical dumps.

• nucleide-r2s ARMI DB-snapshot adapter (crates/r2s/src/snapshot.rs): deck_from_snapshot / R2sWorkflow::from_snapshot / snapshot_workflow build a validated volumes-method ALARA template deck from versionless caller-dumped dicts (opaque zone ids 1:1 from block names, 1:1 zone→mixture with atoms/barn-cm number densities as element vol_fraction, caller-supplied fluxes/cooling/schedule; empty compositions map to void and are skipped). Dict-in only: no HDF5 dependency, no ARMI layout versioning mirrored, synthetic test data only. Composition keys follow the emit ARMI-input rule (post-expansion nuclide keys; elemental keys, bare AM242, and unknown names are errors).

• Python API: nucleide.r2s.r2s_from_snapshot thin wrapper returning {workflow, deck, decks} (workflow summary, canonical template deck, one deck per step).

• Interactive mcnp-io demo (website/src/components/interactive/McnpParser.tsx, docs/tutorials/interactive/mcnp-io.mdx): the xsdir histogram now pairs the existing per-element table-count bar with a table-count bar by library suffix and an AWR-distribution histogram over the staged xsdir_sample.txt.

• Interactive variance-reduction demo (website/src/components/interactive/VrDemo.tsx, docs/tutorials/interactive/variance-reduction.mdx): the MAGIC energy upper-bound step line (hv shape, log energy axis) gains a group-count title.

• Interactive tutorial E2E coverage (website/e2e/smoke.spec.ts): follow-up clicks with per-click output assertions on every demo page — deck-editor Validate + a setter flow + Load L3 sample, variance-reduction Sample index / Sample voxel, materials Mix + To XML, enrichment Optimize M*, activation ALARA-output / FISPACT / R2S mode tabs, mcnp-io wwinp parse, depletion Load sample chain.

• WASM mirrors for the new Rust APIs (bindings/wasm/src/lib.rs, typed in website/src/types/nucleide-wasm.d.ts): emitCards / emitDriftTable over nucleide-emit (all five dialects, EmitOpts overrides, drift rows with per-nuclide drop reasons), emitArmiCards / emitArmiDriftTable over the ARMI bridge (elemental keys, bare AM242, and unknown names throw), doseFactor / dosePerGram over the vendored HNF-5636 tables (pathway air/soil/ingest/inhale, source EPA/DOE/GENII, screening-level only), and r2sFromSnapshot over the snapshot adapter (typed Snapshot*Json inputs returning a {workflow, deck, decks} bundle).

• Interactive emitter demo (website/src/components/interactive/ EmitterDemo.tsx, docs/tutorials/interactive/emitter.mdx): GNDS/ARMI key toggle with inline presets, material name/density/per-dialect options, all five dialect cards plus the drift table, with inline errors for elemental/AM242/bad keys.

• Interactive materials demo dose-per-gram section (website/src/ components/interactive/MaterialBuilder.tsx): pathway/source selects scoring the current formula via dosePerGram, with a screening-only note.

• Interactive activation demo r2s-snapshot mode (website/src/ components/interactive/ActivationDemo.tsx): an inline two-zone JSON snapshot parsed via r2sFromSnapshot into workflow steps plus the template and per-step deck count.

• Interactive tutorial E2E coverage (website/e2e/smoke.spec.ts): emitter GNDS emit plus ARMI re-emit with drift assertion, materials Compute dose value, activation R2S snapshot tab with steps and deck-count outputs.

Fixed

• Interactive tutorial input guards (inline WASM error: before any WASM call instead of NaN passthrough): deck-editor density / material / universe / lattice / fill / mode setters reject non-finite input (website/src/components/interactive/DeckEditor.tsx, docs/tutorials/interactive/deck-editor.mdx); variance-reduction tally / tolerance / null value plus alias-table PDF tokens reject non-finite input instead of silently dropping bad tokens (website/src/components/interactive/VrDemo.tsx, docs/tutorials/interactive/variance-reduction.mdx); depletion atom counts / time step / burnup steps require finite (and positive where applicable) values (website/src/components/interactive/DepletionStep.tsx, docs/tutorials/interactive/depletion.mdx).
• Invalid nuclide-id fallback labels instead of panics (crates/nuclei/src/ lib.rs, crates/nuclei/src/armi.rs): NuclideId::from_nucid on a raw integer outside the validated (Z, A, state) domain previously panicked downstream — out-of-bounds ELEMENTS[z] indexing in to_name, or an .expect() on the element-symbol lookup in nucid_to_armi_label. Both are now total: such ids render the diagnostic fallback Z{z}A{a}[m{s}] (with the same n-prefix/capitalization treatment on the ARMI side), which from_name does not parse. New try_from_nucid / is_valid validate untrusted integers (1 <= Z <= 118, Z <= A <= 999, canonical tail) with the matching Error instead of producing a fallback id.
• Poisoned-lock getters now return errors instead of panicking (bindings/python/src/lib.rs): every PyCascade getter and PyDeckProblem getter/setter previously called .lock().unwrap(), so a poisoned Mutex aborted the interpreter. Each now maps the poison to a ValueError ("cascade lock poisoned" / "deck lock poisoned").
• Cell-FILL matrix cap at 1 000 000 cells (crates/mcnp-io/src/semantic.rs): a hostile FILL range such as -2147483648:2147483647 previously overflowed the i32 width computation (debug panic / release wrap) and could drive a gigabyte allocation. Widths are now computed in i64 with checked multiplication, and any matrix above the cap fails with a cap error before allocation; larger lattices must be built programmatically, not spelled per cell.
• Non-finite n0 validation at the Bateman solve entry (crates/depletion/src/bateman.rs): BatemanCache::solve previously propagated NaN/infinite/negative initial counts into the triangular solves. It now rejects them with a Linalg error (n0 must hold finite atom counts >= 0), matching the existing dt/length-mismatch errors.
• Snapshot-mixture emission returns an error instead of panicking (crates/r2s/src/snapshot.rs, crates/r2s/src/error.rs): the deck_from_snapshot mixture writer previously ended its entry match with unreachable!, so a future non-element variant would panic. It now returns Error::Invalid (the error.rs:16 arm) with the offending entry attached.

[0.3.0] - 2026-09-08

Added

• nucleide-depletion time-series integrators (crates/depletion/src/integrate.rs): Integrator::{Predictor, Cecm, Cf4} with integrate(sys, n0, steps, integrator, order) returning per-node atoms, activity (λ·N), and decay heat (λ·N·E, chain decay_energy → ENDF/B-VII.1 table → zero). Synthetic A→B→C fixture fixtures/depletion/chain_abc.xml plus tests/test_depletion_timeseries.py covering multi-step Bateman accuracy.
• nucleide-nuclei data breadth (no HDF5, no hand values): simple_xs.tsv (241 rows; thermal totals from NIST NCNR bound XS as free-atom xs·(A/(A+1))² + xs_a, fast totals from ENDF/B-VII.1 MF3/MT1 at 14 MeV), scattering_lengths.tsv (267 rows from NIST NCNR Sears-1992, complex lengths by real part, monoisotopic element attributions), decay_energy.tsv (3,557 rows of mean prompt recoverable MeV/decay from ENDF/B-VII.1 MF8/MT457; daughter gammas live on daughter rows, so Cs137 reports prompt 0.179 MeV and Ba137_m1 0.661 MeV — chain codes sum members). All three generated by the committed stdlib-only scripts/gen-nuclear-data.py (run --help for upstream URLs); resonance absorbers without NIST rows and isomers stay absent by construction.
• nucleide-material decay heat: DecayEnergyProvider trait (DecayEnergies, NoDecayEnergies, plus impl for nucleide_nuclei::data::DecayData so depletion gets λ and MeV from one provider with no circular deps), Material::decay_heat / total_decay_heat in watts, and MEV_TO_JOULES. Dose coefficients stay a documented deferral (PyNE’s dbgen/dosefactors*.csv are the future source).
• nucleide-depletion: single decay-energy source of truth — the hand-kept fallback table is removed and decay_energy_mev resolves chain → ENDF/B-VII.1 table → 0.0; new shared decay_energies_by_name helper.
• Python API: nucleide.depletion.deplete_series — multi-step depletion driver over the core integrate series (predictor/cecm/cf4; per-step rates/rates_list), returning times/atoms/activity/ per-nuclide decay_heat maps through the shared resolver (identical to core integrate heats; the core t = 0 row is omitted).
• Python API: nucleide.nuclei (simple_xs, scattering_length, decay_energy) and nucleide.material (decay_heat) thin wrappers over the vendored tables + material analytics.
• Python API: MeshTally.to_list() / totals_list() plain-copy helpers; the zero-copy NumPy bridge (result_array()) stays deferred — no numpy dependency introduced.
• nucleide-mcnp-io full-deck round-trip (crates/mcnp-io/src/cell.rs, surf.rs, problem.rs): typed cell/surface cards (CSG expressions with : union, juxtaposition intersection, parentheses, # complement, * reflecting markers; 40 surface kinds incl. macrobodies), DeckProblem parse/edit/write with format-preserving write-back (byte-identical when unedited; only touched cards re-render canonically) plus set_cell_density / set_cell_material. Synthetic fixtures fixtures/mcnp/inp/deck_{minimal,macro,params}.txt (no license needed).
• nucleide-depletion unit-aware inventories (crates/depletion/src/ inventory.rs): DecayInventory over atom counts with activity/mass/mole units (Bq…Ci, g/kg, mol, time in s/m/h/d/y), fractions, readable half-lives, arithmetic, CSV round-trip, plus cumulative_decays and chain observables (progeny, branching_fraction, decay_mode, chain_edges).
• nucleide-nuclei ARMI dialect bridge (crates/nuclei/src/armi.rs): armi_name_to_nucid (nU235, ZAIDs, MCC3, AAAZZZS), nucid_to_armi_label, mcc3_to_nucid.
• nucleide-material composition checks (crates/material/src/check.rs): check_labels (truncated-label collisions at DIF3D/MC2 6/8-char widths across GNDS/ZAID/Serpent/ALARA/ARMI labels) and conservation audit (non-positive totals, negative masses, duplicates, unknown masses).
• Python API: nucleide.mcnp.DeckProblem (parse_deck/read_deck, dumps, setters, cell/surf/data accessors), nucleide.depletion. Inventory + cumulative_decays/progeny/branching_fraction/ decay_mode/chain_edges, nucleide.nuclei.armi_to_nucid/ nucid_to_armi/mcc3_to_nucid, nucleide.material.check_labels/ audit_material.

[0.2.0] - 2026-09-06

Added

• cccc-io crate: CCCC text-subset parsers (ISOTXS multigroup libraries, RTFLUX/ATFLUX/RZFLUX flux files) + PARTISN deck writer with ISOTXS nuclide mapping and validation (no solver). Synthetic fixtures under fixtures/cccc/ (no license needed).
• fispact-io crate: FISPACT-II inventory-output parser producing ALARA-compatible response frames (output-only, reusing ResponseFrame; no solver). Synthetic fixture fixtures/fispact/inventory.fis (no license needed).
• origen-io crate: scoped ORIGEN 2.2 TAPE readers — TAPE5 input echo, TAPE6 output inventory, TAPE9-style decay constants. Synthetic fixtures under fixtures/origen/ (no license needed).
• r2s crate: scoped R2S workflow builder — zone-to-flux linking from ALARA decks, schedule expansion, per-zone deck emission, and photon-source assembly with a documented uniform-split approximation (total strength preserved; group-wise emission lines stay in ALARA .photonSrc spectra). Transport and activation solving stay out of scope.
• Python API: nucleide.cccc (isotxs_parse, rtflux_parse, partisn_render, partisn_validate), nucleide.fispact (fispact_parse_output), nucleide.origen (origen_parse_tape5/6/9), and nucleide.r2s (r2s_from_deck, r2s_validate, r2s_expand, r2s_assemble) exposing the four new crates as plain dicts/lists.
• alara-io crate: ALARA input-deck, group-flux, material/element/WDR library, activation-output, photon-source, and schedule-expansion parsers (glue only; solver out of scope). Fixtures vendored verbatim from UW ALARA BSD samples under fixtures/alara/ (terms in fixtures/alara/LICENSE.ALARA).
• Python API: nucleide.alara (alara_parse_deck, alara_parse_flux, alara_parse_output, alara_expand_schedule) exposing the alara-io deck/flux/output/schedule glue as plain dicts/lists.

Changed

• Release workflow: removed the non-functional cargo publish --dry-run step for workspace crates and added set -euo pipefail to the publish loop so a failure in one crate stops the job.

Fixed

• Workspace crate dependencies now specify a version requirement so cargo publish accepts them for crates.io.
• scripts/bump-version.sh now updates workspace crate dependency versions alongside [workspace.package] version.

[0.1.0] - 2026-08-30

Added

• depletion: reaction branching_ratio support in chain XML (default 1.0), with branched gains in the burnup matrix.
• depletion: light-particle production in the burnup matrix — He-4/H-1 from alpha/proton decays and from reaction secondaries such as (n,α) and (n,p), when the product nuclide is in the chain.
• depletion: decay branching ratios are renormalized to sum to 1 (largest branch adjusted) for programmatically built chains, matching OpenMC’s chain-generation behavior.
• depletion: CASL-style fission-yield borrowing in chain XML (<neutron_fission_yields parent="X"/> resolves X’s yields, transitively), matching OpenMC.
• nuclei: NuclideId::from_name now accepts PyNE-normalized forms such as "U-235", "u235", and uppercase-M metastable markers.
• Python API: Cascade.solve_multicomponent() (M*-optimizing solve), DepletionSystem builder and solve_vec() for repeated solves without per-call overhead, and MeshTally.total_rel_error.
• Python API: nucleide.data module (fetch, fetch_compendium, default_ref) for downloading repo data files — the Materials Compendium, sample depletion chains — pinned to the installed version’s tag, since the wheel bundles no data files.
• Criterion benchmarks in crates/*/benches/ (cargo bench).
• validation/: runnable cross-code validation harness against PyNE 0.7.5 and OpenMC 0.16.0 (containerized via validation/Containerfile + validation/run_container.sh) with committed results (validation/results.md).
• validation/: full-chain depletion validation on the CASL/VERA simplified chain (downloaded to the git-ignored validation/.cache/), parser cross-validation against PyNE and serpentTools oracles (parsers_vs_refs.py), and generated paper figures (validation/figures/, via make_figures.py).
• JOSS submission materials: paper.md, paper.bib, CITATION.cff, CONTRIBUTING.md, and a draft-PDF workflow.
• bindings/wasm: wasm-bindgen crate exposing a subset of the workspace to the browser, including a WasmMaterialsCompendium API that parses the DOE/PNNL Materials Compendium from its JSON text.
• Documentation website (website/, Astro + @nukehub/docs-kit) with content synced from docs/, theory pages, and interactive WASM tutorials (including a compendium browser that fetches the staged MaterialsCompendium.json and charts compositions with Plotly), deployed to GitHub Pages.
• Interactive tutorials: MCNP meshtal z-slice heatmap with log colorscale and relative-error toggle, xsdir table-count histogram by element, depletion burnup-curve plot, material atom-vs-weight-fraction bar chart, enrichment cascade stage-profile line chart, and MAGIC energy upper-bound step line.
• enrichment: Cascade::stage_profile() returns per-stage assays of the enriching key via the ideal-cascade recurrence, exposed through WASM as WasmCascade.stageProfile().
• Release workflow: vX.Y.Z tags build and publish Python wheels for Linux, macOS, and Windows to PyPI, publish workspace crates to crates.io, and draft a GitHub release from the matching changelog section.

Changed

• Documentation website moved from https://nukehub-dev.github.io/nucleide to a custom domain, https://nucleide.nukehub.org, hosted on GitHub Pages with Cloudflare DNS; the Astro base path changed from /nucleide to /.

• Documentation theory pages (depletion, enrichment, nuclear-data, variance-reduction) now use the citation support from @nukehub/docs-kit: references are declared in frontmatter, cited inline with <Citation />, and rendered as a linked bibliography with copy-to-clipboard export.

• Python API reorganized into domain submodules (nucleide.nuclei, nucleide.material, nucleide.mcnp, nucleide.serpent, nucleide.fluka, nucleide.vr, nucleide.enrichment, nucleide.depletion) mirroring the workspace crates; the top level re-exports the domain submodules alongside version() and __version__.

• vr-tools: MeshSourceSampler now rejects negative or non-finite tally / user-density values with an error instead of silently absolutizing them.

• depletion: reaction loss is subtracted once per reaction type per nuclide; duplicate entries only add branched gains.

• depletion: CRAM pole solves reuse scratch buffers across poles.

• nuclei: FLUKA name lookups use lazily-built maps instead of linear scans.

Fixed

• Documentation website “Edit this page on GitHub” links now point at real repo files (docs-kit sync injects the source path as editPath frontmatter); previously every link 404’d — all were missing the file extension, and the changelog link pointed at docs/changelog instead of the repo-root CHANGELOG.md.

• Documentation website deployment now builds the bindings/wasm package in CI, so the interactive tutorials can load /wasm/nucleide_wasm.js instead of receiving the SPA fallback HTML response.

• Interactive tutorials: file/text inputs now use the docs-kit Textarea auto-resize (content-fitted height, no manual resize grip); kit bumped for a Select dropdown fix so short lists flipped above the trigger no longer float with a gap.

• depletion: Chain::from_xml no longer renormalizes decay branching ratios — file values are used verbatim, matching OpenMC’s Chain.from_xml (renormalization only happens at chain generation). Found by the CASL full-chain validation (e.g. I-128 β⁻ branching ratio 0.931).

• depletion: fission production now uses the yield set at the lowest incident neutron energy, matching OpenMC’s get_default_fission_yields.

• nuclei: NuclideId::new rejects mass numbers above 999, fixing a debug overflow panic from inputs like "U999999".

• depletion: zero/negative/non-finite half-lives and non-positive dt now return errors instead of producing inf/NaN.

• linalg: LU solve validates matrix/RHS dimensions and symbolic-pattern identity instead of panicking inside faer.

• enrichment: validates alpha > 1, assay ordering, and M* bounds; guards secant divide-by-zero; golden-section M* polish no longer trusts unconverged probe solves; recompute_nm reset path no longer uses stale right-hand sides.

• vr-tools: magic no longer panics on tallies with empty energy bounds and validates array lengths and finiteness.

• bindings/wasm: fraction and result maps are now serialized as plain JS objects instead of JS Maps, fixing empty tables in all interactive tutorials.