ACTINV v1.0.1

ACTINV v1.0.1 — release notes

ACTINV 1.0.1 is a backward-compatible performance release. It changes how activation data is prepared and reused,
not the evaluated nuclear data, physics, solver order, public input/output schemas, certificate provenance, inventory
values, or ledger values. The separately versioned nuclear-data catalog remains data-v1.0.0.

Upgrade the Python package with:

python -m pip install --upgrade actinv

Rust CLI users can install this exact patch with:

cargo install --locked --force actinv-cli --version 1.0.1

What users gain

The first calculation prepares deterministic, integrity-checked cache files bound to the exact activation library,
index, group boundaries, spectrum, schema, and algorithm. Later compatible calculations reuse those files
automatically. No specification change is required.

On the frozen public FNS iron example and recorded Intel Core i3-N305 host, the warm path changed from 3.075 seconds
to 1.185 seconds median process wall time and from 1.077 GB to 129 MB peak RSS: 2.595× faster and 8.326× lower peak
memory. The p95 changed from 3.190 seconds to 1.212 seconds, or 2.632× faster. These figures apply to that workload,
host, warm-cache state, and requested outputs; they are not a general comparison with another product. The optional
one-second warm-median stretch goal was missed, while every required performance gate passed.

The initial cache creation measured 5.583 seconds at 130 MB peak RSS and wrote about 282 MB. Deleting the cache is
safe and changes only preparation time. Set ACTINV_CACHE_DIR to choose its location; otherwise ACTINV follows the
platform cache-directory convention.

Verification

• All 167,735 production rows, 710 group boundaries, 33,597,258 retained cross-section values, and 167,735 spectrum
  collapses matched the verified opening representation exactly.
• Twenty-three corruption, truncation, stale-source, schema, offset, count, and integrity plants failed before a
  result was published. Interrupted and concurrent creation produced only validated, byte-identical final files.
• CLI and Python returned the same normalized scientific result and retained the original source paths and SHA-256
  identities in their certificates. Python uses the same Rust cache rather than creating a second library copy.
• The exact Rust format, check, strict-Clippy, and test commands; Python-binding Rust gates; historical scientific
  controls; parser reliability; packaging; and clean-clone controls all passed locally and in GitHub Actions.

The complete machine-readable evidence is in results/p15_*.json, the independently derived verdict is
results/verdict_p15.json, and the closeout narrative is docs/history/sessions/P15.md.

Compatibility and qualification

Existing actinv-spec-1 problems and the data-v1.0.0 files remain compatible. Generated cache files are local
derived data and are not part of a result certificate or release payload. An existing corrupt or incompatible cache
fails closed with a diagnostic instead of being trusted or silently overwritten.

ACTINV remains research-grade software. This patch is not approval for licensing, safety, waste-classification, or
regulatory use; the boundaries in docs/QUALIFICATION.md and the carried limitations in the v1.0.0 notes still apply.

ACTINV v1.0.0

ACTINV v1.0.0 — release notes

ACTINV v1.0 is the first technically complete repository release for reproducible activation and nuclide-inventory
workflows. It combines a deterministic Rust solver and data pipeline with a Python interface, transport-flux imports,
mesh execution, decay-photon sources, radiological responses, and explicitly bounded uncertainty reporting.

This is research-grade software. It is not approved for licensing, safety, waste classification, or regulatory use.
Public tagging and uploads to PyPI, crates.io, and GitHub Releases are separate maintainer actions and are not implied
by the repository's technical release verdict.

Installation

Published Python wheels support Python 3.9 and newer through the stable ABI:

pip install actinv

That one installation provides both import actinv and the actinv command. Native standalone binaries are also
distributed as GitHub release artifacts, and the command can be built from source with:

cargo install --locked --path crates/actinv-cli

Nuclear data are not embedded in either software artifact. The recommended versioned neutron bundle can be installed
and verified with:

actinv data fetch
actinv data verify

See Nuclear data for charged-particle, covariance, offline, and manual-build options.

Added since v0.5

• Strict ENDF-6 MF=33 covariance sidecars tied to the activation library, energy structure, and source manifest.
• Selectable CRAM-16/48 integration, exact local response sensitivities, and first-order MF=33 bands for selected heat
  and activity responses, with covered/uncovered parameters and numerical-method bounds kept separate.
• User-supplied, hash-pinned clearance-index, waste-index, ingestion-dose, and inhalation-dose tables, including
  scenario metadata, coefficient coverage, and an optional complete-coverage requirement.
• Independent re-derivation of all 289 embedded natural-abundance and atomic-mass rows from the Meija et al. primary
  table and AME2020, including byte-identical source regeneration.
• A bounded deterministic input-reliability gate and minimized regression corpus for the supported public interfaces.
• FNG/ITER cell-620 activation-history reproduction from the published research archive, preserving every supplied
  source hash and keeping all generated bulk data outside Git.
• Stable-ABI Python wheels that install both the Python API and full command, standalone release binaries, consistent
  1.0.0 interface versions, and public-release documentation/checklists.
• A separately versioned, immutable data catalog with verified one-command setup for exact validated TENDL-2025
  artifacts and official decay archives.

Validation summary

• P11's complete TENDL-2025 neutron covariance scan covers 2,850 evaluations, 84,489 MF=33 sections, and 285,023
  retained components with no parse error or silent omission. The final sidecar is fresh/cached byte-identical.
• Analytic sensitivities agree with independent finite differences, and reported covariance propagation agrees with
  direct matrix calculations and an independent fixed-seed sampling control within the frozen limits.
• P12 radiological responses are identical through CLI, Python, prepared, and mesh paths; 80 independent response
  comparisons agree exactly, and incomplete activity coverage remains explicit.
• The independent primary-table control reproduces 289 abundance/mass rows, all 84 element sums, and the generated
  Rust table byte-for-byte.
• Repeated 10,000-case input checks produce the same deterministic summary, and the fixed 1,000,000-case partition
  covers all eleven public reader families with no process-level failure below its 1 GiB ceiling. Five pre-run
  allocation/order and encoding findings retain focused regression tests.
• The FNG/ITER control compares Co-58, Tc-99m, Mn-56, and Cr-51 at all 170 endpoints. Its worst relative difference on
  populations at or above one million atoms is 2.88e-14, within the frozen 1e-4 bound. Selected independently read
  reaction rates agree to 3.24e-16 relative.
• The clean-clone release control builds and verifies all three unpacked Rust archives, the standalone 1.0.0 binary,
  a Python 3.9 stable-ABI wheel, and the Python source distribution; package metadata, both licence texts, the wheel
  SBOM, installed console command, import behavior, strict Rust gates, prior verdicts and CLI/Python end-to-end
  behavior all pass.

The complete phase evidence and exact hashes are in results/, protocols/, and the append-only session records.

Carried limitations

• ACTINV performs inventory evolution, not neutron/photon transport, criticality, thermal hydraulics, or a GUI.
• Mesh cells are independent and do not feed composition, flux, or temperature changes back to one another.
• Supported incident particles are neutron, proton, deuteron, and alpha; triton, helion, and gamma activation are not
  implemented.
• Unresolved-resonance treatment is infinite dilution. Finite-dilution self-shielding, probability tables, and
  Bondarenko factors are not implemented.
• Products absent from configured decay files remain named leakage. No missing decay or fission-yield data are
  invented.
• Contact gamma dose is a semi-infinite-slab screening proxy. Photon and neutron transport remain external, and the
  ordinary photon exporters retain a point-at-origin spatial placeholder.
• MF=33 bands cover retained cross-section covariance only. Decay, yield, flux, composition, response-coefficient,
  geometry, and model uncertainties are excluded; missing evaluated cross-covariance remains explicit.
• Radiological tables are not bundled or selected by ACTINV. Users must choose the applicable jurisdiction, edition,
  scenario, intake assumptions, chemical/aerosol form, and margins, and must assess missing coefficient coverage.
• OpenMC, MCNP, and FISPACT imports cover documented, fail-closed subsets with explicit physical normalization. They do
  not validate the originating transport calculation.
• The 132-experiment FNS result and the FNG/ITER cell-620 result apply to their recorded materials, data, histories,
  and responses. Neither is a general safety or shutdown-dose qualification.
• Performance measurements and large-mesh extrapolations are fixture-specific sizing evidence, not guarantees.
• Data inputs and generated bulk libraries are not embedded in the software package or Git repository. The optional
  data release retains public provenance, terms, and SHA-256 identities; users remain responsible for the dataset they
  select.

See Qualification boundary for responsibilities when ACTINV is placed in a controlled analysis
chain and Validation for detailed evidence.

ACTINV data v1.0.0

ACTINV data v1.0.0

This is ACTINV's first immutable, ready-to-use nuclear-data release. It is separate from the software release: install
an ACTINV v1.0.0 standalone command, then let its embedded catalog select and verify only the files you need.

actinv data list
actinv data fetch
actinv data verify

The default tendl-2025-neutron bundle downloads about 139 MiB and installs the exact P10 TENDL-2025 709-group
neutron activation library plus primary ENDF/B-VIII.0 and fallback JEFF-3.3 decay data. The decay archives come
directly from the IAEA and are not assets on this release. Optional bundles provide the matching complete P11 MF=33
neutron covariance sidecar and the P10 proton, deuteron, or alpha 162-group activation libraries.

Every direct asset, official archive, and extracted decay payload has a frozen byte count and SHA-256 in
actinv-data-catalog-v1.0.0.json. Downloads are streamed, bounded, verified, and published atomically. SHA256SUMS
and SIZES cover every attached payload and the catalog. Files are installed under a versioned directory and are
never silently updated.

The processed TENDL-2025 assets are distributed under the CC-BY-4.0 declaration in their source evaluations. See
ACTINV-DATA-NOTICE-v1.0.0.md for creators, citation, source identities, transformations, the two recorded Pb-208
repairs, and the separation between data terms and ACTINV's MIT/Apache-2.0 software licence.

ACTINV and these data remain research-grade. A verified download proves file identity, not fitness for a particular
material, spectrum, safety case, or regulatory use. See the repository's qualification boundary and validation record
before relying on results in a controlled analysis chain.