ACTINV v1.2.1

ACTINV v1.2.1 — release notes

ACTINV 1.2.1 is a patch release. It corrects the self-shielding interpolation, turns several silent
input and data problems into errors or ledger entries, and updates rustls past RUSTSEC-2026-0285.
There are no new features. Schemas and interfaces are unchanged apart from the stricter validation
listed below, and the embedded catalog stays data-v1.1.0.

Upgrade the Python package with:

python -m pip install --upgrade actinv

Rust CLI users can install this exact version with:

cargo install --locked --force actinv-cli --version 1.2.1

Results to re-check

• Self-shielded runs. The σ₀ interpolation weight was mirrored (sigma0_b is descending). Any
  shielded run whose effective σ₀ fell strictly inside the table grid used factors from the wrong point
  in its interval: all dilution: "composition" runs, and any fixed σ₀ other than the grid ends. For
  example, σ₀ = 1e3 b returned the 1e4 b row. This includes the shielded rows folded into MF=33
  uncertainty. Runs at σ₀ = 0.1 b or ≥ 1e10 b were unaffected.
• Compositions with a misspelt or abundance-free element key ("Fee", "Tc"). These silently
  contributed zero atoms, and with atom_fraction the remaining members were renormalised. They now
  fail validation.
• Study refinement, robustness and comparison on array responses (inventory_per_nuclide,
  photon_source_per_group). Refinement reported satisfied and robustness reported 0 ± 0
  regardless of the data. These consumers now require one of the three scalar responses.
• Runs with JEFF-3.3 or UKDD-2020 as the primary decay library. When a state's branching ratios
  summed to less than 1, the unassigned fraction of its decays vanished; JEFF-3.3 Ir-169, for example,
  lists only its 45% alpha branch. The default ENDF/B-VIII.0 data is unaffected and gives byte-identical
  results.

What changes

• Decay-branch accounting. A branching shortfall beyond a 1e-5 rounding allowance is routed to
  leakage. An excess is scaled away, so decay never creates atoms. Both cases are listed in
  ledger.decay_branching_sums_off_unity, which appears only when needed. Radioactive records with no
  decay modes, and branches that resolve to their own parent, also go to leakage.
• Non-finite results fail. A run whose result holds NaN or infinity now stops with the field's path
  (for example steps[3].heat_W_per_g.total is NaN). Previously the value was written as JSON null.
• New validation errors:
  • a positive spectrum.total with an all-zero shape;
  • a duration longer than 64 characters;
  • self_shielding on a charged-projectile spec;
  • study names containing __;
  • more than 4096 robustness.samples;
  • a per-step study spectrum;
  • duplicate (ZA, LISO) decay records;
  • a resolved-range NAPS other than 0 or 1;
  • MF=6/8/9 products without a reaction;
  • MF=33 blocks over the 1 GB array cap;
  • imported meshes over 100 million cells;
  • --grid-density above 64.
• Robustness:
  • actinv data fetch has network timeouts.
  • A cache lock left by a killed process is reclaimed.
  • Mesh resume streams its output.
  • Hash-pinned inputs are parsed from exactly the bytes that were hashed.
  • Trace-mode heat and photon sums no longer depend on hash-map order.
  • build-damage/build-shielding caches are keyed by projectile and code version.
  • Fission-yield files use the canonical ENDF number parser.
• Python binding. run, reverse, cram_step and broaden release the GIL. broaden raises
  ValueError instead of panicking on bad input.
• Desktop and web workbench. Numeric fields refuse "nan" and "inf". Results load off the UI thread.

Security

• rustls 0.23.45 in both lockfiles (RUSTSEC-2026-0285, reachable through actinv data fetch). v1.2.0
  wheels and cargo install --locked builds carry the vulnerable 0.23.43.
• The library crates and the actinv binary declare #![forbid(unsafe_code)].

Qualification boundary

Unchanged — docs/DATA_LIMITATIONS.md and docs/QUALIFICATION.md apply as before. The full list of
changes is in CHANGELOG.md.