chore(main): release nyxus 0.12.0

[github-actions]

🤖 I have created a release beep boop

0.12.0 (2026-09-01)

⚠ BREAKING CHANGES

• omezarr: OME-Zarr support now requires z5 3.0.1 and C++20; xtensor/xsimd are no longer used.

Features

• add 2D test coverage for touch and morphology feature families (3f576d9)
• add IBSI Intensity Histogram features, per-bin HISTOGRAM output, and --mergerois (5b517f4)
• add unittest coverage to all 214 3D features, labeling tests with clear ground truth (b89a9ac)
• Benchmark area fix (#345) (b532f21)
• helpers: add ceil_pow2 (tight next power of two) (d0992dc)
• hounsfield: --preserve-hu CT/Hounsfield intensity preservation (1b5667f)
• hounsfield: keep HU offset base at scanned slide min under fp-options; enable bare CLI --preserve-hu (43f8063)
• hounsfield: wire --preserve-hu through the 3D NIfTI loader (65c9ad9)
• omezarr: migrate to z5 3.0.1 ArrayView API, drop xtensor (225b9b9)
• re-organize and create complete implementation unit test coverage for 492 2D features (#346) (011df60)
• vetting: check_coverage CLI (--check/--write) + drift guard (0ad65ee)
• vetting: check_coverage registry load + schema/token validation (44d2bfa)
• vetting: coverage stats + markdown report generation (0f85581)

Bug Fixes

• 2d zernike: seed the angular tables at repetition 0, not 1 (2eb3cc3)
• 3d ngtdm: default the neighbourhood radius so the family is not NaN (53b04b5)
• 3d-ngldm: DCNU/DCNUN summed the wrong marginal (== GLNU) (d3198e5)
• 3d-shape: correct mislabeled axis lengths + vet 7 morphology features (0af100b)
• 3d: compute 3ROBUST_MEAN (was hardcoded 0) + vet 18 3D first-order features (79d0974)
• audit and fix 2D feature tests 1/n (c5cf260)
• caliper: correct Martin/Nassenstein diameters + vet 12 features vs imea (bc3e92a)
• caliper: float-precision hull rotation + address PR #381 review (3377183)
• categorize and name feature unit tests and expected values (d854679)
• chords: correct max-chord angle index and all-chords histogram source (bd9a930)
• cleare state and start fresh to avoid random test failure. (495f05c)
• convex-hull: Pick's-theorem hull area so SOLIDITY <= 1 (0319932)
• emit per-bin HISTOGRAM values in arrow/parquet output (01ac9c2)
• feature_mgr: extend check_11_correspondence() to Feature3D (b188e50)
• feature_mgr: loop 3D correspondence check from Feature3D::FIRST (9287f3c)
• fractal-dim: adaptive box counting and a clean divider perimeter (c981743)
• fractal-dim: address PR review comments (98a4f2d)
• fractal-dim: align the ROI to the box-counting grid (98e53ec)
• fractal-dim: mean log-log slope for box-count, Richardson D for perimeter (af37142)
• gabor: keep filter-response magnitudes real-valued, then vet GABOR vs scikit-image (2b2c7d7)
• glcm: default co-occurrence offset to 1 and exclude background (02c5a75)
• glcm: normalize HOM2/ENTROPY by sum_p (2D+3D) + fully oracle-vet GLCM (60a1ebb)
• gldm: exclude background from the dependence matrix (f1f8eb4)
• gldzm: build zones from the full 8-neighbourhood (fc0bbde)
• handle multi-valued HISTOGRAM in buffer output (Python pandas/arrow) (246f4dd)
• hexagonality,lstsq: use std::abs so gcc/Linux doesn't truncate doubles to int (88a6f53)
• imq: correct FOCUS_SCORE overflow + abs()-variance, vet 4 imq features vs OpenCV and CellProfiler (7eed701)
• intensity_histogram: report IH_ROBUST_MEAN_IDX as a 1-based bin index (0527f48)
• issue 327 single roi polus (#330) (03eeb0c)
• loader: clamp negative signed-int pixels so int16 CT doesn't wrap to a huge unsigned value (fac1bac)
• make imagecodecs PIP_ONLY_BINARY (b5fa20a)
• migrate build_and_test from macos-15-intel to macos-14 (ac70de1)
• mitigate exposed bug due to incorrect reading of bx in 3d_surface (24d18ee)
• moments: correct Hu invariant h5/h6 formulas (9x bracket, stray +eta03) (73f6ca8)
• moments: use fractional centroid in 2D central moments (drop (int) truncation) (5cbf94a)
• morphology: geodetic length/thickness from real perimeter (was size_t-truncated) (ce7716f)
• neighbors: exact min distance for the touch/radius test (PERCENT_TOUCHING) (92de774)
• neighbors: implement closest-neighbor dist/ang and cap PERCENT_TOUCHING (4322c88)
• ooc: make the 2D out-of-core feature paths match the in-RAM path (9a9ad02)
• python: release an instance's Environment so a recycled id() cannot inherit it (6dd550f)
• remove redundant calculations and avoid flushing valid old values (cdf48cf)
• review comments on PR 352 (ec42edc)
• stop featurize() mutating its shared default name lists (372172e)
• upgrade Docker build to CUDA 12.6 and Python 3.11 (80b548d)
• use raw docstrings to silence invalid escape sequence SyntaxWarning (e3bf05c)
• vetting: band the 3D GLSZM entropy pin for a platform, not for a machine (ab8573e)
• vetting: correct radial follow-up records (7c4657a)
• vetting: let the NGTDM coverage scanner read one-line TEST registrations (51df88e)
• vetting: normalize oracle_coverage.csv to LF, drop dead current_test notes (7dc26ed)
• vetting: write every generated CSV and report with LF (91eb35e)

Performance Improvements

• omezarr: open the z5 dataset once, not per tile (9b36f84)

Documentation

• 3d gldzm: record what the retired coverage sweep checked, and what it did not (483dbbc)
• 3d ngldm: say what the generator verifies, and what the Ns agreement proves (9310932)
• add CLAUDE.md codebase and workflow guide (e301fe6)
• caliper: correct inaccurate/stale comments (no code change) (cc5b4de)
• CLAUDE: add git workflow and tests/vetting conformance rules (a548b57)
• CLAUDE: address PR #388 review — C++20; agentic git workflow (466c038)
• CLAUDE: address PR #388 review — generalize paths, drop z5py pin, add no-local-paths rule (4235e0c)
• imq: say in the matrix that this pass vets the in-RAM paths only (1591760)
• test: correct IQR/QCoD _IDX framing — discrete percentile is IBSI-conformant, not a bug (29214ba)
• tests: add vetting audit baseline, fraclac oracle, and README (e5232c0)
• tests: box-count golden is FracLac-vetted, not a self-consistency pin (8ccc76b)
• tests: neighbor features are partly tool-vetted (CellProfiler), not analytic-only (7910549)
• tests: oracle tool local-setup research (TOOLS.md) (65e0b79)
• tests: spec for the oracle-vetting test framework (f44ff25)
• vetting: add skimage to the SPEC 4 oracle-token set (3ee1dfd)
• vetting: add the missing radial config matrix (fe4ef56)
• vetting: audit the 2D GLCM goldens against fresh oracle runs (0484103)
• vetting: audit the 2D GLRLM goldens against fresh oracle runs (67f24ed)
• vetting: correct 3D first-order oracle accounting (f6ad06a)
• vetting: correct split to 16 ibsi + 4 analytic; file IQR_IDX flooring bug (726b082)
• vetting: correct what the 3D first-order MATLAB oracle backs (2ccd560)
• vetting: design for IBSI-vetted IH dispersion/index tests (07b1b75)
• vetting: drop the references to the deleted 3D morphology MATLAB header (071b52d)
• vetting: fix stale token count in plan example (16 ibsi + 4 analytic) (3255fef)
• vetting: give the constant-ROI cell a SPEC disposition, not an invented one (834e59d)
• vetting: give the neighbour rows a tier, a test name, a benchmark and a matrix (7981b2a)
• vetting: implementation plan for IBSI-vetted IH dispersion/index tests (4dfb1ff)
• vetting: lead the 3D first-order report with the current state (004bfc5)
• vetting: lead the 3D first-order report with the current state (25c57f2)
• vetting: migration map + seeded oracle_coverage.csv (Phase 1, no code changes) (77f3d07)
• vetting: name Octave wherever the matlab oracle is presented (d70e3fe)
• vetting: narrow what the 3D coverage sweep asserts that nothing else does (2a75d34)
• vetting: record licensed 2D morphology provenance (3d039d6)
• vetting: record reconciled IBSI IH phantom config + index base (99df244)
• vetting: record the 2D GLDM audit (3ee009d)
• vetting: record the 2D GLDZM audit (f29924d)
• vetting: record the 2D GLSZM audit (176476a)
• vetting: record the 2D neighbour audit (50d0ff8)
• vetting: record the 2D NGTDM audit (23c609f)
• vetting: record the 3D GLDM default cell and the degenerate ROI it discards (12ef315)
• vetting: record the 3D GLDM recipes, matrix, benchmarks and audit (cac4e99)
• vetting: record the 3D GLSZM audit, its recipes and its benchmark (33e0fbe)
• vetting: record the 3D NGTDM audit, its recipes and its benchmark (55d5e90)
• vetting: record the config scope of the gldm promotions (7 are mode-invariant) (0780912)
• vetting: record the config scope of the glrlm promotions (285bc39)
• vetting: record the config scope of the glszm promotions (c4bb8b1)
• vetting: record what the 3D GLSZM review round measured (b7126de)
• vetting: resolve missing oracles via deep-dive research (Phase 2) (2fa0515)
• vetting: resolve the Phase-2 reconciliation decisions (SPEC 6) (8b45ddd)
• vetting: retract unproven 3D NGLDM "ibsi" promotions; document MIRP disagreement (b924f8e)
• vetting: scaffold matrix/ and oracles/gen_ conventions (GLCM example) (61fdc66)
• vetting: seed config_recipes.md (b64c82b)
• vetting: separate unproven provenance from unreproducible provenance (058cb00)
• vetting: update golden_regen doc for the flat rel=1e-8 tolerance (f84773f)

Code Refactoring

• 3d surface: describe the eigenvalue ordering rather than the defect it replaced (34a74c2)
• address PR #351 review nits (c623405)
• tests: route fractal fixture through the shared masked-loader helper (050e672)

Build System

• arrow: require Boost for Arrow instead of Z5 (43a584f)

Continuous Integration

• release-please: use manifest config so 0.x breaking = minor, and pin 0.12.0 (a500bd9)

---

This PR was generated with Release Please. See documentation.