Albucore public API classification

from albucore import * exposes only routers and shared helpers listed in albucore/__init__.py __all__.

Routers (package `all`)

User-facing entry points with benchmark-driven backends inside:

Arithmetic / weighted: add, add_constant, add_array, add_vector, add_weighted, multiply, multiply_by_constant, multiply_by_vector, multiply_by_array, multiply_add, power, normalize, normalize_per_image
Uint8 LUT (albucore.lut): sz_lut, apply_uint8_lut (StringZilla + OpenCV routing; re-exported on albucore for star imports)
I/O: to_float, from_float
Geometry / misc: hflip, vflip, median_blur, matmul, pairwise_distances_squared
Stats: mean, std, mean_std, reduce_sum (from albucore.stats)
Decorators: see decorators.__all__ in albucore/decorators.py
Geometric: copy_make_border, remap, resize, warp_affine, warp_perspective
Utils: see utils.__all__ in albucore/utils.py
Types / constants: ImageType, ImageUInt8, ImageFloat32, SupportedDType, NormalizationType, ValueType, MAX_OPENCV_WORKING_CHANNELS, etc.
Metadata: __version__, __author__, __maintainer__

Import explicitly from albucore.functions for tests and golden references:

Backend-specific: *_numpy, *_opencv, *_lut, *_cv2, hflip_numpy, vflip_numpy
NumKong helpers: add_weighted_numkong, add_array_numkong, multiply_by_constant_numkong (albucore.functions); add_constant_numkong lives on albucore.weighted only
LUT plumbing: create_lut_array, apply_lut, prepare_value_opencv, apply_numpy

Names prefixed with _ or used only inside albucore (not stable API).

from albucore import * no longer exposes backend-specific helpers (add_opencv, normalize_per_image_numpy, …). Import them explicitly: from albucore.functions import add_opencv.
Deprecated SimSimd aliases are removed; use add_weighted_numkong (and other *_numkong helpers) from albucore.functions if needed.
New stats entrypoints: from albucore.stats import mean, std, mean_std, reduce_sum (also re-exported on albucore for star-import users).