This directory is the reproducible source for NumSharp's public API coverage artifact. It compares the public exports of pinned NumPy 2.4.2 with the public surface of the compiled NumSharp assembly.

python -m pip install numpy==2.4.2
python coverage/generate_coverage.py
python coverage/generate_coverage.py --check
python coverage/audit_documentation.py

Generated, reviewable outputs live in coverage/generated/:

• coverage.json — complete machine-readable inventory used by the documentation dashboard.
• coverage.csv — flat data for spreadsheets and downstream tooling.
• summary.md — human-readable totals and the highest-priority gaps.
• manifest.json — schema, tool versions, scope, and counting rules.

CI generates a fresh copy under artifacts/numpy-numsharp-coverage/, validates every headline-scope link against NumPy's official latest-stable Sphinx inventory, compares the result byte-for-byte with the checked-in dashboard data, and uploads the fresh directory as the numpy-numsharp-api-coverage artifact.

The default denominator includes NumPy top-level callables, ndarray methods and properties, and callables from numpy.random, numpy.linalg, and numpy.fft. NumPy types, constants, and modules are catalogued but do not affect the headline percentage. NumSharp-only APIs are catalogued separately and also do not affect it.

Platform-conditional extended-precision aliases (float96, float128, complex192, and complex256) are excluded so the artifact is byte-identical across Windows and Linux. NumPy's portable longdouble and clongdouble names remain catalogued.

• Exact — the corresponding NumSharp surface has the same public member name.
• Alias — a reviewed or mechanically safe C# equivalent exists under another name or surface.
• Partial — an API exists, but the reviewed mapping has a known semantic limitation.
• Unsupported — a public compatibility symbol exists but does not implement the NumPy capability.
• Missing — no NumSharp public API mapping was found.
• NumSharp-only — an unmatched public member declared by np, NDArray, or NumPyRandom; these rows link directly to their declaration on GitHub.

API availability is not a blanket behavioral-parity claim. Exact edge-case, dtype, layout, and signature parity still requires differential tests. Record reviewed exceptions and cross-surface aliases in coverage/overrides.json; the generator validates every referenced NumSharp target.