pyproject.toml

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "constituent-reconciler"
version = "0.7.0"
description = "Offline-first pipeline that resolves and deduplicates nonprofit constituent records with a human review gate."
readme = "README.md"
requires-python = ">=3.12"
license = { file = "LICENSE" }
authors = [{ name = "Chelsea Kelly-Reif" }]
keywords = [
  "nonprofit",
  "record-linkage",
  "deduplication",
  "entity-resolution",
  "data-intake",
  "human-services",
]
classifiers = [
  "Development Status :: 3 - Alpha",
  "Intended Audience :: Developers",
  "License :: OSI Approved :: Apache Software License",
  "Programming Language :: Python :: 3.12",
  "Topic :: Scientific/Engineering :: Information Analysis",
]
dependencies = [
  "splink>=4.0,<5",
]

[project.scripts]
reconcile = "constituent_reconciler.cli:main"

[project.optional-dependencies]
extract = ["pdfplumber>=0.11"]
ocr = ["pytesseract>=0.3.10"]
# The AI assistant package (constituent_reconciler.assistant) is opt-in and
# never imported by the deterministic pipeline. This extra installs the
# default provider's SDK only; the Bedrock provider's boto3 dependency is
# intentionally left undeclared here, matching extract/seam.py's existing
# BedrockSeam convention (lazy-imported, install-hint-on-missing) rather
# than adding a second, inconsistent way to depend on boto3.
ai = ["anthropic>=0.40"]

[dependency-groups]
dev = [
  "pytest>=8",
  "pytest-cov>=5",
  "mypy>=1.18",
  "ruff>=0.15",
  "pip-audit>=2.7",
  "pdfplumber>=0.11",
  "anthropic>=0.40",
  "boto3>=1.34",
]

[project.urls]
Homepage = "https://github.com/ChelseaKR/constituent-reconciler"
Documentation = "https://github.com/ChelseaKR/constituent-reconciler/tree/main/docs"

[tool.hatch.build.targets.wheel]
packages = ["src/constituent_reconciler"]

[tool.ruff]
line-length = 100
src = ["src", "tests", "tools"]
extend-exclude = ["src/constituent_reconciler/_vendor/genai_telemetry"]

[tool.ruff.lint]
select = ["E", "F", "I", "UP", "B", "W", "S", "C90"]

[tool.ruff.lint.mccabe]
max-complexity = 10

[tool.ruff.lint.per-file-ignores]
# `assert` is the pytest idiom, not a runtime assertion to strip (S101); tests
# also deliberately construct fake credentials (S105/S106), a
# bind-all-interfaces call to assert it is refused (S104), and urlopen calls
# against the in-test loopback server (S310) — all expected-in-tests patterns,
# not findings. test_corpusgen.py's `random.Random` calls exercise the
# corpus generator's error channels directly and are never a security
# decision either (S311). Real code in src/ still gets the full S set with
# per-line noqas where a specific call is a reviewed false positive.
"tests/*" = ["S101", "S104", "S105", "S106", "S110", "S310", "S311"]
# The corpus generator's randomness is for reproducible synthetic test data,
# never a security decision, so `random.Random` (S311, "not for cryptographic
# purposes") is the correct choice here, not a finding.
"tools/corpusgen/*" = ["S311"]

[tool.mypy]
python_version = "3.12"
strict = true
files = ["src", "tests", "tools"]
explicit_package_bases = true
mypy_path = ["src"]

[[tool.mypy.overrides]]
module = [
  "splink.*",
  "pandas.*",
  "duckdb.*",
  "pdfplumber.*",
  "boto3.*",
  "botocore.*",
  "postal.*",
  "pytesseract.*",
]
ignore_missing_imports = true

[[tool.mypy.overrides]]
module = ["constituent_reconciler._vendor.genai_telemetry.*"]
ignore_errors = true

[tool.pytest.ini_options]
testpaths = ["tests"]
addopts = "-q --strict-markers --strict-config --import-mode=importlib --cov=constituent_reconciler --cov-branch --cov-report=term-missing --cov-fail-under=85"

[tool.coverage.run]
branch = true
source = ["constituent_reconciler"]
omit = ["src/constituent_reconciler/_vendor/*"]