pyproject.toml

[project]
name = "islamic-qa-evaluator"
version = "0.1.0"
description = "LLM-as-judge evaluation framework for Islamic Q&A responses"
requires-python = ">=3.10"
dependencies = [
    "litellm>=1.40.0",
    "openai>=1.30.0",
    "anthropic>=0.25.0",
    "pydantic>=2.6.0",
    "pandas>=2.2.0",
    "numpy>=1.26.0",
    "pyyaml>=6.0.1",
]

[tool.ruff]
line-length = 120
target-version = "py311"
exclude = [
    "venv",
    ".venv",
    "data",
]

[tool.ruff.lint]
select = [
    "E",  # pycodestyle errors
    "F",  # pyflakes
    "I",  # isort (import sorting)
    "UP", # pyupgrade (modernization)
    "B",  # flake8-bugbear
    "W",  # pycodestyle warnings
]
ignore = [
    "B008", # FastAPI dependency defaults are function calls by design
    "E501", # line length is enforced by the formatter, not the linter
    "UP042", # StrEnum requires Python 3.11+, project targets 3.10 for compatibility
]

[tool.ruff.lint.isort]
# Keep `from x import a, b as c` on one statement instead of splitting it.
combine-as-imports = true

[tool.ruff.lint.per-file-ignores]
# Tests routinely rebind fixtures and assert on broad exception shapes.
"tests/**" = ["B011"]

[tool.ruff.format]
quote-style = "double"
indent-style = "space"

[tool.mypy]
python_version = "3.11"
# scripts/ and test dirs have no __init__.py; without explicit package bases
# mypy sees e.g. scripts/foo.py as both "foo" and "scripts.foo".
plugins = ["pydantic.mypy"]
mypy_path = "."
explicit_package_bases = true
namespace_packages = true
warn_unused_configs = true
warn_redundant_casts = true
no_implicit_optional = true
exclude = [
    "^venv/",
    "^\\.venv/",
]

# main.py is the reference for the standard CONTRIBUTING.md sets: every
# function there carries parameter and return annotations. The rest of the
# tree is type-checked but not yet required to be fully annotated.
[[tool.mypy.overrides]]
module = "main"
disallow_untyped_defs = true

# store.py has pre-existing type errors unrelated to cache implementation
[[tool.mypy.overrides]]
module = "store"
ignore_errors = true

# google-generativeai and its api_core siblings ship incomplete stubs.
[[tool.mypy.overrides]]
module = [
    "google.*",
    "stellar_sdk.*",
    "redis.*",
    "yaml.*",
    "firebase_admin.*",
]
ignore_missing_imports = true

[tool.evaluation]
correlation_target = 0.85
inter_judge_agreement_target = 0.80
cost_per_eval_limit = 0.05
throughput_per_hour_target = 100
false_positive_rate_max = 0.08
calibration_drift_limit = 0.05