Validation and data models

Library	Use
Pydantic v2	Validation, coercion and serialization at I/O boundaries: HTTP payloads, config files, external API responses, model output.
`dataclasses`	Stdlib internal value objects. `slots=True` and `frozen=True` reduce memory use and prevent mutation.
attrs	Similar scope to dataclasses with more features: validators, converters, `__init__` customization.
pydantic-settings	Loads and validates configuration from environment variables, `.env` files and secrets directories.
`msgspec`	Alternative serialization and validation library with lower overhead, no coercion by default.

Validate at the process boundary and use plain objects internally:

class CreateJob(BaseModel):          # boundary
    symbol: str
    window: int = Field(gt=0, le=512)

@dataclass(frozen=True, slots=True)  # internal
class Job:
    symbol: str
    window: int

Configuration validated at startup fails immediately on a missing or malformed value rather than at first use:

class Settings(BaseSettings):
    database_url: PostgresDsn
    log_level: Literal["DEBUG", "INFO", "WARNING"] = "INFO"
    model_config = SettingsConfigDict(env_file=".env")