Pydantic 2.13

Apply the Python 3 baseline first. These rules cover Pydantic 2.13 strict validation, model boundaries, validators, unions, serialization, settings, secrets, and resource controls.

Model boundaries and strictness

• Use separate models for raw input, validated domain commands, persistence, and public output.
• Set model_config = ConfigDict(extra='forbid') for security-sensitive input; the default extra='ignore' silently discards unexpected fields.
• Set ConfigDict(strict=True) or Field(strict=True) where coercion could change identity, authorization, money, or routing semantics; strict mode is off by default. Check JSON-specific conversions, such as date strings, and do not weaken these boundaries with per-call strict=False.
• Never pass external data to model_construct(): it performs no validation or nested model conversion, so strict=True, extra='forbid', and validators do not make that construction path safe.
• Do not pass unvalidated updates to model_copy(update=...); even deep=True copies values without validating the updates.
• Apply explicit string, numeric, collection, and nesting constraints to every attacker-controlled field.
• Keep ownership, role, tenant, status, price, audit, and server-generated fields out of client-input models.

Validators and context

• Use field validators for local invariants and model validators for cross-field security invariants.
• Keep SkipValidation, plain validators, and wrap validators that skip their handler out of fields that rely on Pydantic's type and constraint validation.
• Keep validators