[TexasCoding]

Wave 3-A of audit umbrella.

Pairs with the existing config-level test_request_bodies_use_extra_forbid to verify behavior, not just configuration. For each of the 23 Request/RequestOrder models in kalshi.models.__all__, builds minimal-valid kwargs via a new auto-builder (tests/_request_model_fixtures.py) and asserts pydantic.ValidationError with an extra_forbidden entry when a phantom field is passed.

23 new parametrized cases + 1 meta-test. Catches the failure mode where a contributor flips extra='allow' and adds a typo'd field — the config test catches the first miss; this test catches the second.

Closes #219
Refs #224

[claude]

Code Review — PR #235

Overall: Solid. The behavioral test is the right complement to the config-level gate. The auto-builder is clean and minimal. Two items worth addressing before merge, a few smaller nits.

---

What the PR does

Adds a parametrized behavioral test (test_request_model_rejects_phantom_field) that constructs each of the 23 Request/RequestOrder models with a minimal-valid kwargs set, then asserts that adding a phantom ghost_field raises ValidationError with an extra_forbidden error. Pairs with the existing config test (test_request_bodies_use_extra_forbid) to form a two-layer guard: config gate catches flipping extra=, behavioral gate catches the downstream consequence. Also adds a meta-test (test_minimal_kwargs_helper_handles_all_request_models) to give a clear failure path when the auto-builder encounters an annotation it can't resolve.

---

Issues

1. Meta-test ordering is inverted (minor correctness issue)

The meta-test docstring says:

"fails before the parametrized phantom test surfaces a confusing per-case error"

But in the file the meta-test is declared after test_request_model_rejects_phantom_field. With pytest's default declaration-order execution, the parametrized test would run and fail with a confusing RuntimeError before the meta-test runs to diagnose it. The two options:

• Move test_minimal_kwargs_helper_handles_all_request_models to just before the parametrized test (fixes ordering as the docstring implies)
• Or soften the docstring to say it's a diagnostic companion, not a pre-flight gate

2. _generate_value silently misses Enum fields

The walker handles primitives, Literal, Union, list, tuple, set, dict, and nested BaseModel. It does not handle Enum subclasses. Because Enum is a type and not a BaseModel, the isinstance(annotation, type) and issubclass(annotation, BaseModel) branch won't catch it, and _generate_value will raise RuntimeError. That's the intended signal — add an override — but it isn't documented. A one-liner in the module docstring alongside the existing list of supported types would make this expectation explicit:

# The walker handles primitives, ``Literal``, ``Union``/``Optional``,
# ``list[X]``, nested ``BaseModel`` subclasses, and ``UUID``. Enum,
# datetime, and other types must be supplied via ``_OVERRIDES``.

---

Smaller nits

• _OVERRIDES by class name vs. class object: The comment explains the reason (cross-module portability). Fine. Worth noting that if two Request models in different submodules ever share a name (unlikely but not impossible), the wrong override would silently win. A short # no two Request models should share a name comment would be a cheap safety note.

• Meta-test catches RuntimeError and TypeError but not ValidationError: If minimal_kwargs produces a value that passes the auto-builder but fails a model validator, the meta-test won't catch it — the parametrized test would blow up instead with a ValidationError. This is probably fine for now (the overrides system exists for exactly this case), but the except clause could say so:

  except (RuntimeError, TypeError) as e:  # ValidationError → override needed, caught by parametrized test

• CLAUDE.md says no multi-paragraph docstrings — test_request_model_rejects_phantom_field has a 3-sentence docstring. Test files arguably warrant more prose, but worth being consistent.

---

What's good

• The _OVERRIDES pattern is exactly the right design for edge cases — explicit, lazy (no import-time side effects), and self-documenting.
• RuntimeError as the signal for unsupported annotations is the right choice: loud failure beats silent wrong value.
• The ghost_field name is unambiguous; won't collide with real fields accidentally.
• Both types.UnionType (Python 3.10 X | Y syntax) and typing.Union are handled — covers both annotation styles Pydantic models may use.
• 150 lines of additions for 23 new test cases and a safety net is genuinely lean.

Fix the ordering issue and add the Enum/datetime note to the module docstring and this is good to go.