STUB-PAYLOAD RE-RUN — e1e22d48db4d
Claim: the custom-provider stub captured a payload
without the 'model' field.

--- suite tests at HEAD (tests/test_role_providers.py::PresetRequestShapeTests):
test_openai_compatible_presets (test_role_providers.PresetRequestShapeTests.test_openai_compatible_presets) ... ok
test_model_override_and_temperature_quirk (test_role_providers.PresetRequestShapeTests.test_model_override_and_temperature_quirk) ... ok
test_openai_compatible_escape_hatch (test_role_providers.PresetRequestShapeTests.test_openai_compatible_escape_hatch) ... ok

----------------------------------------------------------------------
Ran 3 tests in 0.004s

OK

--- direct stub captures at HEAD:
  spec openai-compatible:...|my-model|CUSTOM_KEY_ENV
    url: https://llm.example.com/v1/chat/completions
    payload keys: ['max_tokens', 'messages', 'model', 'temperature']
    payload['model'] = 'my-model'
  spec deepseek:deepseek-reasoner (preset:model override)
    url: https://api.deepseek.com/chat/completions
    payload keys: ['max_tokens', 'messages', 'model', 'temperature']
    payload['model'] = 'deepseek-reasoner'

PROBE OUTCOME: every captured payload at HEAD
carries 'model' — for presets, for preset:model
overrides, and for the custom escape-hatch spec.
The evidence gap the claim rested on does not
reproduce; the refutation's mechanism (model =
desc['model'], override included) is confirmed by
direct capture. Supports the pre-filled
recommendation: sustain refutation.
