Summary

Closes #148.

The Auto Highlights API has been observed in production to return individual auto_highlights_result.results entries with no rank field set. The SDK currently declares rank: float as required, so the Pydantic validator raises:

pydantic.v1.error_wrappers.ValidationError: 1 validation error for TranscriptResponse
auto_highlights_result -> results -> 2 -> rank field required (type=value_error.missing)

The error is raised while parsing the overall TranscriptResponse — so a missing-field-on-one-entry becomes a hard failure that loses the entire transcript (text, utterances, words, sentiment, every other audio-intelligence result), not just the affected highlight.

Fix

 class AutohighlightResult(BaseModel):
     count: int
-    rank: float
+    rank: Optional[float] = None
     text: str
     timestamps: List[Timestamp]

A missing rank parses through cleanly and the absence is observable to callers via rank is None. No silent default-value substitution; the field is genuinely missing, and the type reflects that.

The change is narrowly scoped to the single field surfaced by the reporter. The other fields on AutohighlightResult (count, text, timestamps) remain required — they haven't been seen missing in the wild and aren't speculative-loosened here.

Tests

New regression test in tests/unit/test_auto_highlights.py:

def test_auto_highlights_parses_result_without_rank(httpx_mock: HTTPXMock):
    """
    Regression for #148. ... A missing rank parses as None and the rest
    of the response is preserved.
    """
    mock_response = factories.generate_dict_factory(
        AutohighlightTranscriptResponseFactory
    )()
    del mock_response["auto_highlights_result"]["results"][0]["rank"]

    _, transcript = unit_test_utils.submit_mock_transcription_request(
        httpx_mock, mock_response=mock_response,
        config=aai.TranscriptionConfig(auto_highlights=True),
    )

    assert transcript.error is None
    assert transcript.auto_highlights.results[0].rank is None
    assert transcript.auto_highlights.results[0].text is not None

Verified the test fails on master (current required-rank contract) and passes on this commit.

Test plan

• pytest tests/unit/test_auto_highlights.py -v — 3/3 pass (2 existing + 1 new)
• Verified new test fails when the rank: Optional[float] = None change is reverted via git stash
• Other AutohighlightResult fields (count, text, timestamps) still required — no inadvertent loosening
• Type annotation accurate: Optional[float] lets rank is None work as expected at the call site

Notes for review

• I considered a broader loosening (e.g. making count or text Optional too as forward-compat) but kept the diff scoped to the actually-reported field. If the API is known to omit other fields in newer responses, happy to widen — let me know.