The condition if types is None or item.type not in types: continue has inverted logic. When types is None (intended to mean 'capture all types'), the first condition types is None evaluates to True, causing continue to execute and skip ALL items. This defeats the purpose of having a default behavior that captures everything.

When should_send_default_pii() returns False, _filter_headers() returns AnnotatedValue objects for sensitive headers (like authorization, cookie, etc.). The new _get_request_attributes() function directly assigns these values to attributes. When set_attribute() processes these, format_attribute() will fall back to safe_repr(), producing unhelpful string representations like <AnnotatedValue object at 0x...> instead of properly filtered values. While no actual PII leaks (the original sensitive data is already stripped), the resulting attribute values are not meaningful.

The envelopes_to_spans helper accesses span_json["attributes"] directly, but in _span_batcher.py:116-119, the attributes key is only added to the span JSON when item._attributes is truthy. If a span has no attributes, this will raise a KeyError. This would cause tests to fail when processing spans without attributes.