Implementation:Deepset ai Haystack ExtractedAnswer Dataclass

Overview

ExtractedAnswer is a Python dataclass that represents a span-based answer extracted from a document. It captures the answer text, its position within the source document, surrounding context, a confidence score, and a reference to the source Document. This is a pattern document describing a data structure interface.

Source Location

• File: haystack/dataclasses/answer.py, lines 28-95
• Dataclass: ExtractedAnswer
• Decorator: @dataclass

Import

from haystack.dataclasses import ExtractedAnswer

Fields

Field	Type	Default	Description
query	str	(required)	The original question that produced this answer
score	float	(required)	Confidence score for this answer
data	None	None	The extracted answer text; None indicates a "no answer" prediction
document	None	None	Reference to the source Document from which the answer was extracted
context	None	None	Surrounding text providing context for the extracted answer
document_offset	None	None	Character positions (start, end) of the answer within the source document
context_offset	None	None	Character positions (start, end) of the answer within the context string
meta	dict[str, Any]	{}	Additional metadata (e.g., computed page numbers)

Nested Span Dataclass

@dataclass
class ExtractedAnswer:
    # ... fields ...

    @dataclass
    class Span:
        start: int
        end: int

The Span nested dataclass represents a character range with:

• start (int): Inclusive start character position.
• end (int): Exclusive end character position.

Used for both document_offset and context_offset.

Full Dataclass Definition

@dataclass
class ExtractedAnswer:
    query: str
    score: float
    data: str | None = None
    document: Document | None = None
    context: str | None = None
    document_offset: Optional["Span"] = None
    context_offset: Optional["Span"] = None
    meta: dict[str, Any] = field(default_factory=dict)

    @dataclass
    class Span:
        start: int
        end: int

Methods

to_dict()

def to_dict(self) -> dict[str, Any]:

Serializes the ExtractedAnswer to a dictionary:

• Converts the document field using Document.to_dict(flatten=False).
• Converts document_offset and context_offset Span objects using dataclasses.asdict().
• Wraps all fields using default_to_dict() for Haystack-compatible serialization.

from_dict(data)

@classmethod
def from_dict(cls, data: dict[str, Any]) -> "ExtractedAnswer":

Deserializes an ExtractedAnswer from a dictionary:

• Reconstructs the document field using Document.from_dict().
• Reconstructs document_offset and context_offset as ExtractedAnswer.Span objects.
• Uses default_from_dict() for Haystack-compatible deserialization.

Protocol Conformance

ExtractedAnswer conforms to the Answer protocol (defined in the same module), which requires:

@runtime_checkable
@dataclass
class Answer(Protocol):
    data: Any
    query: str
    meta: dict[str, Any]

    def to_dict(self) -> dict[str, Any]: ...

    @classmethod
    def from_dict(cls, data: dict[str, Any]) -> "Answer": ...

Usage Example

from haystack.dataclasses import ExtractedAnswer

# A regular extracted answer
answer = ExtractedAnswer(
    query="What is Python?",
    score=0.95,
    data="a popular programming language",
    document=some_document,
    context="Python is a popular programming language used worldwide",
    document_offset=ExtractedAnswer.Span(start=12, end=42),
    context_offset=ExtractedAnswer.Span(start=12, end=42),
)

# A "no answer" entry
no_answer = ExtractedAnswer(
    query="What is Python?",
    score=0.05,
    data=None,
    document=None,
)

# Serialization round-trip
answer_dict = answer.to_dict()
restored = ExtractedAnswer.from_dict(answer_dict)

Related Pages

Principle:Deepset_ai_Haystack_Extracted_Answer_Schema