Safe default for Context.score (#1185)

Author: jamesbraza

src/paperqa/types.py

@@ -216,6 +216,9 @@ class Context(BaseModel):
     # or useful excerpts of text
     model_config = ConfigDict(extra="allow")
 
+    # Value was chosen to be below a 0-10 scale, making the 'unset' nature obvious
+    UNSET_RELEVANCE: ClassVar[int] = -1
+
     id: str = Field(
         default=AUTOPOPULATE_VALUE,
         description="Unique identifier for the context. Auto-generated if not provided.",
@@ -224,8 +227,8 @@ class Context(BaseModel):
     context: Annotated[str, StringConstraints(strip_whitespace=True)] = Field(
         description=(
             "Summary of the text with respect to a question."
-            " Can be an empty string if a summary is not useful"
-            " (which should accompany a score of 0)."
+            " Can be an empty string if a summary is not useful/irrelevant"
+            " (which should be paired with a score of 0)."
         )
     )
     question: str | None = Field(
@@ -236,7 +239,15 @@ class Context(BaseModel):
         ),
     )
     text: Text
-    score: int = 5
+    score: int = Field(
+        default=UNSET_RELEVANCE,
+        description=(
+            "Relevance score for this context to the question."
+            " The range used here is 0-10, where 0 is 'irrelevant',"
+            " 1 is barely relevant, and 10 is most relevant."
+            " The default is -1 to have a 'sorting safe' default as sub-relevant."
+        ),
+    )
 
     CONTEXT_ENCODING_LENGTH: ClassVar[int] = 500  # chars
     ID_HASH_LENGTH: ClassVar[int] = 8  # chars