feat: make contexts optional in /api/score

When contexts is absent, metrics that require retrieved_contexts
(faithfulness, context_recall, context_precision, noise_sensitivity)
are automatically skipped and appear in skipped_metrics.
Only answer_relevancy, factual_correctness, semantic_similarity
remain computable without contexts.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

webapp/api/score.py

@@ -73,7 +73,8 @@ def score_sample(
     用于日志记录、质量监控或触发 Agent 自我改进流程。
 
     **contexts 格式**：多个检索片段用 `context_separator`（默认 `" |||| "`）拼接为一个字符串，
-    服务端自动拆分后传入 RAGAS 管道。
+    服务端自动拆分后传入 RAGAS 管道。**contexts 为可选字段**，缺失时自动跳过依赖检索内容的指标
+    （`faithfulness`、`context_recall`、`context_precision`、`noise_sensitivity`）。
 
     **ground_truth 可选**：
     - 提供时：所有指定指标均参与计算。
@@ -99,12 +100,13 @@ def score_sample(
     """
     client = f"{raw_request.client.host}:{raw_request.client.port}" if raw_request.client else "unknown"
     logger.info(
-        "[score] incoming  client=%s  method=%s  content_type=%s  metrics=%s  has_gt=%s",
+        "[score] incoming  client=%s  method=%s  content_type=%s  metrics=%s  has_gt=%s  has_ctx=%s",
         client,
         raw_request.method,
         raw_request.headers.get("content-type", ""),
         request.metrics,
         request.ground_truth is not None,
+        bool(request.contexts),
     )
     settings = _get_settings()
 

webapp/models.py

@@ -384,6 +384,14 @@ _GT_DEPENDENT_METRICS: frozenset[str] = frozenset({
     "noise_sensitivity",
 })
 
+# 需要 contexts 才能计算的指标集合
+_CONTEXT_DEPENDENT_METRICS: frozenset[str] = frozenset({
+    "faithfulness",
+    "context_recall",
+    "context_precision",
+    "noise_sensitivity",
+})
+
 # 所有合法指标名称
 _VALID_METRICS: frozenset[str] = frozenset({
     "faithfulness",
@@ -428,8 +436,9 @@ class ScoreRequest(BaseModel):
 
     question: str = Field(description="问题文本。")
     answer: str = Field(description="待评分的回答。")
-    contexts: str = Field(
-        description="检索上下文字符串，多段之间用 context_separator 拼接。"
+    contexts: str | None = Field(
+        default=None,
+        description="检索上下文字符串，多段之间用 context_separator 拼接。缺失时自动跳过依赖检索内容的指标（faithfulness、context_recall、context_precision、noise_sensitivity）。",
     )
     ground_truth: str | None = Field(
         default=None,
@@ -467,15 +476,23 @@ class ScoreRequest(BaseModel):
         return value
 
     def contexts_as_list(self) -> list[str]:
-        """Split the contexts string into a list of non-empty fragments."""
+        """Split the contexts string into a list of non-empty fragments.
+
+        Returns an empty list when contexts is None or blank.
+        """
+        if not self.contexts:
+            return []
         separator = self.context_separator or " |||| "
         return [part.strip() for part in self.contexts.split(separator) if part.strip()]
 
     def effective_metrics(self) -> list[str]:
-        """Return metrics filtered to exclude GT-dependent ones when ground_truth is absent."""
-        if self.ground_truth is not None:
-            return list(self.metrics)
-        return [metric_name for metric_name in self.metrics if metric_name not in _GT_DEPENDENT_METRICS]
+        """Return metrics filtered to exclude GT-dependent or context-dependent ones when inputs are absent."""
+        result = list(self.metrics)
+        if self.ground_truth is None:
+            result = [m for m in result if m not in _GT_DEPENDENT_METRICS]
+        if not self.contexts:
+            result = [m for m in result if m not in _CONTEXT_DEPENDENT_METRICS]
+        return result
 
 
 class ScoreResponse(BaseModel):