fix: resolve legacy evidence targets without guessing and guide retries

backend/wisegrade/llm.py

@@ -238,11 +238,19 @@ def validate_score(result: ScoreResult, criterion: dict, chunks: list[dict], *,
             # Empty identifiers are accepted only for pre-object legacy rubrics.
             if any(isinstance(value, dict) for value in criterion.get("required_evidence", [])):
                 raise ModelError("模型证据缺少 target_id")
-            target = default_content or (targets[0] if targets else None)
+            associated = [by_target[key] for key in chunk.get("target_ids", []) if key in by_target]
+            if associated:
+                if len(associated) != 1:
+                    raise ModelError(f"模型证据缺少 target_id：chunk_id={chunk['chunk_id']} 对应多个检索目标，请明确选择")
+                target = associated[0]
+            else:
+                target = default_content or (targets[0] if targets else None)
         if target is None:
             raise ModelError("模型证据未关联任何必备证据目标")
         if chunk.get("target_ids") and target["target_id"] not in chunk["target_ids"]:
-            raise ModelError("模型证据未关联该片段的检索目标")
+            allowed = ",".join(str(key) for key in chunk["target_ids"] if key in by_target)
+            raise ModelError(f"模型证据未关联该片段的检索目标：chunk_id={chunk['chunk_id']}，"
+                             f"target_id={target['target_id']}；允许 target_id={allowed}")
         if not page_type_allowed(page_type, target["allowed_page_types"]):
             raise ModelError(f"模型证据来自不允许的页面类型（{page_type}），不能支持「{target['label']}」")
         if target["use"] == USE_CONTENT and not content_evidence_allowed(page_type):