++

backend/app/services/citizen_generator.py

@@ -378,14 +378,15 @@ class CitizenGenerator:
             )
             user_text += f"\n\nTARGET SEGMENTS (distribute citizens across these):\n{seg_text}"
 
-        response = await self.llm.generate(
-            system=system_prompt,
-            user=user_text,
-            json_mode=True,
-            max_tokens=4000,
-        )
-
-        data = parse_json(response)
+        response = await self.llm.generate(
+            system=system_prompt,
+            user=user_text,
+            json_mode=True,
+            max_tokens=4000,
+            purpose="cast_generate",
+        )
+
+        data = parse_json(response, purpose="居民名单生成")
         if isinstance(data, dict) and "citizens" in data:
             cast = data["citizens"]
         elif isinstance(data, list):
@@ -400,25 +401,26 @@ class CitizenGenerator:
             and cast
             and not any(self._has_cjk(str(c.get("name", ""))) for c in cast)
         ):
-            logger.warning("Cast generation returned non-CJK names despite zh locale, retrying. Names: %s", [c.get('name') for c in cast])
+            logger.warning("中文模式下居民姓名仍非中文，将重试。当前姓名：%s", [c.get('name') for c in cast])
             retry_user = user_text + (
                 "\n\n[STRICT REMINDER] All names, roles, personality hooks, and "
                 "stances MUST be written in Chinese characters. Do NOT use English."
             )
-            response = await self.llm.generate(
-                system=system_prompt,
-                user=retry_user,
-                json_mode=True,
-                max_tokens=3000,
-            )
-            data = parse_json(response)
+            response = await self.llm.generate(
+                system=system_prompt,
+                user=retry_user,
+                json_mode=True,
+                max_tokens=3000,
+                purpose="cast_generate_retry",
+            )
+            data = parse_json(response, purpose="居民名单生成重试")
             if isinstance(data, dict) and "citizens" in data:
                 cast = data["citizens"]
             elif isinstance(data, list):
                 cast = data
 
         if len(cast) < count:
-            logger.warning("Cast generation returned %d/%d citizens, padding", len(cast), count)
+            logger.warning("居民生成仅返回 %d/%d 人，将使用默认居民补齐", len(cast), count)
             while len(cast) < count:
                 idx = len(cast)
                 if self._locale == "zh":
@@ -468,12 +470,31 @@ class CitizenGenerator:
                 )
                 prompts.append((self._directive(PERSONA_SYSTEM_PROMPT), f"{world_ctx}\n\n{citizen_ctx}"))
 
-            responses = await self.llm.generate_batch(prompts, json_mode=True, max_tokens=800)
+            # A full profile contains a backstory, goals, beliefs, and
+            # resources. 800 tokens is too small for Chinese output and can
+            # truncate the JSON before the closing braces, causing avoidable
+            # parse failures.
+            responses = await self.llm.generate_batch(
+                prompts,
+                json_mode=True,
+                max_tokens=1400,
+                purpose="persona_fast",
+            )
+            # Sub-purpose differentiates persona_fast vs persona_enrich so
+            # timing stats aren't mixed together.
+            _persona_purpose = "persona_fast"
+            # Sub-purpose: differentiate "persona_fast" from "persona_enrich"
+            # so timing stats don't get bucketed together with the slow
+            # variants.
+            # (this is the persona_fast path — the enrich path is below)
+            # Sub-purpose: differentiate "persona fast" from "persona enrich" so
+            # timing stats don't get bucketed together with the slow variants.
+            # Callers above set this; default to persona_fast if unset.
 
             for i, response in enumerate(responses):
                 idx = batch_start + i
                 member = cast[idx]
-                data = parse_json(response)
+                data = parse_json(response, purpose=f"居民画像生成（批次索引={batch_start + i}）")
 
                 personality_data = data.get("personality", {})
                 personality = Personality(
@@ -536,7 +557,7 @@ class CitizenGenerator:
             max_tokens=3000,
         )
 
-        data = parse_json(response)
+        data = parse_json(response, purpose="居民关系生成")
         if isinstance(data, dict) and "relationships" in data:
             rel_list = data["relationships"]
         elif isinstance(data, list):
@@ -624,7 +645,7 @@ class CitizenGenerator:
 
             for i, response in enumerate(responses):
                 agent = batch[i]
-                data = parse_json(response)
+                data = parse_json(response, purpose=f"居民人生经历生成（居民={agent.name}，ID={agent.id}）")
                 agent.life_state = self._parse_life_history(data, agent)
 
     def _parse_life_history(self, data: dict, agent: AgentPersona) -> LifeState: