feat:

1. 建立问题库大纲，对应每个人生阶段槽位
2. 鼓励使用更生活化的交流语言共情与总结
3. 降低评审模型可能发生截断的概率
4. 成稿质量维度强化情感表达和上下文连贯性

api/app/features/evaluation/judge_manual_service.py

@@ -27,7 +27,7 @@ from app.features.evaluation.eval_trace_service import EvalTraceService
 from app.features.evaluation.judge_schemas import ConversationJudgeOutput
 from app.features.evaluation.judge_service import (
     EvalJudgeService,
-    eval_judge_compare_transcript_each_max_chars_for_context,
+    eval_judge_compare_bundle_caps,
     eval_judge_conversation_transcript_max_chars_for_context,
 )
 from app.features.evaluation.schemas import MemoirSectionBaselineOut
@@ -234,6 +234,7 @@ class EvalJudgeManualService:
                 f"replay_glm5_failed: {replay_result.error or 'unknown error'}"
             )
 
+        _cmp_total, _cmp_per_side = eval_judge_compare_bundle_caps(judge._ctx_tokens)
         bundle: dict[str, Any] = {
             "version": 1,
             "judged_at": datetime.now(timezone.utc).isoformat(),
@@ -250,9 +251,8 @@ class EvalJudgeManualService:
                 conv_cap=eval_judge_conversation_transcript_max_chars_for_context(
                     judge._ctx_tokens
                 ),
-                compare_cap_each=eval_judge_compare_transcript_each_max_chars_for_context(
-                    judge._ctx_tokens
-                ),
+                compare_cap_total=_cmp_total,
+                compare_per_side_cap=_cmp_per_side,
                 fixture_filename=fn,
             ),
             "compare_markdown": "",
@@ -363,6 +363,7 @@ class EvalJudgeManualService:
 
         acc["options"]["judge_model"] = resolved_model
         acc["fixture_filename"] = fn
+        _sse_cmp_total, _sse_cmp_per = eval_judge_compare_bundle_caps(judge._ctx_tokens)
         persist = True
         try:
             yield {
@@ -435,9 +436,7 @@ class EvalJudgeManualService:
                 full_transcript=replay_transcript
             )
             replay_judge = replay_result.output
-            acc["replay_judge"] = (
-                replay_judge.model_dump() if replay_judge else None
-            )
+            acc["replay_judge"] = replay_judge.model_dump() if replay_judge else None
             acc["compare_summary"] = build_conversation_compare_summary(
                 baseline_judge=baseline_judge,
                 replay_judge=replay_judge,
@@ -446,9 +445,8 @@ class EvalJudgeManualService:
                 conv_cap=eval_judge_conversation_transcript_max_chars_for_context(
                     judge._ctx_tokens
                 ),
-                compare_cap_each=eval_judge_compare_transcript_each_max_chars_for_context(
-                    judge._ctx_tokens
-                ),
+                compare_cap_total=_sse_cmp_total,
+                compare_per_side_cap=_sse_cmp_per,
                 fixture_filename=fn,
             )
             yield {
@@ -532,7 +530,9 @@ class EvalJudgeManualService:
 
         fn = (fixture_filename or "").strip() or None
         if not fn:
-            raise EvaluationBadRequestError("请选择基线 MD（fixture_filename）后再重试基准分")
+            raise EvaluationBadRequestError(
+                "请选择基线 MD（fixture_filename）后再重试基准分"
+            )
 
         try:
             turns, _ = read_user_export_fixture(fn)
@@ -568,6 +568,7 @@ class EvalJudgeManualService:
         judge, resolved_model = _make_eval_judge(judge_provider, judge_model)
         if not judge:
             raise EvaluationBadRequestError(_JUDGE_CONFIG_HINT)
+        _rt_cmp_total, _rt_cmp_per = eval_judge_compare_bundle_caps(judge._ctx_tokens)
         baseline_result = await judge.judge_conversation_result(
             full_transcript=baseline_transcript
         )
@@ -590,9 +591,8 @@ class EvalJudgeManualService:
                     conv_cap=eval_judge_conversation_transcript_max_chars_for_context(
                         judge._ctx_tokens
                     ),
-                    compare_cap_each=eval_judge_compare_transcript_each_max_chars_for_context(
-                        judge._ctx_tokens
-                    ),
+                    compare_cap_total=_rt_cmp_total,
+                    compare_per_side_cap=_rt_cmp_per,
                     fixture_filename=fn,
                 ),
                 "compare_markdown": "",
@@ -619,10 +619,7 @@ class EvalJudgeManualService:
                 sse_event="baseline_turn_judge",
             ):
                 idx = row.get("turn_index")
-                if (
-                    isinstance(idx, (int, float))
-                    and row.get("judge") is not None
-                ):
+                if isinstance(idx, (int, float)) and row.get("judge") is not None:
                     acc["baseline_turn_judges"][str(int(idx))] = row["judge"]
 
         acc["compare_markdown"] = ""
@@ -634,9 +631,8 @@ class EvalJudgeManualService:
             conv_cap=eval_judge_conversation_transcript_max_chars_for_context(
                 judge._ctx_tokens
             ),
-            compare_cap_each=eval_judge_compare_transcript_each_max_chars_for_context(
-                judge._ctx_tokens
-            ),
+            compare_cap_total=_rt_cmp_total,
+            compare_per_side_cap=_rt_cmp_per,
             fixture_filename=fn,
         )
         async for piece in judge.stream_conversation_compare(
@@ -682,7 +678,10 @@ class EvalJudgeManualService:
         trace_svc = EvalTraceService(self._db)
 
         def _chapter_evidence_notes(
-            lineage_tier: str, evidence_summary: str, truncated: bool, dropped: list[str]
+            lineage_tier: str,
+            evidence_summary: str,
+            truncated: bool,
+            dropped: list[str],
         ) -> str:
             drops = ",".join(dropped[:12]) if dropped else ""
             return (