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/tests/test_judge_service.py

@@ -11,8 +11,11 @@ from app.features.evaluation.judge_schemas import ConversationJudgeOutput
 from app.features.evaluation.judge_service import (
     EvalJudgeService,
     _build_memoir_judge_prompt,
-    eval_judge_conversation_transcript_max_chars,
+    conversation_judge_transcript_excerpt,
     eval_judge_compare_transcript_each_max_chars,
+    eval_judge_conversation_transcript_max_chars,
+    trim_compare_transcript_pair,
+    turn_judge_prior_excerpt,
 )
 
 
@@ -58,7 +61,9 @@ async def test_judge_conversation_result_preserves_validation_error(
     monkeypatch.setattr("app.features.evaluation.judge_service.allm_json_call", _boom)
 
     svc = EvalJudgeService(object())
-    result = await svc.judge_conversation_result(full_transcript="[Turn 1]\n用户: hi\nAI: hello")
+    result = await svc.judge_conversation_result(
+        full_transcript="[Turn 1]\n用户: hi\nAI: hello"
+    )
 
     assert result.output is None
     assert result.error is not None
@@ -91,7 +96,32 @@ def test_eval_judge_transcript_budget_exceeds_legacy_8192(
     n = eval_judge_conversation_transcript_max_chars()
     assert n > 90_000
     each = eval_judge_compare_transcript_each_max_chars()
-    assert each > 40_000
+    assert each > 85_000
+
+
+def test_conversation_judge_excerpt_appends_boundary_when_clipped() -> None:
+    body = conversation_judge_transcript_excerpt("abcdefghijklmnopqrstuvwxyz", cap=10)
+    assert body.startswith("abcdefghij")
+    assert "评审边界" in body
+    assert "截断稿" in body
+
+
+def test_turn_judge_prior_excerpt_appends_boundary_when_clipped() -> None:
+    body = turn_judge_prior_excerpt("0123456789abcdef", cap=8)
+    assert body.startswith("01234567")
+    assert "上文节选已截断" in body
+
+
+def test_trim_compare_transcript_pair_prefers_trimming_longer_side() -> None:
+    b_out, r_out, b_trunc, r_trunc = trim_compare_transcript_pair(
+        "a" * 100,
+        "b" * 900,
+        total_max_chars=950,
+        per_side_max_chars=None,
+    )
+    assert len(b_out) == 100
+    assert len(r_out) == 850
+    assert b_trunc is False and r_trunc is True
 
 
 def test_eval_judge_transcript_budget_respects_explicit_cap(
@@ -118,9 +148,7 @@ def test_build_memoir_prompt_includes_source_and_reference_evidence() -> None:
 
 
 def test_build_memoir_prompt_requires_conservative_scoring_without_evidence() -> None:
-    prompt = _build_memoir_judge_prompt(
-        memoir_markdown="# 当前正文\n他后来去了南方。"
-    )
+    prompt = _build_memoir_judge_prompt(memoir_markdown="# 当前正文\n他后来去了南方。")
 
     assert "无可用局部对话证据" in prompt
     assert "必须保守打分" in prompt
@@ -156,12 +184,14 @@ def test_compare_summary_surpass_gate_and_truncation_flags() -> None:
         baseline_transcript="A" * 400,
         replay_transcript="B" * 1200,
         conv_cap=1000,
-        compare_cap_each=500,
+        compare_cap_total=1000,
+        compare_per_side_cap=None,
         fixture_filename="golden.md",
     )
     assert summary["mode"] == "ab"
     assert summary["gate"]["status"] in {"parity", "surpass"}
     assert summary["truncation"]["replay_truncated_for_compare"] is True
+    assert summary["evidence_quality"]["scope"] == "partial"
     assert "group_deltas" in summary
 
 
@@ -181,7 +211,8 @@ def test_compare_summary_flags_repeat_issue_as_regression() -> None:
         baseline_transcript="[Turn 1]",
         replay_transcript="[Turn 1]",
         conv_cap=1000,
-        compare_cap_each=500,
+        compare_cap_total=1000,
+        compare_per_side_cap=None,
     )
     assert summary["repeat_issue_detected"] is True
     assert summary["gate"]["status"] == "regressed"