fix(rag): recall complete course homework lists

ai_knowledge_base_v2/apps/backend/tests/test_knowledge_agent.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import asyncio
 import json
 from datetime import datetime
+from types import SimpleNamespace
 
 from sqlalchemy import create_engine, select
 from sqlalchemy.orm import Session
@@ -19,7 +20,7 @@ from app.models.knowledge import (
     KnowledgeVersion,
 )
 from app.models.chat import ChatMessage
-from app.services.knowledge_agent_service import KnowledgeAgentService
+from app.services.knowledge_agent_service import Candidate, KnowledgeAgentService
 
 
 def _database() -> Session:
@@ -187,13 +188,45 @@ def test_homework_overview_expands_practice_terms_and_section_limit():
 
     assert "功课" in terms
     assert "练习" in terms
-    assert KnowledgeAgentService._selection_limit("合一的作业是什么？") == 8
+    assert KnowledgeAgentService._candidate_limit("合一的作业是什么？") == 48
+    assert KnowledgeAgentService._selection_limit("合一的作业是什么？") == 24
+    assert KnowledgeAgentService._candidate_limit("这个练习怎么做？") == 12
     assert KnowledgeAgentService._selection_limit("这个练习怎么做？") == 4
     assert KnowledgeAgentService._title_intent_boost("十七、练习一：风铃式静心", terms) == 20.0
     assert KnowledgeAgentService._title_intent_boost("完整练习的方向", terms) == 3.0
     assert KnowledgeAgentService._title_intent_boost("课程定位", terms) == 0.0
 
 
+def test_homework_overview_keeps_all_numbered_practices_before_selection():
+    candidates = [
+        Candidate(
+            chunk=SimpleNamespace(title=f"练习{index}：课程作业", id=index),
+            section=SimpleNamespace(id=index),
+            knowledge=SimpleNamespace(id=1),
+            version=SimpleNamespace(id=1),
+            lexical_score=float(30 - index),
+        )
+        for index in range(1, 15)
+    ]
+
+    unrelated = Candidate(
+        chunk=SimpleNamespace(title="课程介绍", id=99),
+        section=SimpleNamespace(id=99),
+        knowledge=SimpleNamespace(id=1),
+        version=SimpleNamespace(id=1),
+        lexical_score=100.0,
+    )
+
+    selected = KnowledgeAgentService._select_sections(
+        [unrelated, *candidates],
+        limit=KnowledgeAgentService._selection_limit("原生里的作业内容都有什么"),
+        prefer_numbered_practice=True,
+    )
+
+    assert len(selected) == 14
+    assert all(item.chunk.title.startswith("练习") for item in selected)
+
+
 def test_complete_section_includes_child_headings_but_stops_at_next_peer():
     with _database() as db:
         knowledge = _add_published_knowledge(db, knowledge_id=1, name="合一课程")