Reduce LLM rewrite calls and add report intro conclusion

tests/test_stage4_rewrite.py

@@ -62,7 +62,7 @@ class Stage4RewriteTests(unittest.TestCase):
         self.assertEqual(report["fallback_count"], 1)
         self.assertIn("TimeoutError", report["errors"][0])
 
-    def test_rewrite_items_retries_failed_batch_as_single_items(self):
+    def test_rewrite_items_can_retry_failed_batch_as_single_items_when_enabled(self):
         items = [news_item("a"), news_item("b")]
         calls = []
 
@@ -85,13 +85,53 @@ class Stage4RewriteTests(unittest.TestCase):
                 }
             )
 
-        rewritten, report = rewrite_items(items, llm_call=llm_call, batch_size=2)
+        rewritten, report = rewrite_items(items, llm_call=llm_call, batch_size=2, retry_single_items=True)
 
         self.assertEqual([item.title for item in rewritten], ["title a", "title b"])
         self.assertEqual(report["rewritten_count"], 2)
         self.assertEqual(report["fallback_count"], 0)
         self.assertEqual(calls, [["a", "b"], ["a"], ["b"]])
 
+    def test_rewrite_items_does_not_retry_single_items_by_default(self):
+        items = [news_item("a"), news_item("b")]
+        calls = []
+
+        def llm_call(prompt):
+            payload = json.loads(prompt)
+            calls.append([item["id"] for item in payload["items"]])
+            return "not json"
+
+        rewritten, report = rewrite_items(items, llm_call=llm_call, batch_size=2)
+
+        self.assertEqual(calls, [["a", "b"]])
+        self.assertEqual([item.title for item in rewritten], ["OpenAI launches GPT-5 API", "OpenAI launches GPT-5 API"])
+        self.assertEqual(report["fallback_count"], 2)
+
+    def test_rewrite_items_defaults_to_large_batches_to_reduce_llm_requests(self):
+        items = [news_item(str(index)) for index in range(61)]
+        batch_sizes = []
+
+        def llm_call(prompt):
+            payload = json.loads(prompt)
+            batch_sizes.append(len(payload["items"]))
+            return json.dumps(
+                {
+                    "rewrites": [
+                        {
+                            "id": entry["id"],
+                            "title": entry["title_raw"],
+                            "summary": entry["summary_raw"],
+                            "flags": [],
+                        }
+                        for entry in payload["items"]
+                    ]
+                }
+            )
+
+        rewrite_items(items, llm_call=llm_call)
+
+        self.assertEqual(batch_sizes, [30, 30, 1])
+
     def test_rewrite_items_does_not_retry_single_items_after_transient_http_error(self):
         items = [news_item("a"), news_item("b")]
         calls = 0

tests/test_stage6_guide.py

@@ -24,7 +24,7 @@ def news_item(item_id, title, section="模型与能力"):
 
 
 class Stage6GuideTests(unittest.TestCase):
-    def test_generate_guide_returns_theme_and_valid_threads(self):
+    def test_generate_guide_returns_intro_theme_threads_and_conclusion(self):
         items = [
             news_item("a", "GPT-5 API 发布"),
             news_item("b", "Miso One 开源语音模型"),
@@ -33,6 +33,7 @@ class Stage6GuideTests(unittest.TestCase):
         def llm_call(prompt):
             return json.dumps(
                 {
+                    "intro": "今天的 AI 行业继续围绕模型能力、Agent 产品和基础设施演进展开。",
                     "theme": "模型能力继续向 API 和实时语音两端推进。",
                     "threads": [
                         {
@@ -48,13 +49,16 @@ class Stage6GuideTests(unittest.TestCase):
                             "kind": "thread",
                         },
                     ],
+                    "conclusion": "总体看，模型能力正在进入更多产品入口，生态竞争也在继续加速。",
                 },
                 ensure_ascii=False,
             )
 
         guide, report = generate_guide(items, llm_call=llm_call)
 
+        self.assertEqual(guide["intro"], "今天的 AI 行业继续围绕模型能力、Agent 产品和基础设施演进展开。")
         self.assertEqual(guide["theme"], "模型能力继续向 API 和实时语音两端推进。")
+        self.assertEqual(guide["conclusion"], "总体看，模型能力正在进入更多产品入口，生态竞争也在继续加速。")
         self.assertEqual(len(guide["threads"]), 1)
         self.assertEqual(guide["threads"][0]["item_ids"], ["a", "b"])
         self.assertEqual(report["dropped_thread_count"], 1)
@@ -67,7 +71,9 @@ class Stage6GuideTests(unittest.TestCase):
 
         guide, report = generate_guide(items, llm_call=llm_call)
 
+        self.assertEqual(guide["intro"], "")
         self.assertEqual(guide["theme"], "")
+        self.assertEqual(guide["conclusion"], "")
         self.assertEqual(guide["threads"], [])
         self.assertTrue(report["fallback_used"])
         self.assertIn("TimeoutError", report["errors"][0])

tests/test_stage7_assemble.py

@@ -23,12 +23,13 @@ def news_item(item_id, title, section):
 
 
 class Stage7AssembleTests(unittest.TestCase):
-    def test_assemble_markdown_renders_sections_and_daily_threads(self):
+    def test_assemble_markdown_renders_intro_sections_daily_threads_and_conclusion(self):
         items = [
             news_item("a", "GPT-5 API 发布", "模型与能力"),
             news_item("b", "Anthropic 提交 IPO 文件", "公司与资本"),
         ]
         guide = {
+            "intro": "今天的 AI 行业继续围绕模型、产品和资本展开。",
             "theme": "> 模型和资本两条线都在推进。[1]",
             "threads": [
                 {
@@ -38,10 +39,12 @@ class Stage7AssembleTests(unittest.TestCase):
                     "kind": "thread",
                 }
             ],
+            "conclusion": "总体看，AI 竞争继续从单点模型能力转向产品、基础设施和资本协同。",
         }
 
         md, report = assemble_markdown(items, guide)
 
+        self.assertTrue(md.startswith("## 引言\n\n> 今天的 AI 行业继续围绕模型、产品和资本展开。"))
         self.assertIn("## 导览", md)
         self.assertIn("> 模型和资本两条线都在推进。", md)
         self.assertIn("## 模型与能力", md)
@@ -49,6 +52,7 @@ class Stage7AssembleTests(unittest.TestCase):
         self.assertIn("**2. Anthropic 提交 IPO 文件**", md)
         self.assertIn("## 今日脉络", md)
         self.assertIn("- **模型能力产品化**", md)
+        self.assertTrue(md.endswith("## 总结\n\n> 总体看，AI 竞争继续从单点模型能力转向产品、基础设施和资本协同。"))
         self.assertNotIn("> >", md)
         self.assertNotIn("[1]", md)
         self.assertEqual(report["item_count"], 2)