Refactor AI daily report pipeline

ai_daily_report/guide.py

@@ -0,0 +1,113 @@
+from __future__ import annotations
+
+import json
+import re
+from typing import Any, Callable
+
+from .llm import parse_json_object
+from .models import NewsItem
+
+
+GuideLlmCall = Callable[[str], str]
+
+
+def _clean_text(text: str, limit: int | None = None) -> str:
+    value = re.sub(r"^\s*>\s*", "", text or "").strip()
+    value = re.sub(r"\[\d+\]|\[N\]", "", value)
+    value = re.sub(r"\s+", " ", value).strip()
+    if limit and len(value) > limit:
+        value = value[:limit].rstrip()
+    return value
+
+
+def _build_prompt(items: list[NewsItem]) -> str:
+    payload = {
+        "task": (
+            "Generate a concise AI daily report guide. Return JSON only. Do not use 强信号/中信号/待验证. "
+            "Use a short theme and 2-4 daily threads. Every thread must reference existing item_ids."
+        ),
+        "items": [
+            {
+                "id": item.id,
+                "title": item.title or item.title_raw,
+                "summary": item.summary or item.summary_raw,
+                "section": item.section,
+                "source": item.source_label,
+            }
+            for item in items
+        ],
+        "output_schema": {
+            "theme": "one sentence under 120 Chinese characters",
+            "threads": [
+                {
+                    "title": "thread title",
+                    "text": "one or two sentences",
+                    "item_ids": ["existing item id"],
+                    "kind": "thread|uncertain",
+                }
+            ],
+        },
+    }
+    return json.dumps(payload, ensure_ascii=False)
+
+
+def generate_guide(
+    items: list[NewsItem],
+    *,
+    llm_call: GuideLlmCall,
+) -> tuple[dict[str, Any], dict[str, Any]]:
+    if not items:
+        return {
+            "theme": "",
+            "threads": [],
+        }, {
+            "input_count": 0,
+            "theme_present": False,
+            "thread_count": 0,
+            "dropped_thread_count": 0,
+            "fallback_used": False,
+            "errors": [],
+        }
+
+    try:
+        obj = parse_json_object(llm_call(_build_prompt(items)))
+    except Exception as exc:
+        return {
+            "theme": "",
+            "threads": [],
+        }, {
+            "input_count": len(items),
+            "theme_present": False,
+            "thread_count": 0,
+            "dropped_thread_count": 0,
+            "fallback_used": True,
+            "errors": [f"{type(exc).__name__}: {exc}"],
+        }
+
+    valid_ids = {item.id for item in items}
+    threads: list[dict[str, Any]] = []
+    dropped = 0
+    for thread in obj.get("threads", []) or []:
+        item_ids = [item_id for item_id in thread.get("item_ids", []) if item_id in valid_ids]
+        if not item_ids:
+            dropped += 1
+            continue
+        title = _clean_text(str(thread.get("title") or ""), limit=80)
+        text = _clean_text(str(thread.get("text") or ""), limit=220)
+        if not title or not text:
+            dropped += 1
+            continue
+        kind = thread.get("kind") if thread.get("kind") in ("thread", "uncertain") else "thread"
+        threads.append({"title": title, "text": text, "item_ids": item_ids, "kind": kind})
+
+    theme = _clean_text(str(obj.get("theme") or ""), limit=120)
+    guide = {"theme": theme, "threads": threads}
+    report = {
+        "input_count": len(items),
+        "theme_present": bool(theme),
+        "thread_count": len(threads),
+        "dropped_thread_count": dropped,
+        "fallback_used": False,
+        "errors": [],
+    }
+    return guide, report