Fix report export routing and research sanitization

Author: yjw

src/young_stock/artifacts.py

@@ -15,11 +15,11 @@
 def market_session(now: datetime | None = None) -> str:
     now = now or datetime.now()
     minute = now.hour * 60 + now.minute
-    if 9 * 60 <= minute < 9 * 60 + 30:
+    if 9 * 60 <= minute < 11 * 60 + 30:
         return "早盘"
     if 11 * 60 + 30 <= minute < 13 * 60:
         return "午间"
-    if 9 * 60 + 30 <= minute < 11 * 60 + 30 or 13 * 60 <= minute < 15 * 60:
+    if 13 * 60 <= minute < 15 * 60:
         return "盘中"
     return "盘后"
 

src/young_stock/cli.py

@@ -78,6 +78,10 @@ def _print_first_use_guide() -> None:
     click.echo(f"配置会保存到: {profile_path()}")
 
 
+def _current_report_date() -> str:
+    return datetime.now().strftime("%Y%m%d")
+
+
 @cli.command(help="A-share dashboard: indices, ZT/DT pool, verified A-share fund flow, boards.")
 @_date_opt
 @_refresh_opt
@@ -313,7 +317,7 @@ def _run_llm_replay(date_str: str, kind: str = "replay", symbol: str | None = No
 def replay(date: str | None, refresh: bool) -> None:
     if refresh:
         _core.NO_CACHE = True
-    _run_llm_replay(date or _core.nearest_trade_date())
+    _run_llm_replay(date or _current_report_date())
 
 
 @cli.command(help="Generate deep analysis for one stock using verified young-stock data.")
@@ -520,7 +524,7 @@ def report(date: str | None) -> None:
 
     try:
         markdown_path, pdf_path = export_report_pdf(
-            date or _core.nearest_trade_date(),
+            date or _current_report_date(),
             core=_core,
             profile=load_profile(),
         )

src/young_stock/pdf.py

@@ -3,8 +3,10 @@
 from __future__ import annotations
 
 import contextlib
+from datetime import datetime
 import html
 import io
+import json
 import os
 import re
 import sys
@@ -15,6 +17,7 @@
 
 from .artifacts import ReportArtifacts, ReportIdentity, market_session
 from .local_store import load_store
+from .research_style import sanitize_public_report
 
 
 class PDFDependencyError(RuntimeError):
@@ -94,6 +97,63 @@ def _template_text() -> str:
     )
 
 
+def _parse_identity_from_path(path: Path) -> ReportIdentity | None:
+    match = re.match(r"^(?P<trade_date>\d{8})-(?P<session>早盘|午间|盘中|盘后)-(?P<topic>.+)\.md$", path.name)
+    if not match:
+        return None
+    return ReportIdentity(match.group("trade_date"), match.group("session"), match.group("topic"))
+
+
+def _report_title(markdown: str, trade_date: str) -> str:
+    title_match = re.search(r"^#\s+(.+)$", markdown, flags=re.MULTILINE)
+    return title_match.group(1).strip() if title_match else f"{trade_date} 投资复盘报告"
+
+
+def _report_topic(markdown: str, fallback: str = "A股投资日报") -> str:
+    title = re.sub(r"\s+", "", _report_title(markdown, "")).strip("：:-—")
+    if title in {"", "复盘", "深度复盘"}:
+        return "A股深度复盘"
+    if "日报" in title:
+        return "A股投资日报"
+    return title or fallback
+
+
+def _load_related_evidence(markdown_path: Path) -> dict[str, Any] | None:
+    candidates = []
+    if markdown_path.stem in {"replay", "report"}:
+        candidates.append(markdown_path.with_name("evidence.json"))
+    candidates.append(markdown_path.with_name(f"{markdown_path.stem}-evidence.json"))
+    for candidate in candidates:
+        if not candidate.exists():
+            continue
+        try:
+            return json.loads(candidate.read_text(encoding="utf-8"))
+        except json.JSONDecodeError:
+            return None
+    return None
+
+
+def _canonicalize_markdown(
+    artifacts: ReportArtifacts,
+    trade_date: str,
+    markdown_path: Path,
+    markdown: str,
+    *,
+    now: datetime | None = None,
+) -> tuple[Path, str]:
+    identity = _parse_identity_from_path(markdown_path)
+    evidence = _load_related_evidence(markdown_path)
+    cleaned = sanitize_public_report(markdown, evidence)
+    session = identity.session if identity else market_session(now)
+    topic = identity.topic if identity else _report_topic(cleaned)
+    canonical = artifacts.write_report_markdown(ReportIdentity(trade_date, session, topic), cleaned)
+    return canonical, cleaned
+
+
+def _clean_document_html(document: str) -> str:
+    return re.sub(r"Kami-compatible editorial layout\s*·?\s*", "", document, flags=re.IGNORECASE)
+
+
 def _load_weasyprint() -> Any:
     os.environ.setdefault("XDG_CACHE_HOME", str(Path(tempfile.gettempdir()) / "young-stock-weasy-cache"))
     if sys.platform == "darwin":
@@ -116,9 +176,9 @@ def _default_render(html_path: Path, pdf_path: Path) -> None:
     renderer = _load_weasyprint()
     if renderer is None:
         raise PDFDependencyError(
-            "未安装 PDF 可选依赖。uv tool 用户请运行 "
-            "`uv tool install --force 'young-stock-cli[pdf]'`；"
-            "普通 Python 环境请运行 `python3 -m pip install \"young-stock-cli[pdf]\"`。"
+            "当前环境未检测到 PDF 渲染能力。请先运行 `young init` 检查安装状态；"
+            "若仍缺少依赖，uv tool 用户请执行 `uv tool install --force 'young-stock-cli[pdf]'`，"
+            "普通 Python 环境请执行 `python3 -m pip install \"young-stock-cli[pdf]\"`。"
         )
     renderer(filename=str(html_path), base_url=str(html_path.parent)).write_pdf(str(pdf_path))
 
@@ -145,6 +205,7 @@ def export_report_pdf(
     profile: dict[str, Any] | None = None,
     daily_markdown_factory: Callable[[], str] | None = None,
     render: Callable[[Path, Path], None] | None = None,
+    now: datetime | None = None,
 ) -> tuple[Path, Path]:
     artifacts = ReportArtifacts(trade_date)
     markdown_path = ReportArtifacts.latest_markdown(trade_date)
@@ -159,13 +220,15 @@ def export_report_pdf(
             markdown = _capture_daily(core, trade_date, profile)
         else:
             raise ValueError("没有可用报告；请先运行 `young daily` 或提供日报生成器。")
-        identity = ReportIdentity(trade_date, market_session(), "A股投资日报")
-        markdown_path = artifacts.write_report_markdown(identity, markdown)
+        markdown_path = artifacts.write_report_markdown(
+            ReportIdentity(trade_date, market_session(now), "A股投资日报"),
+            sanitize_public_report(markdown),
+        )
     markdown = markdown_path.read_text(encoding="utf-8")
+    markdown_path, markdown = _canonicalize_markdown(artifacts, trade_date, markdown_path, markdown, now=now)
     body = markdown_to_html(markdown)
-    title_match = re.search(r"^#\s+(.+)$", markdown, flags=re.MULTILINE)
-    title = title_match.group(1).strip() if title_match else f"{trade_date} 投资复盘报告"
-    document = (
+    title = _report_title(markdown, trade_date)
+    document = _clean_document_html(
         _template_text()
         .replace("{{TITLE}}", html.escape(title))
         .replace("{{DATE}}", trade_date)

src/young_stock/reports.py

@@ -22,7 +22,7 @@
 )
 
 LLM_REPORT_PROMPT_VERSION = "stock-analysis-research-language-v2"
-LLM_REPORT_SYSTEM_PROMPT = """你是一名资深A股交易员、量化研究员和风险控制专家。
+LLM_REPORT_SYSTEM_PROMPT = """请基于用户提供的研报证据，撰写正式 A 股投资研究报告。
 你只能使用用户提供的研报证据，不得补写或外推任何缺失数字、日期、来源或持仓。
 按以下顺序输出 Markdown：大盘指数概览、持仓分析、六模块深度复盘、综合持仓建议与风险提示。
 每个有证据的模块给出关键判断、证据、风险/确认条件；建议必须是条件化触发器，不给无条件买卖指令。

src/young_stock/research_style.py

@@ -58,6 +58,11 @@
     r"\bM[1-6]\s*(?:降级|缺失|不可用)",
 )
 
+EDITORIAL_NOISE_PATTERNS = (
+    r"^\s*(?:好的[，,、]?)?.*资深A股交易员.*$",
+    r"Kami-compatible editorial layout",
+)
+
 
 class ResearchStyleError(RuntimeError):
     """Report still contains publication-unsafe internal language."""
@@ -154,23 +159,49 @@ def _unsafe(line: str) -> bool:
     return any(re.search(pattern, line, flags=re.IGNORECASE) for pattern in FORBIDDEN_PATTERNS)
 
 
-def review_research_report(markdown: str, evidence: dict[str, Any]) -> str:
-    """Remove unsafe sentences, add evidence-backed research wording, and validate."""
-    internal_fields = re.findall(r"\bmodules\.([A-Za-z0-9_]+)\.([A-Za-z0-9_.]+)", markdown)
-    known_modules = set((evidence.get("modules") or {}).keys())
-    if any(module not in known_modules for module, _field in internal_fields):
-        raise ResearchStyleError("正式研报包含无法转换的内部字段。")
+def _editorial_noise(line: str) -> bool:
+    return any(re.search(pattern, line, flags=re.IGNORECASE) for pattern in EDITORIAL_NOISE_PATTERNS)
+
+
+def _normalized_line(line: str) -> str:
+    return line.replace("`", "")
+
+
+def sanitize_public_report(markdown: str, evidence: dict[str, Any] | None = None, *, strict: bool = False) -> str:
+    evidence = evidence or {}
+    normalized_markdown = _normalized_line(markdown)
+    if strict:
+        internal_fields = re.findall(r"\bmodules\.([A-Za-z0-9_]+)\.([A-Za-z0-9_.]+)", normalized_markdown)
+        known_modules = set((evidence.get("modules") or {}).keys())
+        if any(module not in known_modules for module, _field in internal_fields):
+            raise ResearchStyleError("正式研报包含无法转换的内部字段。")
+
     safe_lines = []
-    removed = False
+    removed_unsafe = False
+    in_code_block = False
     for line in markdown.splitlines():
-        if _unsafe(line):
-            removed = True
+        stripped = line.strip()
+        if stripped.startswith("```"):
+            in_code_block = not in_code_block
+            removed_unsafe = True
+            continue
+        normalized = _normalized_line(line)
+        if in_code_block or _editorial_noise(normalized):
+            continue
+        if _unsafe(normalized):
+            removed_unsafe = True
             continue
         safe_lines.append(line)
-    if removed:
+    if removed_unsafe:
         insert_at = 1 if safe_lines and safe_lines[0].startswith("#") else 0
         safe_lines[insert_at:insert_at] = ["", *_replacement_lines(evidence), ""]
     reviewed = "\n".join(safe_lines).strip()
-    if _unsafe(reviewed):
+    validation_lines = [_normalized_line(line) for line in reviewed.splitlines()]
+    if any(_unsafe(line) or _editorial_noise(line) for line in validation_lines):
         raise ResearchStyleError("正式研报未通过研究语言审校。")
     return reviewed
+
+
+def review_research_report(markdown: str, evidence: dict[str, Any]) -> str:
+    """Remove unsafe sentences, add evidence-backed research wording, and validate."""
+    return sanitize_public_report(markdown, evidence, strict=True)

tests/test_artifacts.py

@@ -47,8 +47,10 @@ def test_generate_llm_daily_report_uses_evidence_and_returns_metadata():
 
 def test_market_session_labels():
     assert market_session(datetime(2026, 6, 18, 9, 10)) == "早盘"
-    assert market_session(datetime(2026, 6, 18, 10, 30)) == "盘中"
+    assert market_session(datetime(2026, 6, 18, 9, 41)) == "早盘"
+    assert market_session(datetime(2026, 6, 18, 10, 30)) == "早盘"
     assert market_session(datetime(2026, 6, 18, 12, 0)) == "午间"
+    assert market_session(datetime(2026, 6, 18, 14, 10)) == "盘中"
     assert market_session(datetime(2026, 6, 18, 15, 10)) == "盘后"
 
 
@@ -70,3 +72,14 @@ def test_same_session_topic_overwrites_and_cross_session_is_retained(monkeypatch
     assert first == second
     assert second.read_text(encoding="utf-8") == "# new\n"
     assert other.exists()
+
+
+def test_latest_markdown_prefers_identity_path_over_newer_legacy_file(monkeypatch, tmp_path):
+    monkeypatch.setenv("YOUNG_STOCK_HOME", str(tmp_path))
+    artifacts = ReportArtifacts("20260619")
+    legacy = artifacts.write_markdown("replay", "# legacy\n")
+    identified = artifacts.write_report_markdown(ReportIdentity("20260619", "早盘", "A股深度复盘"), "# identity\n")
+
+    legacy.touch()
+
+    assert ReportArtifacts.latest_markdown("20260619") == identified

tests/test_llm_reports.py

@@ -74,6 +74,19 @@ def test_llm_report_output_removes_engineering_language():
     assert "备用路径" not in markdown
 
 
+def test_llm_report_output_strips_fixed_preamble():
+    client = RecordingClient(
+        "# 复盘\n\n"
+        "好的，作为资深A股交易员，以下是今天的复盘。\n"
+        "据公开市场数据，板块轮动加快。\n"
+    )
+
+    markdown, _ = generate_llm_daily_report({"modules": {}, "_meta": {}}, client)
+
+    assert "资深A股交易员" not in markdown
+    assert "板块轮动加快" in markdown
+
+
 def test_llm_methodology_context_is_research_only():
     client = RecordingClient("# 复盘\n\n据公开市场数据，市场震荡。")
 
@@ -84,6 +97,7 @@ def test_llm_methodology_context_is_research_only():
     )
 
     system_text = "\n".join(message["content"] for message in client.messages if message["role"] == "system")
+    assert "资深A股交易员" not in system_text
     assert "报告固定顺序" in system_text
     assert "fallback" not in system_text
     assert "脚本" not in system_text

tests/test_pdf.py

@@ -24,9 +24,10 @@ def fake_render(html_path: Path, pdf_path: Path):
 
     markdown_path, pdf_path = export_report_pdf("20260618", render=fake_render)
 
-    assert markdown_path == source
+    assert markdown_path != source
+    assert markdown_path.name.startswith("20260618-")
     assert pdf_path.read_bytes().startswith(b"%PDF")
-    assert calls[0][0].name == "replay.html"
+    assert calls[0][0].name.startswith("20260618-")
 
 
 def test_export_report_auto_generates_daily_markdown(monkeypatch, tmp_path):
@@ -60,10 +61,27 @@ def test_export_report_reuses_saved_diary_text(monkeypatch, tmp_path):
     assert "Diary report" in markdown_path.read_text()
 
 
+def test_export_report_strips_layout_noise_and_fixed_preamble(monkeypatch, tmp_path):
+    monkeypatch.setenv("YOUNG_STOCK_HOME", str(tmp_path))
+    ReportArtifacts("20260618").write_markdown(
+        "replay",
+        "# 复盘\n\n好的，作为资深A股交易员，以下是今天的报告。\nKami-compatible editorial layout · 内容仅供复盘参考\n正文\n",
+    )
+
+    markdown_path, pdf_path = export_report_pdf(
+        "20260618",
+        render=lambda html, pdf: pdf.write_bytes(b"%PDF-clean"),
+    )
+
+    assert "资深A股交易员" not in markdown_path.read_text(encoding="utf-8")
+    html_text = pdf_path.with_suffix(".html").read_text(encoding="utf-8")
+    assert "Kami-compatible editorial layout" not in html_text
+
+
 def test_default_renderer_has_clear_optional_dependency_error(monkeypatch, tmp_path):
     monkeypatch.setenv("YOUNG_STOCK_HOME", str(tmp_path))
     ReportArtifacts("20260618").write_markdown("daily", "# 日报")
     monkeypatch.setattr("young_stock.pdf._load_weasyprint", lambda: None)
 
-    with pytest.raises(PDFDependencyError, match=r"uv tool install --force"):
+    with pytest.raises(PDFDependencyError, match=r"young init"):
         export_report_pdf("20260618")

tests/test_report_routing.py

@@ -0,0 +1,28 @@
+from pathlib import Path
+
+from click.testing import CliRunner
+
+import young_stock.cli as cli_module
+from young_stock.cli import cli
+
+
+def test_replay_and_report_default_to_current_calendar_date(monkeypatch, tmp_path):
+    monkeypatch.setenv("YOUNG_STOCK_HOME", str(tmp_path))
+    monkeypatch.setattr(cli_module._core, "nearest_trade_date", lambda: "20260618")
+    monkeypatch.setattr(cli_module, "_current_report_date", lambda: "20260619")
+
+    replay_calls = []
+    monkeypatch.setattr(cli_module, "_run_llm_replay", lambda date_str, kind="replay", symbol=None: replay_calls.append((date_str, kind, symbol)))
+    monkeypatch.setattr(
+        "young_stock.pdf.export_report_pdf",
+        lambda trade_date, **kwargs: (Path(f"/tmp/{trade_date}.md"), Path(f"/tmp/{trade_date}.pdf")),
+    )
+
+    runner = CliRunner()
+    replay_result = runner.invoke(cli, ["replay"])
+    report_result = runner.invoke(cli, ["report"])
+
+    assert replay_result.exit_code == 0
+    assert report_result.exit_code == 0
+    assert replay_calls == [("20260619", "replay", None)]
+    assert "/tmp/20260619.pdf" in report_result.output

tests/test_research_style.py

@@ -73,6 +73,21 @@ def test_review_research_report_handles_markdown_wrapped_engineering_phrases():
     assert "科创板与创业板活跃样本数为 14 家" in reviewed
 
 
+def test_review_research_report_strips_fixed_preamble_and_layout_noise():
+    markdown = (
+        "# 复盘\n\n"
+        "好的，作为资深A股交易员，以下是今天的正式报告。\n"
+        "据公开市场数据，市场震荡整理。\n"
+        "Kami-compatible editorial layout · 内容仅供复盘参考\n"
+    )
+
+    reviewed = review_research_report(markdown, sample_evidence())
+
+    assert "资深A股交易员" not in reviewed
+    assert "Kami-compatible editorial layout" not in reviewed
+    assert "市场震荡整理" in reviewed
+
+
 def test_review_research_report_rejects_unrecognized_internal_field():
     with pytest.raises(ResearchStyleError):
         review_research_report("# 复盘\n\n`modules.UNKNOWN.secret` 为 true。", sample_evidence())