The temptation

A dense Chinese A4 PDF framed around a $200K FI portfolio looks like a new product. New domain. New worker. New Resend project. Maybe a cute investment-report.xingai.app.

ADR-032 already rejected that path for Daily Stock Market Intelligence. ADR-040 applies the same rule to 每日投资智报.

Ship it inside Invest AI. Keep Premium Daily Brief, pre-market / after-close PDFs, and ADR-032 HTML reports alone.

What shipped

Package path:

stock-ai-worker/market_cache_worker/investment_zhibao/
config/investment_zhibao/{user.yaml,portfolio.csv,policy.yaml}

CLI:

python -m market_cache_worker.investment_zhibao validate-data|generate|send|run-daily

Catalog entry: daily-investment-zhibao (ADR-025 report types).

Live mode reads Yahoo bars in batch, stamps market_as_of from real bar dates, and prints day-change %. CAPE / Forward P/E stay N/A until we have an honest source. Missing data stays blank. Fake precision is a product bug.

Email reuses existing Resend env keys (RESEND_API_KEY / INVEST_AI_RESEND_API_KEY). Dry-run is the default path until secrets exist in the environment.

Boundaries that matter

Surface	Owns
Worker package	Snapshot, calc, risk buckets, reco, PDF, email
FastAPI	Report type metadata only — no recomputation
XNP	Future transport only — not in the hot path today

Same decision cache boundary as Premium Daily Brief: worker owns semantics; delivery formats ride on that payload.

What is still open

• Scheduler / cron hook for run-daily
• Live CAPE / fuller fundamentals without inventing numbers
• PDF denser toward the ~8-page reference layout
• One live send once Resend keys are on the worker host

Takeaway

Matching a reference PDF is a packaging problem, not a greenfield product. Extending Invest AI kept Resend, the reports catalog, and the worker/cache rule in one place.

Further reading: ADR-040 · system design article in xingai-enterprise-ai-design.