Research recipes: choose an entry point by question

Prepare the corresponding data first, then run the queries. The basic demo contains only a small range of market data; it cannot be used directly for financial statement, full-market universe or derivatives research.

What you want to verify	Recipe	Prerequisites
The difference between unadjusted and hfq returns	Price adjustment research baseline	demo --research; requires TDX and Sina
The financial statement version visible on a rebalance date	PIT financial statement cross-section	A production lake with financial statements and contemporaneous observation evidence; a fresh backfill may yield no strict results
SQL aggregation and Polars feature processing	DuckDB and Polars	The relevant data has been compacted; the price adjustment example also needs factors
Term structure, option chains and Greeks	Commodity futures and options	Explicitly enabled, backfilled per exchange and verified
Computing 15 / 30 / 60-minute bars on the fly and storing them in the lake	15 / 30 / 60-minute bars	1m or 5m enabled and backfilled

Conventions shared by all recipes

1. Use the configuration each page specifies. Across working directories or with MCP, use an absolute config path and an absolute data.root.
2. Price adjustment research uses strict_adj=True; PIT uses an explicit pit_mode="strict"; for universes, choose a versioned profile or strict_universe=True.
3. The coverage bounds from list_datasets() are a starting point for finding data; they do not prove there are no gaps inside the window. Judge with the actual dates, quality reports and research gates.
4. Pin dependency revisions, query parameters, software version and the necessary evidence. Use research snapshots for long-term retention; the recipes are not promises of backtest returns or investment performance.

For API details, see the query guide and the Python API.