Ziwei Doushu open-source engine releases 56.6 GB dataset of 675k horoscopes

Ziwei Doushu is an open-source engine for Ziwei Doushu (Purple Star Astrology), a traditional Chinese astrology system. The project released version 4.0 of its interpretation dataset on October 15, containing 674,976 horoscopes covering every real calendar date from 1950 to 2026 across 12 time periods and both genders. Each horoscope includes 13-dimensional interpretations totaling approximately 8.8 million segments and 20.4 billion Chinese characters. The engine code is under MIT license; the dataset uses CC BY 4.0. The project also added regression tests and CI via GitHub Actions, expanded the engine from 181 to 641 lines in core algorithm code plus 8 new modules, and corrected star brightness defaults to match the classical Ziwei Doushu Quanshu text.

Engine and dataset details

• Dataset scale: 674,976 horoscopes × 13 dimensions = 8,774,688 interpretation segments; 56.6 GB uncompressed, 161 MB compressed with zstd in 77 year-segmented files
• Coverage: Gregorian dates 1950–2026, all real month/day combinations, 12 traditional Chinese hours (时辰), male/female
• Engine expansion: Core algorithm.ts grew from 181 to 641 lines; 8 new TypeScript modules added (1,367 lines) for true solar time correction, Chinese DST 1986–1991, lunar/solar conversion, school configuration for star brightness variants, twelve spirits, and fate-period context
• Quality controls: 5 regression tests added via npm test, running automatically on GitHub Actions for pushes and PRs; 5 golden-standard horoscopes verified palace-by-palace for star placement and brightness
• Corrections in v4.0: Removed unverifiable citations, deleted impossible four-transformation (四化) combinations, added conditional qualifiers to ambiguous statements, revised health organ mapping to follow Ni Hai Xia's method of kill-star concentration by earthly branch rather than five-element star theory
• Licensing: Code MIT; dataset CC BY 4.0; classical texts public domain; Ni Hai Xia's original lecture quotes under CC BY-NC-SA 4.0

Why it matters

The project represents a rare example of open-sourcing traditional cultural knowledge with structured data and verifiable algorithms, making it reproducible for researchers and developers. The large deterministic dataset could serve as training or evaluation material for natural language generation in Chinese, or for studying systematic biases in traditional interpretive systems.