qimen-dunjia-engine
Qimen Dunjia (奇门遁甲) chart calculation engine for Node.js — deterministic, no AI/LLM dependency, pure calculation.
时家转盘法 · 置闰符头定元 · 值使独立转门
实测数据(每一条都可复现,脚本在 verify/):44,064 次全量遍历崩溃 0、不变式违反 0;41 种格局零命中盘 0%、无恒真规则、无死规则;拆补 vs 置闰分歧率 41.6%;两批书载案例审计(142 例与 79 例,各自独立口径,不可混算)。
Why this engine
Most open-source Qimen libraries cut corners. This one implements the classical mainstream ruleset faithfully:
Features
- Hour-based rotating-plate method (时家转盘), Chabu (拆补), Zhirun (置闰) and Shen Jie Qi (超神接气) determination
- Full plate: earth plate (地盘), heaven plate (天盘), 9 stars (九星), 8 doors (八门), 8 gods (八神), hidden stems (遁干)
- Formation detection (格局): 41 formations detected in real computation — Fu-Yin (伏吟), Fan-Yin (反吟), 门迫, 击刑, 入墓, 青龙返首, 飞鸟跌穴, 玉女守门, 三奇得使, 十干克应 combinations, and more
- Horse palace (马星), void branches (旬空), year/month stems lodging
- Four-pillar (干支) calculation for year/month/day/hour via lunar-javascript
- Zero network, zero AI — same numbers every run
Install
Usage
See examples/basic.js for a runnable demo that prints a full nine-palace grid.
Output reference
Scope
This package computes charts only. Interpretation, judgment rules (用神/断法), and case libraries are not included — those are product-layer assets of QiMenSeek.
The engine is流派-aware in one dimension: it implements the mainstream hour-based rotating-plate tradition (时家转盘). It does not compute flying-plate (飞盘) charts.
Test & Verification
That only covers three sample charts. For real numbers, run the measurement suite —
every figure quoted in this README comes from these scripts, and you will get the
same output on your own machine:
See verify/README.md for what each script proves, what it does
not prove, and the measurement traps we hit while writing them.
Known limitation we do not hide
The 142-case classical library (verify/book_case_library_tagged.json, 95.8% with page
citations) reproduces the book-stated ju number 64.9% of the time under Chaibu
(63/97 testable) vs 42.6% under Zhirun (40/94). Only 51 of 142 cases carry an
expected verdict, and 45 of those are OCR-extracted and still flagged needsReview.
So: use the library for regression exploration, not as an accuracy claim.
The second, independent library (verify/book499_case_library.json, 79 testable cases
from a 1990s casebook) reproduces the book-stated ju 82.3% under Chaibu (65/79) and
91.1% under the Shen Jie Qi mode (72/79). Note what that mode is: it is fitted from
this casebook's own data — it explains this book, it does not claim to be "more
classical". The default remains Chaibu. Five boundary cases stay unexplained, listed by
the script itself. The two libraries never mix: different books, different denominators.
Live tools
End-user guides & articles
For people facing real decisions — not just developers — these plain-language guides explain the
system (no mysticism, full reasoning, every claim sourced to your chart's symbols) and link back to
the free tools:
Index: articles/README.md
License
MIT