ml8s/liki

Liki — Professional fortune-telling skill

If you understand fortune-telling, use Liki.
Layout can be reproduced · Assertions can be traced · Boundaries are clearly stated Bazi · Ziwei · Six Yao · Qimen · Date selection in the lunar calendar · Feng shui · Naming

Why Liki

Directly ask large models for fortune-telling, and if you ask the same question three times, you might get three answers. And fate theory is exactly like “a tiny error can lead to a huge error”: if the solar term boundary is off by one day, the year pillar changes, and all the subsequent judgments will be wrong.

So Liki made every “easily mistaken” step an explicit process—multi-source reconciliation of the disk, assertions from the rule truth table, time normalized before testing. **Conclusions are “calculated,” not generated by the model on the spot. **

• ✅ Disk layout can be reproduced — Same input, always the same dial surface; Solar terms and solar calculations are processed by the astronomical calendrical engine
• ✅ Assertive Statements Can Be Traced — Every conclusion is based on factors, recalling the original texts of ‘Three Lives Tonghui’ and ‘Di Tian Sui’
• ✅ Boundaries are clearly stated — If the timing is sensitive, calibrate; if evidence is insufficient, say it is insufficient; never tacit or fabricated

Three Special Projects

Here are the three most easily overlooked and most distinguishable places for Liki.

Special Feature 1 | Layout Engine: Calculated “in opposition” with three open-source implementations

The layout engine is not written all at once. 192 Solar term boundary cases (2024–2027, 24 solar terms per year ×2 anchor points per term) and three open-source implementations: lunar-python, sxtwl, bazi-calculator Four-way comparison, with only one strategy: all_oracles_must_agree—any inconsistency is not allowed. Ziwei implements data source alignment with Iztro compatibility.

Parts where the engine does not match the classic are corrected and solidified as anti-regression anchors according to the original text:

Proofreading example of ‘Qiutong Baojian’: **Yi wood faces yang, specifically taking Bing fire; Water mostly uses Wu as assistant. ** Engine output is matched verbatim with this original text, and a full comparison is made with 120 original reference tables.

Special Topic Two | Direct Judgment: 829 Real Exam Questions + Classic Fit Truth Table

The assertions are not model-generated; they are “hit” from the truth table: 829 assertions, 893 conditional groups, 1,355 conditional rows, with 511 natal factors and 111 annual factors at the bottom. A hit means a direct judgment; if not, it is stated outright—there’s no room for spontaneous rhetoric. Each statement comes with its own basis and classic source, such as this real data:

Two design details:

• The annual prediction must include current year’s trigger factor; otherwise, data verification will directly show errors—to prevent “natal constitution hitting repeatedly year after year”;
• 160 Dao Numer Competition Real Questions (Mingli160) is used for rule chain coverage and zero-hit checks, serving as a manual calibration/evaluation set.

Special Topic Three | Exam Time: Even if you have a birth certificate

The most common mistake in fortune-telling is precisely “determining the time”:

• Hospital clocks record standard time, not the sun at the moment of birth—due to different longitudes, true solar time may differ by half an hour;