Why AI Cannot Directly Perform a Reading: The Engineering Logic of Charting First, Then Interpreting

Asking a large language model directly to "cast my Zi Wei Dou Shu chart for me" will most likely yield an incorrect chart —
and one that's wrong in a way that's hard to detect: the star positions look plausible, but the entire chart is a hallucination.

Why Large Language Models Cannot Cast Charts Accurately

Zi Wei Dou Shu charting requires precise astronomical and calendrical calculations: lunar calendar conversion, solar term timings, Five-Element Bureau (五行局) numbers,
and over a hundred steps of star-placement rule derivations. These are deterministic problems for table-lookup computation, but for a language model they amount to "continuing text based on impressions." Even if the model has memorized ninety percent of the rules, the remaining ten percent is enough to place the Life Palace (命宫) in the wrong palace or assign the Si Hua (四化) to the wrong stars.

Qi Men Dun Jia (奇门遁甲) is even more demanding: Yang Dun and Yin Dun, the Chai-Bu method (拆补法) and intercalation, the rotation logic of Zhi Fu (值符, Chief) and Zhi Shi (值使, Envoy) — one wrong step and the entire chart is wrong.

The Architecture of This Site: Calculation Belongs to the Engine, Language Belongs to AI

Birth data input → Deterministic charting engine (iztro-py / lunar-python / custom Qi Men)
                → Structured chart facts (star positions across the twelve palaces / Si Hua / luck cycles)
                → AI is only responsible for translating the facts into an interpretation

Each of the three layers does its own job:

• Engine: The same input always produces the same output, cross-checked palace by palace against external charting sites, fully reproducible
• Facts layer: The results page can expand to show "Chart Facts" — what the AI sees and what you see are the same document
• AI layer: A mandatory five-stage process (facts → pattern → itemized analysis → luck cycles → advice), citing the standards of the *Gu Sui Fu* (《骨髓赋》); if the output is too short or fails, credits are automatically refunded

Why This Architecture Is Trustworthy

Placing the AI in the role of "commentator" rather than "fortune teller" is the most effective way to eliminate hallucination:
it doesn't need to "calculate correctly" — it only needs to "explain correctly" — and whether the explanation is correct, you can verify at any time against the chart facts.

This is also why we have opened the "Chart Facts" panel directly: publicly available evidence is more reliable than promises of reliability.