Guide · English · US & Europe first

Why ChatGPT gets BaZi wrong

ChatGPT and similar models predict text — they do not run a fixed BaZi engine. Ask twice and you can get two different charts. Day Master, pillars, and day-boundary rules require deterministic calendar math. Flow+ computes those rules in code and publishes the methodology so the same inputs always produce the same chart.

The failure mode is structural

Large language models sample plausible sentences. They are excellent at explaining concepts and terrible at guaranteeing arithmetic across solar terms, timezones, and day boundaries.

BaZi is unforgiving: a wrong day boundary flips the day stem. That is not a “prompt engineering” problem; it is the wrong class of tool.

What “deterministic” means here

Flow+ engines take birth data in and return the same pillars every time for the same method. Master Xie (our AI mentor) explains what the engines already computed — it does not invent a chart.

We publish methodology details so practitioners and skeptics can audit the choices (including Early Zi day boundary).

A quick test you can run

Ask any general LLM for your Four Pillars twice with identical birth data. If the stems change, you have your answer. Then run the same data through Flow+’s free Day Master tool and compare consistency.

Who this is for

Builders, skeptics, and curious English-speaking users in the US and Europe who want self-insight without black-box mysticism — and without an AI that hallucinates stems.

FAQ

Can I use ChatGPT to explain my Flow+ chart?
You can paste results for discussion, but computation should stay in a deterministic engine. Flow+ separates calculation from explanation for that reason.
Does Flow+ use AI at all?
Yes — for explanation and mentoring after the chart is computed. The pillars and Day Master come from engines, not from the LLM inventing them.
Where do I verify the method?
Start at /methodology, then compute your Day Master at /try-archetype.

Compute a real Day Master

Free · no account · deterministic engines