Every step is shown here — including the ones we got wrong.
Your four pillars are calculated twice, by two independent calendar engines (lunar-python and cnlunar). We only write once both agree. If they disagree, we stop and look.
Strength balance, the ten-god relations, branch interactions (clash, combination, punishment, harm) and solar terms are all computed in code under fixed rules — no AI is involved in those steps.
Why: AI is good at sentences and bad at arithmetic and lookup tables. Anything that can be computed, we don't let the model touch.
A model writes the prose. But every number, technical term and quotation in it is checked line by line against the fact pack for your chart. Anything that can't be traced back to your chart is blocked and rewritten.
There is also a short list of things we never say — next section.
One real example: an early build said "weak day master". That was a guess, not a calculation. We moved that judgement into code. What you read now has been computed.
This isn't a posture. It's written into the pipeline: text that breaks these rules cannot ship.
The Chinese and English editions are written separately — separate contracts, separate banned-word lists, separate validators, sharing one chart.
The same chart, said the way each language actually says it. Not a Chinese sentence run through a translator.
We publish misses as well as matches.
Read it and think "that isn't me" — reply and tell us. We will redo it once, or refund it in full.
This is a self-reflection aid. It is not medical, psychological or legal advice, and it does not promise any outcome.
They are used for two things: casting the chart and writing the report. Not for training. Not shared. Not sold.
We don't publish or display your report — unless you ask us to.