How the chart is computed
You do not have to accept our interpretation. You should be able to reproduce our calculation. This page states every rule we apply.
The line we draw
The chart is computed. The wording is generated. These are two different operations with two different failure modes, and conflating them is how this category loses credibility.
| Layer | Produced by | Reproducible? |
|---|---|---|
| Pillars, stems, branches | Rule engine | Yes — identical every time |
| Element weighting | Rule engine | Yes — identical every time |
| Phase segmentation | Rule engine | Yes — identical every time |
| Plain-language reading | Language model | No — wording varies |
If a service cannot tell you which parts are calculated and which are generated, you cannot evaluate anything it outputs.
Time and calendar
Most disagreements between tools come from here, not from the tradition itself.
- ·Time zone — resolved from birthplace via the IANA database, at the historical date, not today's offset
- ·Daylight saving — applied if it was in effect at that place and date; many tools silently skip this
- ·Calendar — Gregorian input converted to the traditional lunisolar calendar, including leap months
- ·True solar time — optional, off by default; when on, the longitude correction is shown
- ·Day boundary — the late-night hour crossing midnight is handled explicitly and disclosed
Historical time zone data changes. A birth in a region that shifted its offset may resolve differently than it would using present-day rules. We use the historical rule and show which one was applied.
Unknown birth hour
When the hour is unknown, the hour pillar is marked undetermined. Anything derived from it is suppressed rather than estimated, and the reading says so plainly.
A guessed hour pillar produces a confident-looking chart that is simply wrong. Silence is the honest output here.
Constraints on the AI layer
The model receives a structured chart — never a free-text description of you — and operates inside fixed boundaries.
- ·No dates, deadlines or specific future events
- ·No monetary amounts or financial guidance
- ·No medical, psychological, pregnancy or mortality claims
- ·No legal advice
Language models make mistakes, including confident ones. Interpretations can be wrong, internally inconsistent, or simply unhelpful. Weigh them accordingly.
Limits we cannot remove
- ·No predictive validity — no traditional system has demonstrated the ability to forecast events under controlled conditions
- ·Interpretive divergence — schools disagree; a different lineage would read the same chart differently
- ·Input sensitivity — a one-hour error in birth time can change the hour pillar entirely
- ·Language variance — the same chart described twice will not use the same words
These are properties of the domain, not defects in our implementation. No amount of engineering removes them.