Accuracy
What is checked, and what is not
We do not publish an accuracy percentage. Saju is an interpretive tradition, and a percentage would be a claim we cannot support. What we can tell you is exactly which parts of the calculation are tested and which are not.
Checked
- The same birth details always produce the same chart. The calculation is deterministic and the result carries the version it was produced under.
- The difference between universal time and terrestrial time is applied, so solar term instants are not shifted by the roughly one minute that correction is worth.
- Day pillars were checked against 475 published reference dates spanning 1981-02-15 to 2058-12-28. Every one matched, and the days either side of each did not.
- The chart does not change with the time zone of the machine it is calculated on.
- Historical time zone offsets and daylight saving changes are applied from the standard database, including skipped and repeated hours.
- Month boundaries switch at the calculated solar term instant, checked one second before, at, and one second after each of the twelve terms.
- Longitude and equation-of-time corrections are checked against worked examples.
- Every sentence in a paid report is checked against the chart it claims to come from before delivery.
Not checked yet
- The absolute instant of each solar term can be off by up to 13 minutes, measured against published times.
- The school convention behind the day boundary has not yet been confirmed by a practitioner, so charts for births between 23:00 and midnight may be read differently elsewhere.
- Korean lunar calendar input is not supported.
- Luck cycles are calculated, but the ten-year periods have not yet been compared against an independent published reference.
- Hidden stems are switched off. The table is built and checked against Korean and English-language sources, but every chart on this site is calculated without it, which makes our element counts and our root counting more conservative than a reading that uses it.
This list shrinks as work lands. It is not marketing copy; it is the current state.
How far off the solar terms are
We compared our calculated solar term instants against published times across 1959–2050 — 59 terms in total.
| Largest difference | 13 minutes |
|---|---|
| Average difference | -3 minutes (we run early) |
| Measured on | 2026-10-02 |
This is a measurement, not an estimate. It is also the reason a birth moment that falls within this window of a solar term is marked unresolved rather than assigned to a month pillar.
The comparison spans four different historical offsets for Korea, including the half hour it kept until 1961 and the summer time of 1960 and 1987. Our chart matched the published instants under all of them, which is the part most calculators get wrong for older births.
Bringing this below a minute needs a high precision ephemeris. We would rather publish the number we have than round it out of sight.
If your chart differs from another site
Compare the rules first. A different day boundary, a different time standard, or a different reference day will change the day and hour pillars without either calculation being wrong. Your result page shows every value that went into the calculation, so the difference can be located.