Chart methods and open data
AI BaZi chart calculation check
Responses archived in September 2026 · Published 7 October 2026
We sent 12 fictional birth dates to models with web search disabled and archived 60 Chinese and English responses. Comparing all four pillars with our archived engine output, the ChatGPT Chinese group matched in 7 of 12 responses and the English group in 6 of 12. Claude's Chinese and English groups and Gemini's Chinese group matched in 0 of 12 each. Responses missing pillars stay in each group’s denominator.
This checks Four Pillars calculation by the archived model versions. The reference comes from kaucim.ai; other schools may use different day-boundary and birth-hour rules. These cases exclude Zi-hour and solar-term boundaries and do not test interpretation, prediction or search citations.
Results by model and language
Model names are those reported in the archived responses. Each model-language group ran once per input; these counts are not a general accuracy estimate.
| Reported version | Language | All four match | Incomplete extraction |
|---|---|---|---|
| gpt-5.5-2026-04-23 | English | 6/12 | 0/12 |
| claude-sonnet-5 | English | 0/12 | 9/12 |
| gpt-5.5-2026-04-23 | Chinese | 7/12 | 0/12 |
| claude-sonnet-5 | Chinese | 0/12 | 3/12 |
| gemini-2.5-flash | Chinese | 0/12 | 0/12 |
Sample and comparison method
- A fixed seed of 20260923 generated 12 fictional birth dates from 1980 to 2005, all in Hong Kong. We excluded 23:00–00:59 and dates within one day of a solar-term month change.
- The reference is archived output from kaucim.ai’s lunar-typescript engine, using the two-hour branch of the recorded clock time without solar-time correction. Pillars are ordered year, month, day, hour.
- Responses came through the DataForSEO LLM Responses interface with web search disabled. Chinese groups cover ChatGPT, Claude and Gemini; English groups cover ChatGPT and Claude. This dataset contains no English Gemini group.
- We first extract stem-branch pairs labelled year, month, day and hour. If all labels are not found, we use the first four pairs in the response. A full match requires agreement on every pillar.
Automatic extraction can miss a valid answer or pick an example from an explanation. Incomplete extraction records insufficient data for a four-pillar comparison. Full answers and extracted values are available for manual review.
English responses, case by case
A case ID identifies the same birth data across models. Each row shows engine reference pillars followed by extracted pillars. Missing values are labelled “missing”.
| Case and model | Engine reference | Extracted answer | Match |
|---|---|---|---|
| b1 · ChatGPT | 乙丑 · 己丑 · 丙寅 · 丁酉 | 乙丑 · 己丑 · 丙寅 · 丁酉 | Yes |
| b2 · ChatGPT | 丁丑 · 壬子 · 辛卯 · 辛卯 | 丁丑 · 壬子 · 辛丑 · 辛卯 | No |
| b3 · ChatGPT | 壬午 · 戊申 · 庚午 · 乙酉 | 壬午 · 戊申 · 庚辰 · 乙酉 | No |
| b4 · ChatGPT | 甲戌 · 癸酉 · 癸丑 · 甲寅 | 甲戌 · 癸酉 · 壬亥 · 壬寅 | No |
| b5 · ChatGPT | 己巳 · 壬申 · 丙寅 · 戊戌 | 己巳 · 壬申 · 戊寅 · 壬戌 | No |
| b6 · ChatGPT | 甲子 · 辛未 · 庚申 · 癸未 | 甲子 · 辛未 · 戊子 · 己未 | No |
| b7 · ChatGPT | 壬申 · 庚戌 · 戊辰 · 丙辰 | 壬申 · 庚戌 · 戊辰 · 丙辰 | Yes |
| b8 · ChatGPT | 戊寅 · 戊午 · 壬寅 · 辛亥 | 戊寅 · 戊午 · 壬子 · 辛亥 | No |
| b9 · ChatGPT | 甲申 · 壬申 · 己卯 · 癸酉 | 甲申 · 壬申 · 己卯 · 癸酉 | Yes |
| b10 · ChatGPT | 甲申 · 壬申 · 戊辰 · 壬戌 | 甲申 · 壬申 · 戊辰 · 壬戌 | Yes |
| b11 · ChatGPT | 庚申 · 乙酉 · 壬辰 · 庚戌 | 庚申 · 乙酉 · 壬辰 · 庚戌 | Yes |
| b12 · ChatGPT | 戊辰 · 壬戌 · 癸卯 · 丁巳 | 戊辰 · 壬戌 · 癸卯 · 丁巳 | Yes |
| b1 · Claude | 乙丑 · 己丑 · 丙寅 · 丁酉 | missing · missing · missing · missing | No |
| b2 · Claude | 丁丑 · 壬子 · 辛卯 · 辛卯 | missing · missing · missing · missing | No |
| b3 · Claude | 壬午 · 戊申 · 庚午 · 乙酉 | missing · missing · missing · missing | No |
| b4 · Claude | 甲戌 · 癸酉 · 癸丑 · 甲寅 | missing · missing · missing · missing | No |
| b5 · Claude | 己巳 · 壬申 · 丙寅 · 戊戌 | missing · missing · missing · missing | No |
| b6 · Claude | 甲子 · 辛未 · 庚申 · 癸未 | missing · missing · missing · missing | No |
| b7 · Claude | 壬申 · 庚戌 · 戊辰 · 丙辰 | 壬申 · 庚戌 · 庚寅 · 戊辰 | No |
| b8 · Claude | 戊寅 · 戊午 · 壬寅 · 辛亥 | 戊寅 · 戊寅 · missing · missing | No |
| b9 · Claude | 甲申 · 壬申 · 己卯 · 癸酉 | 甲申 · 壬申 · 丙戌 · 丁酉 | No |
| b10 · Claude | 甲申 · 壬申 · 戊辰 · 壬戌 | missing · missing · missing · missing | No |
| b11 · Claude | 庚申 · 乙酉 · 壬辰 · 庚戌 | 庚申 · 甲申 · 丙寅 · 戊戌 | No |
| b12 · Claude | 戊辰 · 壬戌 · 癸卯 · 丁巳 | missing · missing · missing · missing | No |
Download and reproduce the comparison
You can correct extracted values and rerun the summary while retaining the original answers. To check the engine, enter the recorded date, gender and two-hour branch in the free calculator and leave solar-time correction off.
Use a calculated chart when asking AI
Check the pillars and calculation rules first, then ask AI to explain Ten Gods or help frame questions. Birth times near a day, solar-term or hour boundary need a clear choice of calculation rules.