Accuracy: cross-checked against other charting software

Every number on this page is generated from the external cross-check cases kept in our repository and asserted on every change; nothing here is typed by hand.

Results are deterministic: the same input and the same engine version give byte-identical results. Current engine version: te-e248fd2794. Your agent can rerun the public cases itself with the free tool verify_vectors.

Summary

CastExternal sourcesCasesComparedMatchedNot compared
BaZi (Four Pillars) (day changes at 23:00)Yiyunpan (yypan.cn), Yuanhenglizhen (china95.net)2222220
BaZi (Four Pillars) (day changes at 0:00)Yiyunpan (yypan.cn)1212120
Zi Wei Dou ShuYuanhenglizhen (china95.net)1616160
Da Liu RenYuanhenglizhen (china95.net), Huayi (k366.com), Yi'anju (zhouyi.cc)9288884
Qi Men Dun Jia (rotating plate)Yuanhenglizhen (china95.net)2423231
Qi Men Dun Jia (flying plate)Yuanhenglizhen (china95.net)3330
Liu YaoYuanhenglizhen (china95.net), Yi'anju (zhouyi.cc)1212120
Plum Blossom (Mei Hua Yi Shu)Yuanhenglizhen (china95.net)9990

BaZi (Four Pillars) 八字

Four pillars (year, month, day, hour) from the clock time (no longitude, as charted on the sites), with the day changing at 23:00.

External sourcesCasesComparedMatchedNot compared
Yiyunpan (yypan.cn)1212120
Yuanhenglizhen (china95.net)1010100

Options: day_boundary = zi23. Cases: packages/reading/cases/bazi-external-dayboundary.json, asserted by scripts/check-bazi-day-boundary.ts.

No mismatches.

Four pillars (year, month, day, hour) from the clock time (no longitude, as charted on the sites), with the day changing at midnight.

External sourcesCasesComparedMatchedNot compared
Yiyunpan (yypan.cn)1212120

Options: day_boundary = midnight. Cases: packages/reading/cases/bazi-external-dayboundary.json, asserted by scripts/check-bazi-day-boundary.ts.

No mismatches.

Reported only: late-Zi hour stem counted from the same day's stem (the other reading), 0 / 10. Counter-check: starting the 23:00 hour stem from the same day's day stem gives a different hour pillar on these cases, so they really separate the two readings.

Where schools or sites differ

  • Schools differ on whether the day changes at 23:00 or at midnight. Yiyunpan was charted with both settings (不分子时 / 分子时); Yuanhenglizhen has no setting and changes the day at 23:00. Each batch matches when the same option is chosen. We take: Both are offered: options.day_boundary = zi23 (default, as on mingge.org) or midnight. (34 cases: yypan zi23 ×12, china95 zi23 ×10, yypan midnight ×12)
    易运盘 https://www.yypan.cn/bazi/index.php → result.php(标准盘,不填出生地 = 钟表时间,男)。zi23 = 「不分子时」(zishi=1),midnight = 「分子时」(zishi=2);页面自述「不分早夜子时 23 点换日柱,分早夜子时 24 点换日柱」,且「子时的天干都是用第二天的日干根据五鼠遁遁出来的」。2026-11-17 在已登录的 Chrome 里按页面表单提交、从结果页抽四柱(未存页面)
  • With the day changing at midnight, 23:00-0:00 keeps the day's own day pillar, and schools differ on the hour stem: Yiyunpan starts it from the next day's stem; starting it from the same day's stem matches none of these cases. We take: The hour stem from the next day's day stem (五鼠遁 on the next day), as Yiyunpan. (10 cases)
    midnight 的时柱:日柱按当天,时干按次日日干起五鼠遁(夜子时)。另一说「时干按当天日干」(2000-05-25 23:30 会得壬子时)与易运盘不合,不取

Check it yourself

verify_vectors with {"door":"bazi"} returns 24 of these cases (5 by default, up to 100 with limit). Send each call as is and compare the fields in paths with expect. Example:

{
  "call": {
    "tool": "bazi_pillars",
    "input": {
      "birth": {
        "year": 2000,
        "month": 5,
        "day": 25,
        "hour": 23,
        "minute": 30
      },
      "gender": "male",
      "options": {
        "day_boundary": "zi23"
      }
    }
  },
  "expect": {
    "year": "庚辰",
    "month": "辛巳",
    "day": "甲申",
    "hour": "甲子"
  }
}

Source notes (from the case files, in Chinese): 易运盘 https://www.yypan.cn/bazi/index.php → result.php(标准盘,不填出生地 = 钟表时间,男)。zi23 = 「不分子时」(zishi=1),midnight = 「分子时」(zishi=2);页面自述「不分早夜子时 23 点换日柱,分早夜子时 24 点换日柱」,且「子时的天干都是用第二天的日干根据五鼠遁遁出来的」。2026-11-17 在已登录的 Chrome 里按页面表单提交、从结果页抽四柱(未存页面)。元亨利贞 https://www.china95.net/paipan/bazi/ → bazi_show.asp(普通方式排盘 taiyang=0,经度 120,男);该站无早晚子时设置项,实测 = 子初换日。

Zi Wei Dou Shu 紫微斗数

Births in a leap lunar month, from the clock time (no longitude, as charted on the site): Five Elements class, Life and Body palace, and for all twelve palaces the stem, the palace name and the 43 stars placed there (the 14 main stars and the month-, day- and hour-based minor stars); 640 fields in all.

External sourcesCasesComparedMatchedNot compared
Yuanhenglizhen (china95.net)1616160

Options: leap_month = self. Cases: packages/ziwei/cases/ziwei-external-leap.json, asserted by scripts/check-ziwei-leap.ts.

No mismatches.

Reported only: with the default leap rule (split15), 8 / 16. Counter-check: with the default split at the 15th, only births on days 1-15 match; every birth from the 16th on lands in a different Life palace, so these cases really separate the two rules.

Where schools or sites differ

  • Schools differ on how a birth in a leap month counts. Yuanhenglizhen counts the whole leap month as the month itself; MingGe's default splits it at the 15th (from the 16th it counts as the next month), which differs on the births from the 16th on. The Complete Book (紫微斗数全书) counts the whole leap month as the next month. We take: Split at the 15th by default (options.leap_month = split15, as on mingge.org); self reproduces the site exactly. The Complete Book's whole-month-as-next-month rule is not offered: the charting library (iztro) has no such mode. (8 cases)
    china95 无闰月设置项;页面农历行把闰月生人印成本月数(如 2020 闰四月十六 → 04月16日),安命按本月 = 本站 leap_month: self(iztro fixLeap:false)
  • Tian Ma (天马) is not compared: it is placed from the year branch and has nothing to do with leap months. For 2025 (乙巳) the site puts it in 巳, while the common rule (巳酉丑马在亥) and MingGe put it in 亥. We take: The common rule (巳酉丑 → 亥). (16 cases)
    天马不比:它按年支安、与闰月无关;china95 对 2025 乙巳年把天马安在巳(通行「巳酉丑马在亥」,iztro 同),与本题无关的分歧,不进比对
  • When a palace holds more than nine of the compared stars, the site prints only the last character of the extra ones (e.g. 台 for 三台); such a star counts as matched when that character is printed in the same palace. We take: Not a difference of school. (5 cases)
    palaces 按地支:stem 宫干、name 宫名、stars = 下列星曜在该宫者(排序);overflowTail = 该宫超出 9 格、china95 只印了尾字的星(如「台」= 三台)。ju 五行局、ming 命宫支、shen 身宫支

Check it yourself

verify_vectors with {"door":"ziwei"} returns 16 of these cases (5 by default, up to 100 with limit). Send each call as is and compare the fields in paths with expect. Example:

{
  "call": {
    "tool": "ziwei_board",
    "input": {
      "birth": {
        "year": 2017,
        "month": 7,
        "day": 23,
        "hour": 10,
        "minute": 0
      },
      "gender": "male",
      "options": {
        "leap_month": "self"
      }
    }
  },
  "expect": {
    "five_elements_class": "金四局",
    "life_palace_branch": "寅",
    "body_palace_branch": "子",
    "major_stars": {
      "寅": [],
      "卯": [
        "廉贞",
        "破军"
      ],
      "辰": [],
      "巳": [
        "天府"
      ],
      "午": [
        "天同",
        "太阴"
      ],
      "未": [
        "武曲",
        "贪狼"
      ],
      "申": [
        "太阳",
        "巨门"
      ],
      "酉": [
        "天相"
      ],
      "戌": [
        "天机",
        "天梁"
      ],
      "亥": [
        "紫微",
        "七杀"
      ],
      "子": [],
      "丑": []
    }
  }
}

Source notes (from the case files, in Chinese): 元亨利贞 china95 紫微斗数排盘(https://www.china95.net/paipan/ziwei/ → ziwei.asp,天盘,无经度 = 钟表时间)。china95 无闰月设置项;页面农历行把闰月生人印成本月数(如 2020 闰四月十六 → 04月16日),安命按本月 = 本站 leap_month: self(iztro fixLeap:false)

Da Liu Ren 大六壬

Month General, Noble Person, day pillar, the four lessons and the three transmissions.

External sourcesCasesComparedMatchedNot compared
Yuanhenglizhen (china95.net)6664642
Huayi (k366.com)1210102
Yi'anju (zhouyi.cc)1414140

Options: month_general = per case (all zhongqi); noble_person = per case (jiaxu or jiayang, as the site was set). Cases: packages/liuren/cases/liuren-external.json, asserted by scripts/check-liuren.ts.

No mismatches.

Where schools or sites differ

  • Schools differ on the day/night Noble Person table. Yuanhenglizhen was charted with both tables; each batch matches when the same table is chosen. We take: Both are offered: options.noble_person = jiaxu (甲戊庚牛羊, default, the external sites' default) or jiayang (甲羊戊庚牛). (88 cases: china95 jiaxu ×20, china95 jiayang ×44, k366 jiaxu ×10, zhouyi jiayang ×14)
    元亨利贞 https://www.china95.net/paipan/sanshi/sanshi.asp(tyqmlr=2 大六壬)。缺省贵人甲戊庚牛羊(id 带 def);另一批以甲羊戊庚牛取(gya / k1 / edge / yj / sh)。
  • Near a mid-month solar term the site still uses the previous Month General: its term moment is later than the astronomical one, so the whole plate shifts by one place. Not compared. We take: The astronomical moment of the solar term (tyme4ts). (2 cases: c95-edge-20260823-1030, c95-yj-20260923-0900)
    元亨利贞的中气交气时刻晚于 tyme4ts:tyme4ts 定将中气为 处暑 2026年8月23日 10:18:49,站上此刻仍用 午 将(本站 巳),整盘差一位,非四课三传之误。
  • The site's page contradicts itself (its day pillar and its four lessons belong to different days; a caching error on the site). Not compared. We take: Not a difference of school. (2 cases: k366-20261008-1000, k366-20261009-1000)
    华易返回页自相矛盾:页头日柱 乙卯,四课一课之干却是 丁(四课三传是别一日的盘,与同批另一页逐字相同),站点缓存错乱。
  • At 23:00 the site's header still shows the old day pillar while its lessons already use the next day. The day pillar is not compared; the lessons match. We take: The day changes at 23:00 (options.day_boundary = zi23, default); midnight is available. (1 cases: k366-20261017-2330)
    华易页头 23 时仍写 甲子 日,但四课按次日(一课之干 乙)排 —— 页头日柱未随 23:00 换日,四课 / 贵人按次日,与本站 zi23 口径一致。
  • For daytime casts the site uses the night Noble Person table. The Noble Person is not compared; lessons and transmissions do not depend on it and match. We take: Daytime casts use the day table. (11 cases)
    易安居昼占也用夜贵表(站上 丙戌日昼占出 亥,即甲羊戊庚牛的夜贵),站点的昼夜取表错;四课三传与贵人无关,照断。

Check it yourself

verify_vectors with {"door":"liuren"} returns 88 of these cases (5 by default, up to 100 with limit). Send each call as is and compare the fields in paths with expect. Example:

{
  "call": {
    "tool": "cast_liuren",
    "input": {
      "topic": "other",
      "d": "2026-10-10",
      "t": "10:00",
      "tz": "Asia/Shanghai",
      "options": {
        "month_general": "zhongqi",
        "noble_person": "jiaxu"
      }
    }
  },
  "expect": {
    "day_ganzhi": "丁巳",
    "month_general": "辰",
    "noble": "亥",
    "lessons": [
      "丁/午",
      "午/巳",
      "巳/辰",
      "辰/卯"
    ],
    "transmissions": [
      "卯",
      "寅",
      "丑"
    ]
  }
}

Source notes (from the case files, in Chinese): 元亨利贞 https://www.china95.net/paipan/sanshi/sanshi.asp(tyqmlr=2 大六壬)。缺省贵人甲戊庚牛羊(id 带 def);另一批以甲羊戊庚牛取(gya / k1 / edge / yj / sh)。 华易 https://m.k366.com/cm/pp/liuren.htm(guishi=0,缺省贵人甲戊庚牛羊)。 易安居 https://www.zhouyi.cc/zhouyi/liuren/liuren.php(无贵人选项;昼占误用夜贵表)。

Qi Men Dun Jia 奇门遁甲

Ju number, Duty Star, Duty Door, and star, door, deity, heaven stem and earth stem of the eight outer palaces (rotating plate).

External sourcesCasesComparedMatchedNot compared
Yuanhenglizhen (china95.net)2423231

Options: yuan_rule = per case (chaibu or zhirun, as the site was set). Cases: packages/qimen/cases/qimen-external.json, asserted by scripts/check-qimen.ts.

No mismatches.

Flying plate: star, door, deity and earth stem of all nine palaces.

External sourcesCasesComparedMatchedNot compared
Yuanhenglizhen (china95.net)3330

Options: plate = fei (not offered by the paid tool). Cases: packages/qimen/cases/qimen-external-fei.json, asserted by scripts/check-qimen.ts.

No mismatches.

Reported only: heaven stems (flying plate), 10 / 27. Reported only, not asserted: the site's rule for flying-plate heaven stems could not be reconstructed (it writes 甲, which never shows on a plate).

Where schools or sites differ

  • Schools differ on how the round and Ju number are fixed. The site was charted with both split-and-patch (拆补) and intercalation (置闰); each matches when the same method is chosen. We take: All three are offered: options.yuan_rule = zhirun (default, as on mingge.org), chaibu or maoshan. (23 cases: china95 chaibu ×11, china95 zhirun ×12)
    元亨利贞网 在线奇门排盘 https://www.china95.net/paipan/qimen/ (转盘,fh0 = 拆补无闰、fh1 = 超接置闰),2026-10-19 取
  • When the Duty Star is the centre star, the site names it 天禽; MingGe names the star it travels with, 天芮. Same palace, so it counts as a match. We take: 天芮 (天禽 lodges with 天芮 in the rotating plate). (2 cases: chaibu-20261215-1000, zhirun-20261215-1000)
    外部第二来源对拍:局数 / 值符 / 值使 / 外八宫星门神与天地盘干逐格。china95 把天禽值符写作「天禽」,本站写「天芮」(同宫即同);其 cells 不含天禽随天芮挂的第二干。原始响应在 scratch/2026-10-19_rebu-qimen/dl/china95/。
  • The site's solar-term moment differs from the astronomical one by more than an hour; the one chart between the two moments is not compared. We take: The astronomical moment of the solar term (tyme4ts). (1 cases: chaibu-20261008-1500)
    china95 寒露交节写 16:08,tyme4ts 14:29(热卜 14:28)—— 两家交节时刻之间那一盘不比

Check it yourself

verify_vectors with {"door":"qimen"} returns 23 of these cases (5 by default, up to 100 with limit). Send each call as is and compare the fields in paths with expect. Example:

{
  "call": {
    "tool": "cast_qimen",
    "input": {
      "topic": "other",
      "d": "2026-10-16",
      "t": "15:51",
      "tz": "Asia/Shanghai",
      "options": {
        "yuan_rule": "chaibu"
      }
    }
  },
  "expect": {
    "ju": "阴遁三局",
    "zhifu": "天柱@1",
    "zhishi": "惊门@1",
    "palaces": {
      "1": "天柱 惊门 值符 癸+庚",
      "2": "天辅 杜门 六合 乙+己",
      "3": "天蓬 休门 九地 庚+戊",
      "4": "天任 生门 玄武 壬+乙",
      "6": "天芮 死门 螣蛇 己+丁",
      "7": "天英 景门 太阴 辛+癸",
      "8": "天心 开门 九天 丁+壬",
      "9": "天冲 伤门 白虎 戊+辛"
    }
  }
}

Matched externally but not offered as a public case: 3 (flying plate (飞盘): the paid tool serves the rotating plate only, because MingGe's verdicts are written for it).

Source notes (from the case files, in Chinese): 元亨利贞网 在线奇门排盘 https://www.china95.net/paipan/qimen/ (转盘,fh0 = 拆补无闰、fh1 = 超接置闰),2026-10-19 取。外部第二来源对拍:局数 / 值符 / 值使 / 外八宫星门神与天地盘干逐格。china95 把天禽值符写作「天禽」,本站写「天芮」(同宫即同);其 cells 不含天禽随天芮挂的第二干。 元亨利贞 china95 飞盘(fs=0)× 拆补无闰(fh=0)三盘,逐宫抄录九神 / 九星 / 九门 / 天盘干 / 地盘干。三盘都是阴遁。星 / 门 / 神 / 地盘干与引擎逐宫一致;天盘干 china95 规则未能反推(会写出「甲」),check-qimen §7 只报数不判。

Liu Yao 六爻

Hexagram from six given line values; branch, element and Six Relative of every line of the changed hexagram.

External sourcesCasesComparedMatchedNot compared
Yuanhenglizhen (china95.net)6660
Yi'anju (zhouyi.cc)6660

Options: changed_line_relatives = ben. Cases: packages/liuyao/cases/liuyao-external-bianrelative.json, asserted by scripts/check-liuyao.ts.

No mismatches.

Reported only: changed-line relatives with the other school (bian), 12 / 72. Counter-check: with the changed hexagram's palace instead, fewer lines match, so these pages really separate the two schools.

Where schools or sites differ

  • Schools differ on which palace gives the Six Relatives of a changed line. Both sites use the original hexagram's palace. We take: The original hexagram's palace (options.changed_line_relatives = ben, default); bian is available. (12 cases)
    values 为初爻→上爻的 6/7/8/9;changedLines 为变卦初爻→上爻(index0 = 初爻)的六亲、地支、五行。两站变爻六亲都按本卦宫(bianRelative: "ben")

Check it yourself

verify_vectors with {"door":"liuyao"} returns 12 of these cases (5 by default, up to 100 with limit). Send each call as is and compare the fields in paths with expect. Example:

{
  "call": {
    "tool": "cast_liuyao",
    "input": {
      "topic": "other",
      "mode": "manual",
      "values": "987877",
      "d": "2026-10-19",
      "t": "10:00",
      "tz": "Asia/Shanghai",
      "options": {
        "changed_line_relatives": "ben"
      }
    }
  },
  "expect": {
    "original": "风火家人",
    "changed": "风山渐",
    "changed_lines": [
      "妻财 辰土",
      "子孙 午火",
      "官鬼 申金",
      "妻财 未土",
      "子孙 巳火",
      "兄弟 卯木"
    ]
  }
}

Source notes (from the case files, in Chinese): values 为初爻→上爻的 6/7/8/9;changedLines 为变卦初爻→上爻(index0 = 初爻)的六亲、地支、五行。两站变爻六亲都按本卦宫(bianRelative: "ben")

Plum Blossom (Mei Hua Yi Shu) 梅花易数

Time cast: original, mutual and changed hexagram, moving line and which trigram is Body.

External sourcesCasesComparedMatchedNot compared
Yuanhenglizhen (china95.net)9990

Options: year_rule = lichun. Cases: packages/meihua/cases/meihua-external-china95.json, asserted by scripts/check-meihua.ts.

No mismatches.

Reported only: with the default year rule (lunar), 5 / 9. Counter-check: with the year changing at Chinese New Year instead, fewer pages match, so these pages really separate the two rules.

Where schools or sites differ

  • The book counts the year from Chinese New Year; the site counts it from Lichun (start of spring), with month and day still lunar. We take: Chinese New Year by default (the book's method, options.year_rule = lunar); lichun reproduces the site exactly. (9 cases)
    立春换年(年支按立春交节时刻)+ 农历月日(闰月按本月数)+ 动爻定体用。本站 yearRule:"lichun" 应逐条相合

Check it yourself

verify_vectors with {"door":"meihua"} returns 9 of these cases (5 by default, up to 100 with limit). Send each call as is and compare the fields in paths with expect. Example:

{
  "call": {
    "tool": "cast_meihua",
    "input": {
      "topic": "other",
      "way": "now",
      "d": "2025-01-31",
      "t": "12:00",
      "tz": "Asia/Shanghai",
      "options": {
        "year_rule": "lichun"
      }
    }
  },
  "expect": {
    "original": "天地否",
    "mutual": "风山渐",
    "changed": "风地观",
    "moving_line": 4,
    "body_seat": "lower"
  }
}

Source notes (from the case files, in Chinese): 立春换年(年支按立春交节时刻)+ 农历月日(闰月按本月数)+ 动爻定体用。本站 yearRule:"lichun" 应逐条相合

BaZi verdict regression (not an external cross-check)

BaZi structure verdicts: 273 cases (engine/test_cases.json). Hand-checked charts with the expected structure (格局), strength and pattern flags, asserted on every engine change. This is a regression of MingGe's own verdicts, not a cross-check against another charting site; the BaZi pillars are cross-checked above.

Determinism

Same input, same engine version, byte-identical result. Casts with an explicit d and t never depend on the moment of the request; coin tosses and lot draws are random by design and are not part of these cases. Engine changes are listed in the changelog.

中文说明

本页数字全部由仓库里的外部对拍样例生成,每次改动都由检查脚本断言,不手写。同一入参、同一引擎版本,结果逐字相同(现行引擎版本 te-e248fd2794);免费工具 verify_vectors 可取样例自己核对。