关于「灾异年表的随机置换检验:事后拟合为何几乎总能成功」,《太乙神数的同余结构与博弈边界》给出的定量结论是什么?

结论:第六章在合成年表上做随机置换检验,最「显著」的格局族系 p 值为 0.501,计入积年与容差的选择后,至少一条路径名义显著的机会是 0.867。

数据口径:书中参数为示意;全部计算由书中程序生成;书中灾异年表为固定种子生成的合成数据。 第六章在合成年表上做随机置换检验,最「显著」的格局族系 p 值为 0.501,计入积年与容差的选择后,至少一条路径名义显著的机会是 0.867。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.501

years 1100, calamity years 224, base rate 0.204
conditions searched73
best: 'Taiyi 8'39 of 138 years, rate 0.283
nominal one-sided p0.0165
Bonferroni p1.000
conditions with nominal p < 0.053
permutation family-wise p (1000 relabellings)0.501
查看完整验算程序
# Chapter 6. p61: a random-permutation test of an after-the-fact fit.
# The chronicle is synthetic: each year 1000-2099 is marked as a
# "calamity year" with probability 0.2 (seeded), independent of every
# god. We then search all single-feature conditions (convention C) for
# the one whose years carry the most calamities, and judge that best
# fit against 1000 random relabellings of the same chronicle.
import math
import random

# Ring of sixteen gods, clockwise from zi; ring slots that are palaces
# carry the Taiyi palace number (Jinjing vol. 2 example: from Dayi (hai)
# 1, Dizhu (zi) 8, Hede (gen) 3, Gaocong (mao) 4, total 16 before xun 9).
RING = ("zi", "chou", "gen", "yin", "mao", "chen", "xun", "si",
        "wu", "wei", "kun", "shen", "you", "xu", "qian", "hai")
PAL = {0: 8, 2: 3, 4: 4, 6: 9, 8: 2, 10: 7, 12: 6, 14: 1}
POS = {p: k for k, p in PAL.items()}
TY_ORDER = (1, 2, 3, 4, 6, 7, 8, 9)          # skips the centre 5
# Wenchang starts at Wude (shen) and walks clockwise; qian and kun each
# hold it two years, so one circuit of 16 gods takes 18 years.
WC_SEQ = (11, 12, 13, 14, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 10)
BR_POS = (0, 1, 3, 4, 5, 7, 8, 9, 11, 12, 13, 15)   # 12 branches on ring


def taiyi(j):
    r = j % 360 % 72 % 24 or 24
    return TY_ORDER[(r - 1) // 3]


def wenchang(j):
    r = j % 360 % 72 % 18 or 18
    return WC_SEQ[r - 1]


def jishen(j):
    r = j % 60 or 60                     # from yin, counter-clockwise
    return BR_POS[(2 - (r - 1)) % 12]


def shiji(j):
    # Jishen laid on Hede (gen, slot 2); read where Wenchang falls
    return (wenchang(j) + 2 - jishen(j)) % 16


def count(start, ty):
    stop = POS[ty]
    k, tot = start, PAL.get(start, 1)
    if k == stop:
        return tot
    k = (k + 1) % 16
    while k != stop:
        tot += PAL.get(k, 0)
        k = (k + 1) % 16
    return tot


def host(j):
    return count(wenchang(j), taiyi(j))


def guest(j):
    return count(shiji(j), taiyi(j))


def general(s):
    # counts reduce by their units: 24 sets 4; 10 sets 1 (vol. 2)
    r = s % 10
    return r if r else (s // 10) % 10 or 1


def patterns(j):
    t = POS[taiyi(j)]
    w, g = wenchang(j), shiji(j)
    hg, gg = general(host(j)), general(guest(j))

    def near(a):
        return min((a - t) % 16, (t - a) % 16) in (1, 2)
    return {"yan": g == t, "po": near(w), "ji": near(g),
            "qiu": taiyi(j) in (hg, gg), "guan": hg == gg,
            "ge": gg + taiyi(j) == 10}


WUFU = (1, 3, 9, 7, 5)
DAYOU = (7, 8, 9, 1, 2, 3, 4, 6)
YEARS = list(range(1000, 2100))


def conditions(off):
    # every single-feature condition as a bitmask over YEARS
    masks = {}

    def put(name, i):
        masks[name] = masks.get(name, 0) | (1 << i)
    for i, y in enumerate(YEARS):
        j = y + off
        put("Taiyi {}".format(taiyi(j)), i)
        put("Wenchang {}".format(RING[wenchang(j)]), i)
        put("Shiji {}".format(RING[shiji(j)]), i)
        put("Jishen {}".format(RING[jishen(j)]), i)
        put("Wufu {}".format(WUFU[j % 225 // 45]), i)
        put("Dayou {}".format(DAYOU[j % 288 // 36]), i)
        for k, v in sorted(patterns(j).items()):
            if v:
                put("pattern " + k, i)
        if host(j) > guest(j):
            put("host longer", i)
        if host(j) >= 10:
            put("host >= 10", i)
    return masks


def tail(n, k, p):
    # P(X >= k) for Binomial(n, p)
    return math.fsum(math.comb(n, i) * p ** i * (1 - p) ** (n - i)
                     for i in range(k, n + 1))


def best(lab, masks, tabs):
    out = (2.0, "")
    for name in sorted(masks):
        m = masks[name]
        pv = tabs[name][(lab & m).bit_count()]
        if pv < out[0]:
            out = (pv, name)
    return out


if __name__ == "__main__":
    rng = random.Random(20261112)
    n = len(YEARS)
    flags = [rng.random() < 0.2 for _ in YEARS]
    lab = 0
    for i, f in enumerate(flags):
        lab |= f << i
    k_all = lab.bit_count()
    p0 = k_all / n
    masks = conditions(10153917)
    tabs = {}
    for name, m in masks.items():
        c = m.bit_count()
        tabs[name] = [tail(c, k, p0) for k in range(c + 1)]
    pv, name = best(lab, masks, tabs)
    m = masks[name]
    print(" years {}, calamity years {}, base rate {:.3f}".format(
        n, k_all, p0))
    print(" conditions searched: {}".format(len(masks)))
    print(" best: '{}': {} of {} years, rate {:.3f}".format(
        name, (lab & m).bit_count(), m.bit_count(),
        (lab & m).bit_count() / m.bit_count()))
    print(" nominal one-sided p = {:.4f}".format(pv))
    print(" Bonferroni p = {:.3f}".format(min(1.0, pv * len(masks))))
    nom = len([x for x in masks if tabs[x][(lab & masks[x]).bit_count()]
               < 0.05])
    print(" conditions with nominal p < 0.05: {}".format(nom))
    ge = 0
    idx = list(range(n))
    for _ in range(1000):
        rng.shuffle(idx)
        perm = 0
        for i in range(n):
            if flags[idx[i]]:
                perm |= 1 << i
        if best(perm, masks, tabs)[0] <= pv:
            ge += 1
    print(" permutation family-wise p (1000 relabellings) = {:.3f}"
          .format(ge / 1000))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《太乙神数十六宫游走、积年同余周期与主客算筹博弈边界》第 6 章「灾异年表的随机置换检验:事后拟合为何几乎总能成功」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-154 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026154,
  author    = {Fan, Yafu},
  title     = {Taiyi Shenshu: Sixteen-Palace Walks, Accumulated-Year Congruence Cycles and the Game Bounds of Host and Guest Counting Rods},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论