5D Library › 塔罗超几何抽样、雷诺曼网格拓扑与卡巴拉生命树图论 · English
第五章 信息量、重复与「总有特别之处」的基率
一个牌阵能区分多少种情形
一个牌阵所能携带的最大信息量,是可区分布局数的以 2 为底的对数。这是上限:它假设每一种布局都被读成不同的意思。
推导
# Chapter 5. p11: How much a spread can say, and how often cards repeat.
# (1) Information: log2 of the number of distinguishable layouts.
# (2) Repeats: two independent 10-card spreads from a full deck share
# S cards, S hypergeometric with mean 100/78. "The same card again"
# is the rule, not the exception.
# (3) Over a run of readings: P(a named card appears at least once in
# r spreads of 10).
from fractions import Fraction
from math import comb, factorial, log2
def bits(n):
return log2(n)
if __name__ == "__main__":
rows = [("3 cards, upright only", 78 * 77 * 76),
("3 cards with reversals", 78 * 77 * 76 * 8),
("Celtic Cross, reversals",
factorial(78) // factorial(68) * 2 ** 10),
("Lenormand Grand Tableau", factorial(36))]
print(" layout bits")
for name, n in rows:
print(" {:24s} {:7.2f}".format(name, bits(n)))
share = [Fraction(comb(10, s) * comb(68, 10 - s), comb(78, 10))
for s in range(11)]
mean = sum(s * p for s, p in enumerate(share))
print("\n two Celtic Crosses: shared cards S")
for s in range(5):
print(" S = {} {:.4f}".format(s, float(share[s])))
print(" E[S] = {:.4f}; P(S >= 1) = {:.4f}; P(S >= 2) = {:.4f}".format(
float(mean), float(1 - share[0]), float(1 - share[0] - share[1])))
print("\n a named card in r spreads of 10:")
for r in (1, 3, 5, 10, 20):
p = 1 - (Fraction(68, 78)) ** r
print(" r = {:2d} P(at least once) = {:.4f}".format(r, float(p)))
one_of_22 = 1 - Fraction(comb(56, 10), comb(78, 10))
print(" P(a Celtic Cross holds at least one major) = {:.4f}".format(
float(one_of_22)))
# layout bits
# 3 cards, upright only 18.80
# 3 cards with reversals 21.80
# Celtic Cross, reversals 71.99
# Lenormand Grand Tableau 138.09
#
# two Celtic Crosses: shared cards S
# S = 0 0.2311
# S = 1 0.3916
# S = 2 0.2644
# S = 3 0.0925
# S = 4 0.0183
# E[S] = 1.2821; P(S >= 1) = 0.7689; P(S >= 2) = 0.3773
#
# a named card in r spreads of 10:
# r = 1 P(at least once) = 0.1282
# r = 3 P(at least once) = 0.3374
# r = 5 P(at least once) = 0.4964
# r = 10 P(at least once) = 0.7464
# r = 20 P(at least once) = 0.9357
# P(a Celtic Cross holds at least one major) = 0.9717三张不带逆位的牌阵约 18.80 比特,带逆位为 21.80 比特;带逆位的凯尔特十字为 71.99 比特;雷诺曼大桌阵的 36 张排列为 138.09 比特。比特数很大,说明「同一个牌阵再次出现」几乎不可能;但这并不意味着解读真的用到了这么多信息。解读通常把布局压缩成少数几个特征(几张大阿卡纳、主导花色、几张关键牌的位置),实际被使用的信息量要小得多。
「同一张牌又出现了」
两次独立的凯尔特十字共享的牌数 也服从超几何分布:从 78 张中取 10 张,其中「上一次出现过的」有 10 张。共享牌数的期望为 1.2821;至少共享一张的概率为 0.7689,至少共享两张为 0.3773。换言之,接连两次牌阵出现同一张牌是常态,而不是例外。
对一张指定的牌,连续 次十张牌阵中至少出现一次的概率为 :五次时为 0.4964,十次时为 0.7464,二十次时为 0.9357。一个凯尔特十字中至少有一张大阿卡纳的概率为 0.9717。
注
同一张牌「反复出现」的体验。 对经常抽牌的人,某张特定的牌在十次之内出现的概率约四分之三;而在 78 张牌中,总会有若干张牌在近期出现过两三次。被注意到的是那几张,没有反复出现的七十多张不会被记起。
多重比较:总有一种特征
读者在一个牌阵中会同时检视许多可能的「特征」。程序选取八种常被提到的特征,在随机牌阵上计算它们的基率:F1 四张以上大阿卡纳;F2 某个花色四张以上;F3 三张以上宫廷牌;F4 两张以上 A;F5 七张以上或三张以下逆位;F6 结果位置(第 10 张)为大阿卡纳;F7 缺少某个花色;F8 同一花色中三张连续的数字牌。
推导
# Chapter 5. p12: Something always stands out - pattern base rates.
# Eight features readers often remark on in a Celtic Cross, each scored
# on random spreads (cards and orientations both random):
# F1 four or more majors F2 one suit four or more times
# F3 three or more courts F4 two or more aces
# F5 7+ or 3- reversed F6 a major in the outcome position (10)
# F7 a suit missing entirely F8 three pips in sequence in one suit
# We report each rate, the expected number of features present and the
# chance that at least one (two, three) is present.
import random
DECK = [("M", i) for i in range(22)] + \
[(s, r) for s in "WCSP" for r in range(1, 15)]
def features(hand, rev):
minors = [c for c in hand if c[0] != "M"]
suits = [sum(c[0] == s for c in minors) for s in "WCSP"]
pips = {(s, r) for s, r in minors if r <= 10}
seq = any((s, r) in pips and (s, r + 1) in pips and (s, r + 2) in pips
for s in "WCSP" for r in range(1, 9))
nrev = sum(rev)
return [len(hand) - len(minors) >= 4, max(suits) >= 4,
sum(r >= 11 for _, r in minors) >= 3,
sum(r == 1 for _, r in minors) >= 2,
nrev >= 7 or nrev <= 3, hand[9][0] == "M",
min(suits) == 0, seq]
if __name__ == "__main__":
rng = random.Random(71012)
T = 200000
hits = [0] * 8
dist = [0] * 9
for _ in range(T):
hand = rng.sample(DECK, 10)
rev = [rng.random() < 0.5 for _ in range(10)]
f = features(hand, rev)
for i, x in enumerate(f):
hits[i] += x
dist[sum(f)] += 1
for i, h in enumerate(hits):
print(" F{} {:.4f}".format(i + 1, h / T))
mean = sum(k * d for k, d in enumerate(dist)) / T
print(" expected features per spread: {:.3f}".format(mean))
for k in (1, 2, 3):
print(" P({} or more features) = {:.4f}".format(
k, sum(dist[k:]) / T))
# F1 0.2952
# F2 0.2865
# F3 0.3331
# F4 0.0783
# F5 0.3449
# F6 0.2838
# F7 0.4346
# F8 0.0469
# expected features per spread: 2.103
# P(1 or more features) = 0.9099
# P(2 or more features) = 0.6741
# P(3 or more features) = 0.3685八种特征的基率从 0.0469(三张连号)到 0.4346(缺花色)不等;一个随机牌阵平均同时具有 2.103 种特征。至少具有一种的概率为 0.9099,两种以上为 0.6741,三种以上为 0.3685。也就是说,一个随机的凯尔特十字有九成以上的机会「显出某种特别之处」,有三分之二的机会同时显出两种。
证据
「指定的特征」与「某种特征」的区别是多重比较问题。 若事先只指定一种特征,其基率可能只有百分之五左右;若事后在八种特征中挑出发生了的那一种,基率就变成九成。这与第三章「本宫」的例子是同一个道理。
假说 · 尚未验证
作者的假说(尚未验证): 解读者在随机牌阵中报告「本次牌阵特别有意义」的比例,与该牌阵所具有的上述特征个数正相关,而与牌阵是否为当事人所抽无关。什么会证明它错: 若在盲法条件下,解读者对当事人所抽牌阵与随机替换牌阵的「意义感」评分有稳定差异,而且差异不能由特征个数解释,这一假说就不成立。
小结
牌阵的组合空间很大,但实际解读只使用少数特征。连续两次牌阵共享至少一张牌的概率约为 0.7689;随机牌阵具有八种常见特征中至少一种的概率为 0.9099。在讨论任何一次牌阵「意义」之前,应当先承认这些基率。
Fan, Y. (2026). Tarot Hypergeometric Sampling, Lenormand Grid Topology and Kabbalistic Tree Graphs: Reversal Binomial Mixtures, Grand-Tableau Manhattan Distances and Twenty-Two-Edge Spectral Centrality [塔罗超几何抽样、雷诺曼网格拓扑与卡巴拉生命树图论] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090191 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).