5D Library › Tarot Hypergeometric Sampling, Lenormand Grid Topology and Kabbalistic Tree Graphs · 中文
Chapter 8. A Four-Step Check and Two Pre-Registered Studies
A four-step check
The results of the first seven chapters compress into a checklist to use before any spread or Grand Tableau. It judges no reading right or wrong; it only makes sure the discussion rests on the right base rates.
- Write down the sampling rule first. How many cards, with or without replacement, with or without reversals, which layout. Without this none of the base rates of Chapters 1 and 3 can be computed.
- Record how cards are turned. Riffles produce no reversals; if reversals are used, state whether cards are turned one by one, by splitting and turning half, or in packets, because the three give entirely different correlations of reversals (Chapter 2).
- Name the features to look at in advance. Only a feature written down beforehand can be judged by its own base rate; picking out afterwards whichever of many features happened raises the base rate from a few per cent to nine in ten (Chapters 3 and 5).
- Replace “was it accurate?” by blind matching. Vague statements fit anyone equally, and selective recall raises remembered accuracy (Chapter 6); the only way to test whether readings discriminate is to have people pick out their own reading from several without knowing which is which.
Study A: a blind matching test
Each participant poses a question; a reader who never meets the participant deals a Celtic Cross and writes a reading. The participant then receives four readings without labels, one from the spread dealt for them and three from spreads dealt for other participants, and chooses the one that fits best. The chance hit rate is 1/4.
Calculation
# Chapter 8. p16: Two pre-registered studies and their sizes.
# (A) Blind matching: each participant receives four unlabelled Celtic
# Cross readings written by an interpreter who never meets them: one
# from a spread dealt for their question, three from spreads dealt for
# other participants. Chance of choosing their own: 1/4. Exact
# one-sided binomial test, alpha 0.05, power 0.90.
# (B) Shuffle audit: readers shuffle as they always do; we record the
# orientation of the top 10 cards. Under independent turning the
# number of changes has mean 4.5 and s.d. 1.5. Size to detect a mean
# of 4.2 (a mild clustering) two-sided at 0.05 with power 0.90.
from math import comb, ceil
from statistics import NormalDist
def tail(n, k, p):
return sum(comb(n, i) * p ** i * (1 - p) ** (n - i)
for i in range(k, n + 1))
def plan(p0, p1, alpha=0.05, power=0.90):
for n in range(20, 5000):
k = next(k for k in range(n + 1) if tail(n, k, p0) <= alpha)
if tail(n, k, p1) >= power:
return n, k, tail(n, k, p0)
if __name__ == "__main__":
for p1 in (0.32, 0.35):
n, k, a = plan(0.25, p1)
print(" (A) hit rate {:.2f} vs 0.25: {} participants, at least {}"
" hits".format(p1, n, k))
print(" exact size {:.4f}".format(a))
z = NormalDist().inv_cdf
for mu1 in (4.2, 4.0):
n = ceil(((z(0.975) + z(0.90)) * 1.5 / (4.5 - mu1)) ** 2)
print(" (B) mean changes {:.1f} vs 4.5: {} audited spreads".format(
mu1, n))
print("\nReading: the matching test needs a few hundred participants;")
print("the shuffle audit is cheap, because orientation is mechanical.")
# (A) hit rate 0.32 vs 0.25: 359 participants, at least 104 hits
# exact size 0.0485
# (A) hit rate 0.35 vs 0.25: 179 participants, at least 55 hits
# exact size 0.0485
# (B) mean changes 4.2 vs 4.5: 263 audited spreads
# (B) mean changes 4.0 vs 4.5: 95 audited spreads
#
# Reading: the matching test needs a few hundred participants;
# the shuffle audit is cheap, because orientation is mechanical.If the true hit rate is 0.32, an exact one-sided binomial test at 0.05 with power 0.90 needs 359 participants and at least 104 hits, with an actual type I error of 0.0485; if the true hit rate is 0.35, it needs 179 participants and at least 55 hits. The readings should keep only specific statements, with vague ones removed by an independent reviewer before analysis, so that the Barnum effect does not flatten any difference.
Study B: a shuffle audit
Readers shuffle as they always do, and the orientation of the top ten cards is recorded. Under independent turning the number of changes between neighbours has mean 4.5 and standard deviation 1.5. If the true mean is 4.2 (mild clustering), a two-sided test at 0.05 with power 0.90 needs 263 audited spreads; if it is 4.0, it needs 95. The models of Chapter 2 predict that turning packets pushes the mean down to between 1 and 2, which a handful of spreads would show. The study is cheap because orientation is a purely mechanical fact.
Hypothesis · not established
Pre-registered predictions (not yet tested): in Study A the hit rate does not differ significantly from 0.25; in Study B readers who turn packets have a mean number of changes significantly below 4.5, while readers who split, turn half and riffle finely do not differ significantly from 4.5. What would show them wrong: a hit rate in Study A significantly above 0.25, or no difference between the two habits in Study B.
Note
Conduct and publication. The code, exclusion rules and analysis plan of both studies should be registered publicly before data are collected; the analysis is carried out by an independent person with no financial relationship with the author; the results are published whatever they are.
Summary of the book
A tarot spread is hypergeometric sampling, a Lenormand Grand Tableau a random permutation on a grid, and the Tree of Life a graph with one reflection symmetry and Tiphereth at its centre. Whether reversals are independent is decided by the mechanics of turning; “touching”, “in its own house” and “always something special” all have high base rates that can be computed exactly; vague statements and selective recall are enough to explain most of the experience of an “accurate” reading. Every conclusion comes from the structure of the symbol systems themselves and involves no claim that any card meaning comes true.
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).