5D Library › 多模态历法张量融合、模九十谐波拨盘与跨系统共振诊断 · English
第八章 约定预算与一项预注册的「共振年份」检验
给任何多系统合盘软件的四个问题
本章把全书变成四个问题,可以向任何多系统合盘软件——包括作者自己的——提出。它们不是择时或决策建议。
一、每套系统的输出有多少种,集中在哪里? 称骨只有 51 种总重,而且集中在中间(第一章)。
二、哪些结论来自约定,哪些来自几何? 子平冬夏的调候由几何决定,春秋依赖加权;四化表与排盘规则各派不一(第二、三章)。
三、偶然的基率是多少? 拨盘上约 81.60 个偶然图形,每年约 135.4 次太阳弧接触,40 年中约 10.00 年偶然「共振」(第四、六章,p07、p08、p13)。
四、各系统之间的相关测量过吗? 没有测量相关,就不能把「几套系统都这么说」当作更强的证据(第五、六章)。
约定预算
推导
# Chapter 8. [p15] A convention budget for a fused five-system report.
# Each line is a choice two engines can make differently (or an input
# error), with the share of charts it changes, from earlier programs:
# bone weight, birth hour off by one double-hour (p02)
# Ziping weighting 0.6 instead of 0.7 (p04)
# Zi Wei geng-stem ji convention (p06)
# Zi Wei lunar day off by one (p06)
# If the choices act independently, the share of births for which two
# engines that differ in all of them give the same fused report is the
# product of the unchanged shares. The dial orb (1.5 or 1.0 degrees)
# changes the expected picture count from 81.60 to 54.40 (p07).
BUDGET = (("bone weight, hour off by one", 0.8333),
("Ziping weighting 0.6 vs 0.7", 0.0500),
("Zi Wei geng convention", 0.0934),
("Zi Wei lunar day off by one", 0.4735))
if __name__ == "__main__":
same = 1.0
for name, s in BUDGET:
same *= 1 - s
print(" {:32s} {:.4f}".format(name, s))
print(" conventions only (Ziping, geng): differ on {:.4f}".format(
1 - (1 - 0.0500) * (1 - 0.0934)))
print(" all four, inputs included: differ on {:.4f}".format(
1 - same))
print(" dial pictures per chart, orb 1.5 vs 1.0: 81.60 vs 54.40")
print("\nReading: the conventions alone change about one fused report")
print("in seven; an uncertain birth hour or lunar day changes most.")
print("A fused report must print its conventions and its inputs'")
print("uncertainty, or two honest engines will contradict each other.")
# bone weight, hour off by one 0.8333
# Ziping weighting 0.6 vs 0.7 0.0500
# Zi Wei geng convention 0.0934
# Zi Wei lunar day off by one 0.4735
# conventions only (Ziping, geng): differ on 0.1387
# all four, inputs included: differ on 0.9244
# dial pictures per chart, orb 1.5 vs 1.0: 81.60 vs 54.40
#
# Reading: the conventions alone change about one fused report
# in seven; an uncertain birth hour or lunar day changes most.
# A fused report must print its conventions and its inputs'
# uncertainty, or two honest engines will contradict each other.把前面算出的约定与输入误差排在一起:只看两项流派约定(子平加权 0.6 对 0.7、紫微庚干四化),两个取法不同的引擎就会对 0.1387 的出生给出不同的合盘报告;再加上出生时辰差一格与农历日差一天,比例升到 0.9244。拨盘的容许度从 1.5 度改为 1.0 度,每张星盘的偶然图形从 81.60 个降到 54.40 个。一份合盘报告若不印出它的约定与输入的不确定性,两个诚实的引擎就会互相矛盾。
一项预注册的检验
假说 · 尚未验证
研究方案。 一套完整的合盘引擎(五套系统的规则、约定、共振评分公式)在招募前公开登记,此后不再改动。每位参与者提交出生资料与 20 个成年年份,并把其中发生最重大人生事件的年份密封提交给第三方。独立分析者用登记的引擎为这 20 年按共振评分排序,不看密封的年份。主要结局:事件年份的排名分位(1/20 到 20/20);没有信息时均值为 0.525。检验:单侧检验,,把握度 0.90,备择假设为均值 0.45。由与作者无经济关系、不知道事件年份的独立分析者执行,结果不论正负都发表 (Carlson 1985)。
推导
# Chapter 8. [p16] A pre-registered test of "resonance years".
# Each participant lists her adult years (20 of them) and, sealed, the
# year of her most significant life event. A registered fusion engine
# ranks the 20 years by its resonance score without seeing the event.
# The outcome is the event year's rank as a fraction (1/20 .. 20/20).
# H0: uniform ranks, mean 0.525, sd as computed exactly below. H1: the
# mean rank fraction is 0.45 (events in the engine's upper half). One-
# sided test at 0.05 with power 0.90 (normal approximation), and a
# check of the size by simulation (20,000 trials, seed 70).
import math
import random
from statistics import NormalDist
YEARS = 20
RANKS = [(k + 1) / YEARS for k in range(YEARS)]
if __name__ == "__main__":
mu = sum(RANKS) / YEARS
sd = math.sqrt(sum((r - mu) ** 2 for r in RANKS) / YEARS)
z = NormalDist().inv_cdf
delta = mu - 0.45
n = math.ceil(((z(0.95) + z(0.90)) * sd / delta) ** 2)
print(" H0 mean {:.3f}, sd {:.4f}; H1 mean 0.450".format(mu, sd))
print(" participants needed: {}".format(n))
crit = mu - z(0.95) * sd / math.sqrt(n)
rng = random.Random(70)
rej = sum(sum(rng.choice(RANKS) for _ in range(n)) / n <= crit
for _ in range(20000)) / 20000
print(" reject H0 when the mean rank fraction <= {:.4f}".format(crit))
print(" simulated size under H0: {:.4f}".format(rej))
print("\nReading: a test of resonance years needs fewer than two hundred")
print("participants and a sealed event year; without the seal, any")
print("engine can find resonance in the year the client names.")
# H0 mean 0.525, sd 0.2883; H1 mean 0.450
# participants needed: 127
# reject H0 when the mean rank fraction <= 0.4829
# simulated size under H0: 0.0495
#
# Reading: a test of resonance years needs fewer than two hundred
# participants and a sealed event year; without the seal, any
# engine can find resonance in the year the client names.需要 127 位参与者;平均排名分位不高于 0.4829 即拒绝原假设,模拟得到的实际检验水准为 0.0495。检验的关键是密封:若事件年份在排序之前就已告知引擎或解读者,任何一套引擎都可以在那一年找到「共振」——第六章显示,偶然一致本来就很常见。
小结
四个问题——输出有多少种、结论来自约定还是几何、偶然基率多少、相关测过没有——可以向任何多系统合盘软件提出。两项流派约定就使 0.1387 的合盘报告不同;加上输入误差则为 0.9244。检验「共振年份」的预注册方案需要 127 位参与者与一个事先密封的事件年份。
参考文献
Fan, Y. (2026). Multimodal Calendrical Tensor Fusion, Modulo-90 Harmonic Dials and Cross-System Resonance Diagnostics: Climate-Balance Vectors, Flying-Transformation Matrices and Solar-Arc Midpoint Trees [多模态历法张量融合、模九十谐波拨盘与跨系统共振诊断] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090189 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).