关于「方法:对数几率、相关折扣、九级阶梯与事后计分」,《多维神谕冲突消解与权重裁决》给出的定量结论是什么?

结论:第一章立方法:权重 1.0 等于宣称 0.7311 的命中率,要 27 次有结果的判定才能证实,适配最差的一格要 6872 次;相关系数 0.5 时,三个同源证人只值 1.5 个,不打折的合成在自称九成以上把握的判断里实际只对了 0.8534。穷举 8,748 种裁决,五票合成最高只到 1.9415,所以全力进攻与撤退只能经过「证据闸门」打开:盘外硬证据(已经签订的订单、已经到账的款项、已经生效的判决或监管决定、已经发生的违约)的对数几率,事先登记,默认为零。一套毫无信息的合成,期望 Brier 分数 0.2782,比一直说五五开更差,只记 20 次时却仍有 0.2352 的机会显得有用。

数据口径:本书不重新起盘,只接收前面各卷交来的一张票,用对数几率把五张票按「问事类型 × 时间尺度」加权合成;读同一份出生资料的三个系统与读发问时刻的两个系统,各自按设计效应 1 + (n - 1)ρ 打折;合成值落到九级行动阶梯上,再配一段对客户说的合成话术。 全部数字由书中程序生成;本书不是投资、法律、医疗或心理意见。

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: 0.5, 0.7311, 1.0, 27, 6872

  weight w   LR e^w   implied hit rate   logged votes needed to beat a coin
     0.5    1.6487          0.6225                    101
     0.6    1.8221          0.6457                     71
     0.7    2.0138          0.6682                     53
     0.8    2.2255          0.6900                     41
     0.9    2.4596          0.7109                     33
     1.0    2.7183          0.7311                     27
effective weights w x fit: 180 cells, smallest 0.06 (relocate liuyao natal)
                             -> hit rate 0.5150, needs 6872 votes
                             largest 1.00 -> hit rate 0.7311
share of cells whose implied hit rate is below 0.55: 0.178
for comparison: a hit rate of 0.55 needs 617 votes, 0.60 needs 153 votes
1 independent agreeing votes of weight 1.0: odds x 2.7183, p 0.7311
2 independent agreeing votes of weight 1.0: odds x 7.3891, p 0.8808
3 independent agreeing votes of weight 1.0: odds x 20.0855, p 0.9526
查看完整验算程序
# Chapter 1. p11: Every weight is a claim about accuracy, and every claim has a price in logged decisions.
# A vote of weight w multiplies the odds of "attack is right" by the likelihood ratio LR = e^w. If a system is
# symmetric (right with probability a whether the right call is attack or defend), its LR is a / (1 - a), so
# the weight implies a hit rate a = 1 / (1 + e^-w). The effective weight at a time scale is w x fit.
# Sample size to show a hit rate a beats a coin (one-sided alpha 0.05, power 0.80, normal approximation):
#   n = ((z_alpha * 0.5 + z_beta * sqrt(a (1 - a))) / (a - 0.5))^2,  z_alpha 1.6449, z_beta 0.8416.
import math

WEIGHT = {
    "money":    {"bazi": 0.9, "ziwei": 1.0, "qimen": 0.8, "liuyao": 0.7, "west": 0.5},
    "contract": {"bazi": 0.6, "ziwei": 0.8, "qimen": 1.0, "liuyao": 0.9, "west": 0.6},
    "partner":  {"bazi": 0.8, "ziwei": 1.0, "qimen": 0.7, "liuyao": 0.8, "west": 0.7},
    "love":     {"bazi": 0.7, "ziwei": 1.0, "qimen": 0.5, "liuyao": 0.8, "west": 0.9},
    "launch":   {"bazi": 0.7, "ziwei": 0.8, "qimen": 1.0, "liuyao": 0.7, "west": 0.6},
    "relocate": {"bazi": 0.8, "ziwei": 0.9, "qimen": 0.9, "liuyao": 0.6, "west": 0.5},
}
FIT = {
    "bazi":   {"natal": 1.0, "decade": 0.9, "year": 0.7, "month": 0.5, "day": 0.3, "moment": 0.2},
    "ziwei":  {"natal": 1.0, "decade": 1.0, "year": 0.9, "month": 0.6, "day": 0.3, "moment": 0.2},
    "qimen":  {"natal": 0.2, "decade": 0.3, "year": 0.5, "month": 0.8, "day": 1.0, "moment": 1.0},
    "liuyao": {"natal": 0.1, "decade": 0.2, "year": 0.4, "month": 0.7, "day": 0.9, "moment": 1.0},
    "west":   {"natal": 0.6, "decade": 0.5, "year": 0.6, "month": 0.7, "day": 0.7, "moment": 0.8},
}
ZA, ZB = 1.6449, 0.8416


def hit(w):
    return 1 / (1 + math.exp(-w))


def n_needed(a):
    return math.ceil(((ZA * 0.5 + ZB * math.sqrt(a * (1 - a))) / (a - 0.5)) ** 2)


if __name__ == "__main__":
    print("  weight w   LR e^w   implied hit rate   logged votes needed to beat a coin")
    for w in (0.5, 0.6, 0.7, 0.8, 0.9, 1.0):
        print("   {:5.1f} {:9.4f} {:15.4f} {:22d}".format(w, math.exp(w), hit(w), n_needed(hit(w))))
    eff = sorted((WEIGHT[q][s] * FIT[s][t], q, s, t) for q in WEIGHT for s in FIT for t in FIT[s])
    lo, hi = eff[0], eff[-1]
    print("effective weights w x fit: {} cells, smallest {:.2f} ({} {} {})".format(
        len(eff), lo[0], lo[1], lo[2], lo[3]))
    print("                             -> hit rate {:.4f}, needs {} votes".format(
        hit(lo[0]), n_needed(hit(lo[0]))))
    print("                             largest {:.2f} -> hit rate {:.4f}".format(hi[0], hit(hi[0])))
    share = sum(1 for e in eff if hit(e[0]) < 0.55) / len(eff)
    print("share of cells whose implied hit rate is below 0.55: {:.3f}".format(share))
    print("for comparison: a hit rate of 0.55 needs {} votes, 0.60 needs {} votes".format(
        n_needed(0.55), n_needed(0.60)))
    for k in (1, 2, 3):
        print("{} independent agreeing votes of weight 1.0: odds x {:.4f}, p {:.4f}".format(
            k, math.exp(k), hit(k)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《多维神谕冲突消解、权重裁决与总参谋行动处方母典》第 1 章「方法:对数几率、相关折扣、九级阶梯与事后计分」(版本 0.1.0). 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-132 · v0.1.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026132,
  author    = {Fan, Yafu},
  title     = {The Multi-Oracle Conflict-Resolution and Chief-of-Staff Action Codex},
  year      = {2026},
  publisher = {5D Library},
  edition   = {0.1.0},
  note      = {DOI pending}
}

本书其他章节结论