关于「违和与因果推断:一路信号失真时,观者为何拆开而不是合并」,《三秒薄切片信任》给出的定量结论是什么?

结论:第三章用共同源因果推断说明,差距超过 3.99 个视觉标准差时观者把违和的一路拆开,而一路噪声变大会让诚实场景的误拆率从 0.0235 升到 0.0984。

数据口径:书中参数为示意;全部计算由书中程序生成;本书只写如何让真实、可兑现的信号在前三秒清晰一致,不写任何伪造资质或假权威的手法,不宣称任何气味的健康功效,是非个人化经营建议。 第一章区分信度与效度,示意参数下三秒判断只达到效度上限的 0.612,与长期评价的一致率为 0.579。

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: 0.0235, 0.0984

s_afalse split (honest)caught (offset 4)balanced acc
0.50.02350.88510.931
1.00.03180.72500.847
2.00.05240.41620.682
3.00.07700.28740.605
4.00.09840.23260.567
Cleaner weak channel: fewer honest scenes misread, more real
mismatches noticed.
查看完整验算程序
# Chapter 3. p32: how noise in one channel makes honest, congruent signals
# look like a mismatch. Same observer as p31 (s_v = 1, s_p = 10, p_c =
# 0.5). An honest scene: both channels report the same quality q ~ N(0,
# 10^2). A mismatched scene: the auditory claim is offset by 4 units from
# what the visual channel shows. The observer splits (treats the cues as
# separate) when P(common) < 0.5. We vary the auditory noise s_a, which
# stands for a muffled microphone, a hurried voice or a script read badly.
# 20,000 scenes per cell; illustrative parameters.
import math
import random

SV, SP, PC, OFFSET, N = 1.0, 10.0, 0.5, 4.0, 20000


def post_common(xv, xa, sa):
    sv, sp = SV, SP
    v1 = sv * sv * sa * sa + sv * sv * sp * sp + sa * sa * sp * sp
    q1 = ((xv - xa) ** 2 * sp * sp + xv * xv * sa * sa
          + xa * xa * sv * sv) / v1
    l1 = math.exp(-0.5 * q1) / math.sqrt(v1)
    va, vb = sv * sv + sp * sp, sa * sa + sp * sp
    l2 = math.exp(-0.5 * (xv * xv / va + xa * xa / vb)) / math.sqrt(va * vb)
    return PC * l1 / (PC * l1 + (1.0 - PC) * l2)


if __name__ == "__main__":
    rng = random.Random(41932)
    print("  s_a  false split (honest)  caught (offset {:.0f})  "
          "balanced acc".format(OFFSET))
    for sa in (0.5, 1.0, 2.0, 3.0, 4.0):
        fs = 0
        hit = 0
        for _ in range(N):
            q = rng.gauss(0.0, SP)
            if post_common(q + rng.gauss(0.0, SV),
                           q + rng.gauss(0.0, sa), sa) < 0.5:
                fs += 1
            if post_common(q + rng.gauss(0.0, SV),
                           q + OFFSET + rng.gauss(0.0, sa), sa) < 0.5:
                hit += 1
        f, h = fs / N, hit / N
        print("  {:.1f}       {:.4f}               {:.4f}            "
              "{:.3f}".format(sa, f, h, 0.5 * (h + 1.0 - f)))
    print("Cleaner weak channel: fewer honest scenes misread, more real")
    print("mismatches noticed.")

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《三秒薄切片信任捕获:视听嗅三通道同频一致性与瞬间权威建立母典》第 3 章「违和与因果推断:一路信号失真时,观者为何拆开而不是合并」(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-419 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026419,
  author    = {Fan, Yafu},
  title     = {Three-Second Thin-Slicing Trust: Visual-Auditory-Olfactory Congruence and Instant Authority Signalling},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论