关于「混杂的偏差地板:旁观日志为什么越长越稳地错」,《先手探针与隐藏事实》给出的定量结论是什么?

结论:第一章量出旁观日志的偏差地板:示意参数下地板为 +0.150,不随日志变长而下降,只需 27 位随机访客,真实可买的探针就胜过无限长的日志。

数据口径:书中参数为示意,全部计算由书中程序生成,数据全部是合成数据。 本书是非个人化经营建议,不构成法律、税务或投资意见,不向任何人承诺收入。

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: 0.150

true effect of the high tier: -0.080
npassive meanpassive rmseprobe meanprobe rmse
400+0.07540.1603-0.07860.0398
1,600+0.06980.1510-0.07920.0189
6,400+0.07160.1518-0.07980.0107
25,600+0.07060.1508-0.07880.0045
bias floor of the passive contrast: +0.150
passive limit as n grows: +0.070
查看完整验算程序
# Chapter 1. p11: passive bias stays put, probe error shrinks with n.
# Model (illustrative, synthetic): a hidden segment U (1 = high intent,
# share 0.5) changes both which price a visitor saw and whether they buy.
# Passive log: P(high price | U=1) = 0.8, P(high price | U=0) = 0.2,
# because past discounts went mostly to newcomers. Purchase probability:
# 0.10 + 0.25 U - 0.08 [high price]. True effect of the high price: -0.08.
# Probe: the same visitors, but the price tier is assigned by a coin.
import math
import random

BASE, LIFT, EFFECT = 0.10, 0.25, -0.08
P_HIGH = {1: 0.8, 0: 0.2}


def draw(rng, n, randomise):
    buys = {0: 0, 1: 0}
    seen = {0: 0, 1: 0}
    for _ in range(n):
        u = 1 if rng.random() < 0.5 else 0
        if randomise:
            x = 1 if rng.random() < 0.5 else 0
        else:
            x = 1 if rng.random() < P_HIGH[u] else 0
        p = BASE + LIFT * u + EFFECT * x
        seen[x] += 1
        if rng.random() < p:
            buys[x] += 1
    if seen[0] == 0 or seen[1] == 0:
        return 0.0
    return buys[1] / seen[1] - buys[0] / seen[0]


def summary(vals):
    mean = math.fsum(vals) / len(vals)
    rmse = math.sqrt(math.fsum((v - EFFECT) ** 2 for v in vals) / len(vals))
    return mean, rmse


if __name__ == "__main__":
    rng = random.Random(39511)
    reps = 100
    print("true effect of the high tier: {:+.3f}".format(EFFECT))
    print("     n   passive mean  passive rmse   probe mean  probe rmse")
    for n in (400, 1600, 6400, 25600):
        pv = [draw(rng, n, False) for _ in range(reps)]
        rv = [draw(rng, n, True) for _ in range(reps)]
        pm, pr = summary(pv)
        rm, rr = summary(rv)
        print("{:>6,}   {:>+10.4f}  {:>11.4f}   {:>+10.4f}  {:>10.4f}"
              .format(n, pm, pr, rm, rr))
    shift = LIFT * (P_HIGH[1] - P_HIGH[0])
    print("bias floor of the passive contrast: {:+.3f}".format(shift))
    print("passive limit as n grows: {:+.3f}".format(EFFECT + shift))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《高能动性主动推理、意念强度先手扰动与商业隐藏事实勘探母典》第 1 章「混杂的偏差地板:旁观日志为什么越长越稳地错」(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-395 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026395,
  author    = {Fan, Yafu},
  title     = {High-Agency Active Inference, Intentional Perturbation Probes and Hidden Market-Truth Exploration},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论