关于「实战工具与检查器规格:假知剥离台账、四步剃刀评分器与预注册检验」,《解释深度剃刀与校准》给出的定量结论是什么?
结论:第八章给出实战工具与检查器规格:假知剥离台账、蒸馏小工具「四步剃刀与决策前假设清单评分器」、按四件套分组的检查器、预注册检验与推翻条件;检验自评下降需 88 名评分者,检验 Brier 小幅改善每组需 1,047 人。书中以标注框区分已发表的证据、可运行的 Python 计算(22 段程序)、原典与作者的假说。
数据口径:书中参数为示意,数据全部合成,全部计算由书中程序生成。 第一章在合成数据里量出装置类自评下降 1.283 分,事实类下降区间从 -0.263 到 0.063,跨过零,并给出剃刀与检索的分流规则。
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: 1047, 88
| H1 raters for a 0.30 drop | 88 |
| H2 raters for a 0.30 domain gap | 88 |
| H3 forecasts per person m, persons per group, total forecasts | |
| m | 20 sd 0.0490 n/group 1,047 total 41,880 |
| m | 40 sd 0.0374 n/group 611 total 48,880 |
| m | 80 sd 0.0300 n/group 393 total 62,880 |
| m | 160 sd 0.0255 n/group 284 total 90,880 |
| H4 resolved forecasts per bin | 97 |
查看完整验算程序
# Chapter 8. p82: sample sizes for the pre-registered tests.
# H1 (rating drop, mechanism claims): paired, d_z = 0.30, two-sided
# alpha 0.05, power 0.80: n = ((z_a + z_b) / d_z)^2.
# H2 (domain gap): drop on mechanism claims exceeds drop on fact claims
# by 0.30 rating-sd, within person: same formula.
# H3 (Brier gain from razor + reasons-against): two groups of decision
# makers, each scored on m resolved forecasts; person-level sd of the
# mean Brier = sqrt(tau^2 + sigma^2 / m) with sigma = 0.20 per
# forecast, tau = 0.02 between persons; smallest gain worth acting
# on = 0.006 (about the delta 0.4 row of p51).
# H4 (reliability table): forecasts per bin so the observed frequency is
# known to +-0.10 at 95%: n = 1.96^2 * 0.25 / 0.10^2.
import math
from statistics import NormalDist
Z = NormalDist()
ZA, ZB = Z.inv_cdf(0.975), Z.inv_cdf(0.80)
K = (ZA + ZB) ** 2
print("H1 raters for a 0.30 drop: {}".format(math.ceil(K / 0.30 ** 2)))
print("H2 raters for a 0.30 domain gap: {}".format(math.ceil(K / 0.09)))
print("H3 forecasts per person m, persons per group, total forecasts")
for m in (20, 40, 80, 160):
sd = math.sqrt(0.02 ** 2 + 0.20 ** 2 / m)
n = math.ceil(2.0 * K * sd ** 2 / 0.006 ** 2)
print(" m = {:3d} sd {:.4f} n/group {:5,} total {:7,}".format(
m, sd, n, 2 * n * m))
print("H4 resolved forecasts per bin: {}".format(
math.ceil(ZA ** 2 * 0.25 / 0.10 ** 2)))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《解释深度错觉破除、认知边界校准与真知假知剥离剃刀母典》第 8 章「实战工具与检查器规格:假知剥离台账、四步剃刀评分器与预注册检验」(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-396 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026396,
author = {Fan, Yafu},
title = {Shattering the Illusion of Explanatory Depth, Epistemic Calibration and the Razor Separating True Knowledge from Assumed Familiarity},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}