关于「实证闭环:标准作业模板与检查器规格」,《注意力残留与冲刺排程》给出的定量结论是什么?
结论:第八章给出标准作业模板、蒸馏小工具「切换残留预算与冲刺排程器」的检查器规格与预注册检验,主检验需要 250 回切换、约 6.2 周。书中以标注框区分已发表的证据、可运行的 Python 计算(20 段程序)、原典与作者假说。
数据口径:书中参数为示意,全部计算由书中程序生成。 本书不是医疗、睡眠、营养或心理健康建议,不宣称任何功效。
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: 250, 6.2
H1 switches per arm 125; total 250; weeks at 40 per week 6.2 H2 decisions per group 199; total 398; weeks at 15 per week 26.5 SPRT bounds on the log-likelihood ratio: lower -1.558 upper 2.773 per-switch LLR increment, Poisson y errors (closed arm): y = 0: +0.3600 y = 1: +0.0033 y = 2: -0.3533 y = 3: -0.7100 refutation line: H1 fails if the 95 % interval of RR includes 0.95 or more
查看完整验算程序
# Chapter 8. p82: sample sizes and stopping bounds for the pre-registration
# H1 (primary): written closure lowers first-ten-minute errors. Exits are
# randomised per switch (coin flip), outcome Poisson, mu0 = 1.2,
# target rate ratio 0.70; two-sided alpha 0.05, power 0.80.
# n per arm = (z_a + z_b)^2 (1/mu1 + 1/mu0) / (ln RR)^2
# H2 (secondary): decisions taken inside the high window are reversed on
# review less often (0.10 vs 0.20); n per group by the two-proportion
# normal formula. Decisions are assigned to windows by the scheduler,
# so H2 is observational unless decision times are randomised.
# Wald SPRT for H1 (weekly review): continue while
# ln(beta / (1 - alpha)) < LLR < ln((1 - beta) / alpha).
import math
from statistics import NormalDist
ND = NormalDist()
ZA, ZB = ND.inv_cdf(0.975), ND.inv_cdf(0.80)
SW_WEEK, DEC_WEEK = 40, 15 # illustrative logging rates
mu0, rr = 1.2, 0.70
mu1 = mu0 * rr
n1 = math.ceil((ZA + ZB) ** 2 * (1.0 / mu1 + 1.0 / mu0) / math.log(rr) ** 2)
print("H1 switches per arm {:,}; total {:,}; weeks at {} per week {:.1f}"
.format(n1, 2 * n1, SW_WEEK, 2 * n1 / SW_WEEK))
p1, p2 = 0.10, 0.20
pb = (p1 + p2) / 2.0
num = (ZA * math.sqrt(2.0 * pb * (1.0 - pb))
+ ZB * math.sqrt(p1 * (1.0 - p1) + p2 * (1.0 - p2))) ** 2
n2 = math.ceil(num / (p1 - p2) ** 2)
print("H2 decisions per group {:,}; total {:,}; weeks at {} per week {:.1f}"
.format(n2, 2 * n2, DEC_WEEK, 2 * n2 / DEC_WEEK))
alpha, beta = 0.05, 0.20
upper = math.log((1.0 - beta) / alpha)
lower = math.log(beta / (1.0 - alpha))
print("SPRT bounds on the log-likelihood ratio: lower {:.3f} upper {:.3f}"
.format(lower, upper))
print("per-switch LLR increment, Poisson y errors (closed arm):")
for y in range(0, 4):
inc = y * math.log(rr) - (mu1 - mu0)
print(" y = {}: {:+.4f}".format(y, inc))
print("refutation line: H1 fails if the 95 % interval of RR includes 0.95"
" or more")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《多战线指挥官注意力残留清零、超日节律能量调配与跨场景秒级切换母典》第 8 章「实证闭环:标准作业模板与检查器规格」(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-362 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026362,
author = {Fan, Yafu},
title = {Eliminating Attention Residue in Multi-Arena Command, Ultradian Energy Pacing and Rapid Context-Switching Protocols},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}