关于「实证闭环:智能体并行数上限、分工表与预注册检验」,《团队规模的协调边界》给出的定量结论是什么?

结论:第八章给出智能体并行数上限与分工表的实证闭环、蒸馏小工具「团队规模最优计算器」与预注册检验方案:人均比值低于 0.781 即拒绝「无损失」假设。书中以标注框区分已发表的证据、可运行的 Python 计算(18 段程序)、原典与作者的假说。

数据口径:书中参数为示意;全部计算由书中程序生成;本书属于非个人化经营建议。 第四章以本出版局每波四卷并行为示意场景:容量峰值在 6.78 个窗口,四个窗口已拿到峰值的 89.6 %,协调者的审稿队列把上限压在四。

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: 0.781

z_a1.9600, z_b = 0.8416
delta sd0.15 sd=0.25 sd=0.40
0.3244 10 24
0.2009 25 63
0.10036 99 252
0.050142 393 1005
standardised d for human groups (sd1):
d0.2: 393 d=0.5: 63 d=0.8: 25
delta0.324, sd = 0.25: 10 waves per arm, se 0.112
reject 'no loss' if observed ratio < 0.781
查看完整验算程序
# Chapter 8. p82: pre-registered test: waves needed per arm.
# Outcome: gate-green volumes per agent per wave. Arm 1 runs one agent,
# arm 2 runs four in parallel. Null: no loss (per-agent ratio 1.0).
# Model prediction (p41): 0.676. Two-sided alpha = 0.05, power 0.80,
# normal approximation per arm: m = 2 (z_a + z_b)^2 sd^2 / delta^2,
# where sd is the wave-to-wave standard deviation in solo units
# (illustrative). Second block: Cohen's standardised d for human teams.
import math
from statistics import NormalDist

Z = NormalDist()
ZA = Z.inv_cdf(0.975)
ZB = Z.inv_cdf(0.80)


def per_arm(delta, sd):
    return math.ceil(2.0 * (ZA + ZB) ** 2 * sd * sd / (delta * delta))


if __name__ == "__main__":
    print("z_a = {:.4f}, z_b = {:.4f}".format(ZA, ZB))
    print("  delta    sd=0.15  sd=0.25  sd=0.40")
    for delta in (0.324, 0.20, 0.10, 0.05):
        print("  {:5.3f}  ".format(delta) + "  ".join(
            "{:7d}".format(per_arm(delta, sd)) for sd in (0.15, 0.25,
                                                          0.40)))
    print("standardised d for human groups (sd = 1):")
    print("  " + "  ".join("d={:.1f}: {}".format(d, per_arm(d, 1.0))
                           for d in (0.2, 0.5, 0.8)))
    m = per_arm(0.324, 0.25)
    se = 0.25 * math.sqrt(2.0 / m)
    print("delta = 0.324, sd = 0.25: {} waves per arm, se {:.3f}"
          .format(m, se))
    print("reject 'no loss' if observed ratio < {:.3f}".format(
        1.0 - ZA * se))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《林格曼效应与布鲁克斯定律、团队与智能体军团规模的协调开销最优边界母典》第 8 章「实证闭环:智能体并行数上限、分工表与预注册检验」(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-322 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026322,
  author    = {Fan, Yafu},
  title     = {The Ringelmann Effect and Brooks's Law: the Optimal Size of Teams and Agent Squads under Coordination Overhead},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论