关于「临界群规模:网络多大时声誉机制失效」,《商帮多边惩罚与买手信任网络》给出的定量结论是什么?

结论:第四章求临界群规模:链状熟人网络在示意参数下只有几十到一百多人,均匀混合网络要到 33742 人。

数据口径:书中代理人、订单与网络参数均为示意;全部计算由书中程序生成;本书是非个人化的经营分析,不构成经营、法律或理财建议。 第四章求临界群规模:链状熟人网络在示意参数下只有几十到一百多人,均匀混合网络要到 33742 人。

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: 33742

multilateral benchmark d* = G/(G+r) = 0.8333
Nchain k=1chain k=2mixed k=1
100.83330.83330.8333
200.83330.83330.8333
500.87060.83470.8334
1000.92260.87180.8336
2000.97370.92340.8342
10001.00000.99940.8416
10000nonenone0.8841
critical N (first size at which honesty fails)
d = 0.90: chain k=1 76, chain k=2 149, mixed k=1 33742
d = 0.95: chain k=1 142, chain k=2 282, mixed k=1 > 10^7

核心公式

\[q_t = H\,(1 - i_t),\]
查看完整验算程序
# Chapter 4. p41: the critical coalition size at which reputation fails.
# Model: a theft at period 0 is known only to the victim. News spreads
# as SI: on a chain of correspondents (k members further in each
# direction per period) or in a mixed coalition (each informed member
# writes k letters a round); M letter rounds fit in one hiring period.
# From period 1 the cheater finds an opening with probability H and is
# rehired if that member has not heard, so q_t = H (1 - i_t). Honest
# value V = w / (1 - d); boycotted cheater
# U_t = q_t V + (1 - q_t)(w0 + d U_{t+1}); honesty iff V >= w + G + d U_1.
# d*(N) by bisection. G = 5, r = 1, w0 = 1, H = 0.5, M = 12 illustrative.
# 模型:SI 扩散决定再雇用概率,求声誉失效的临界群规模。
import math

G, R, W0, H = 5.0, 1.0, 1.0, 0.5


def informed_path(n, k, kind, m=12):
    path, i, t = [], 1.0 / n, 0
    while i < 1.0 - 1e-12 and t < 100000:
        for _ in range(m):
            t += 1
            if kind == "chain":
                i = min(1.0, (1.0 + 2.0 * k * t) / n)
            else:
                i = 1.0 - (1.0 - i) * (1.0 - 1.0 / (n - 1)) ** (k * i * n)
        path.append(i)
    return path


def honest_margin(d, path):
    w = W0 + R
    v = w / (1.0 - d)
    u = W0 / (1.0 - d)
    for i in reversed(path):
        q = H * (1.0 - i)
        u = q * v + (1.0 - q) * (W0 + d * u)
    return v - math.fsum([w, G, d * u])


def d_star(path):
    lo, hi = 0.0, 0.999999
    if honest_margin(hi, path) < 0.0:
        return None
    for _ in range(50):
        mid = 0.5 * (lo + hi)
        if honest_margin(mid, path) >= 0.0:
            hi = mid
        else:
            lo = mid
    return hi


def fails(n, k, kind, d):
    ds = d_star(informed_path(n, k, kind))
    return ds is None or ds > d


def critical_n(k, kind, d):
    lo, hi = 3, 6
    while not fails(hi, k, kind, d):
        lo, hi = hi, hi * 2
        if hi > 10 ** 7:
            return "> 10^7"
    while hi - lo > 1:
        mid = (lo + hi) // 2
        if fails(mid, k, kind, d):
            hi = mid
        else:
            lo = mid
    return str(hi)


def fmt(x):
    return " none " if x is None else "{:.4f}".format(x)


if __name__ == "__main__":
    print("multilateral benchmark d* = G/(G+r) = {:.4f}".format(G / (G + R)))
    print("   N     chain k=1  chain k=2  mixed k=1")
    for n in (10, 20, 50, 100, 200, 1000, 10000):
        row = [d_star(informed_path(n, k, kind))
               for k, kind in ((1, "chain"), (2, "chain"), (1, "mix"))]
        print("  {:<6d}  {}     {}     {}".format(n, *map(fmt, row)))
    print("critical N (first size at which honesty fails)")
    for d in (0.90, 0.95):
        out = [critical_n(k, kind, d)
               for k, kind in ((1, "chain"), (2, "chain"), (1, "mix"))]
        print("  d = {:.2f}: chain k=1 {}, chain k=2 {}, mixed k=1 {}"
              .format(d, *out))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《商帮多边惩罚声誉机制、马格里布商人代理博弈与买手社群信任网络母典》第 4 章「临界群规模:网络多大时声誉机制失效」(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-222 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026222,
  author    = {Fan, Yafu},
  title     = {Multilateral Punishment in Merchant Networks: Maghribi Agency Games and Trust Networks in Buyer Communities},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论