关于「细尺码表还是品牌残差:五种选码策略的误差预算」,《尺码体系与退货最小化》给出的定量结论是什么?

结论:第二章把一次选码的误差拆成五块:身体尺码表的退货概率是 0.2170,细尺码表 0.1760,完美尺码表也只有 0.1636,而品牌残差法的闭式解是 0.0302,如实模拟留存单品的筛选后是 0.0499。

数据口径:书中的品牌、门店、价格与身体尺寸参数均为示意;全部计算由书中程序生成;本书是非个人化经营建议,只涉及服装尺寸的测量,不作任何健康判断。 第二章把一次选码的误差拆成五块:身体尺码表的退货概率是 0.2170,细尺码表 0.1760,完美尺码表也只有 0.1636,而品牌残差法的闭式解是 0.0302,如实模拟留存单品的筛选后是 0.0499。

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: 0.0302, 0.1636, 0.1760, 0.2170

polmeas+idealeasebrandstylesdP(ret)P(small)
A2.2502.2504.0000.6403.0230.21700.1085
B0.4902.2504.0000.6402.7170.17600.0880
B00.0002.2504.0000.6402.6250.16360.0818
D6.2500.0000.1900.6402.6610.16850.0842
C1.3600.0000.1900.6401.4800.03020.0151
grade alone (sd -> 0): P(ret) = 0.0000
brand share of A's variance: 0.438
查看完整验算程序
# Chapter 2. p21: Error budget and fit-return probability of five sizing
# policies. Error of the garment chest against the customer's ideal:
#   e = r + N(0, s^2), r ~ Uniform(-g/2, g/2) from the 4 cm grade,
# and the item comes back for fit when |e| > T = 4 cm. The integral of
# the normal cdf has the closed form G(x) = x Phi(x) + phi(x), so every
# probability below is exact. All standard deviations are illustrative.
#   A  body chart: self-measure 1.5, ease 1.5, brand 2.0, style 0.8
#   B  finer body chart and measuring guide: self-measure 0.7
#   B0 perfect body chart and perfect tape: self-measure 0
#   D  garment (flat) chart plus one favourite garment measured at home
#   C  brand-residual recommender from 3 kept items, brand offsets
#      estimated from 5 store-measured garments per brand
import math
from statistics import NormalDist

Z = NormalDist()
GRADE, T = 4.0, 4.0
MEAS, EASE, TAU, STYLE = 1.5, 1.5, 2.0, 0.8
KEPT, FLAT, MB, NKEPT, HOME = 2.0, 1.0, 5, 3, 1.5
POP = math.sqrt(7.0 ** 2 + EASE ** 2)   # sd of the ideal chest


def gint(x):
    return x * Z.cdf(x) + Z.pdf(x)


def probs(sd, shift=0.0, tlo=T, thi=T, g=GRADE):
    h = g / 2.0

    def below(c):   # P(r + N < c), r uniform on [-h, h]
        return sd / g * (gint((c + h) / sd) - gint((c - h) / sd))
    small = below(-tlo - shift)
    large = 1.0 - below(thi - shift)
    return small, 1.0 - small - large, large


def v_off(m=MB):
    return 1.0 / (1.0 / TAU ** 2 + m / FLAT ** 2)


def v_resid(n=NKEPT, m=MB):
    obs = KEPT ** 2 + v_off(m)
    post = 1.0 / (1.0 / POP ** 2 + n / obs)
    return post + v_off(m) + STYLE ** 2


budgets = [
    ("A", [MEAS ** 2, EASE ** 2, TAU ** 2, STYLE ** 2]),
    ("B", [0.7 ** 2, EASE ** 2, TAU ** 2, STYLE ** 2]),
    ("B0", [0.0, EASE ** 2, TAU ** 2, STYLE ** 2]),
    ("D", [KEPT ** 2 + HOME ** 2, 0.0, v_off(), STYLE ** 2]),
    ("C", [v_resid() - v_off() - STYLE ** 2, 0.0, v_off(), STYLE ** 2]),
]
print("pol  meas+ideal  ease   brand  style  sd     P(ret)  P(small)")
for name, v in budgets:
    sd = math.sqrt(math.fsum(v))
    s, f, l = probs(sd)
    print("%-3s  %6.3f      %5.3f  %5.3f  %5.3f  %5.3f  %.4f  %.4f"
          % (name, v[0], v[1], v[2], v[3], sd, 1.0 - f, s))
print("grade alone (sd -> 0): P(ret) = %.4f" % (1.0 - probs(1e-9)[1]))
print("brand share of A's variance: %.3f"
      % (TAU ** 2 / math.fsum(budgets[0][1])))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《人体测量主成分尺码体系、虚荣尺码漂移与跨境退货率最小化母典》第 2 章「细尺码表还是品牌残差:五种选码策略的误差预算」(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-215 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026215,
  author    = {Fan, Yafu},
  title     = {Anthropometric Principal-Component Sizing, Vanity-Size Drift and Cross-Border Return-Rate Minimisation},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论