What does Sizing Systems and Return Minimisation conclude, in numbers, about “Closing the Loop: a Buyer's Direct-Message Size Recommender, a Return-Reason Code Table and a Pre-Registered Test”?

Answer: Chapter 8 gives a buyer's direct-message size recommender (fit probability 0.9441 in the worked example), a return-reason code table and a pre-registered test plan. Marked boxes separate published evidence, runnable Python calculations (21 programs), classical sources and the author's hypotheses.

Scope of the numbers: Its brands, shops, prices and body dimensions are illustrative; all calculations are generated by its programs; the book is not individual business advice, deals only with the measurement of garment sizes, and makes no health judgement. Marked boxes separate published evidence, runnable Python calculations (21 programs), classical sources and the author's hypotheses.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.9441

brand  n  shrunk offset  sd
Alder  3  +2.434         0.693
Birch  2  -1.868         0.825
Cedar  5  +0.407         0.550
Dune   1  +2.838         1.078
Elm    3  -0.117         0.693
kept: Alder M, Cedar L, Elm L -> ideal 109.49 cm (sd 1.20)
target Birch, offset -1.868, total sd 1.659
  M   mean error -5.35  small 0.7927  fit 0.2073  large 0.0000
  L   mean error -1.35  small 0.0553  fit 0.9441  large 0.0006
  XL  mean error +2.65  small 0.0000  fit 0.7926  large 0.2073
  XXL mean error +6.65  small 0.0000  fit 0.0553  large 0.9447
recommend L (P(fit) 0.9441)
Show the full program
# Chapter 8. p81: The buyer's direct-message size recommender, worked.
# Store data: flat-measured deviations (cm) of label M from the house
# standard 106 cm for each brand; offsets are shrunk with tau = 2.0 and
# a per-garment sd of 1.28 (p41). Customer data: the items she kept.
# Each kept item says her ideal chest is near nominal(label) + offset,
# with sd sqrt(2.0^2 + v_off); a N(106, 7.16^2) prior completes the
# precision-weighted estimate. For the target brand every label gets
# P(too small), P(fit), P(too large) with T = 4 cm and style sd 0.8.
# Brand names and numbers are invented for the demonstration.
import math
from statistics import NormalDist

Z = NormalDist()
TAU, SDG, KEPT, STYLE, T = 2.0, 1.28, 2.0, 0.8, 4.0
PRIOR_MU, PRIOR_SD = 106.0, 7.16
LABELS = ("XS", "S", "M", "L", "XL", "XXL")
MEAS = {"Alder": [2.1, 3.4, 2.8], "Birch": [-1.9, -2.6],
        "Cedar": [0.4, 1.1, -0.2, 0.6, 0.3], "Dune": [4.0],
        "Elm": [-0.8, 0.5, -0.1]}


def nominal(label):
    return 106.0 + 4.0 * (LABELS.index(label) - 2)


def offset(brand):
    xs = MEAS.get(brand, [])
    n = len(xs)
    if n == 0:
        return 0.0, TAU ** 2
    k = TAU ** 2 / (TAU ** 2 + SDG ** 2 / n)
    return k * math.fsum(xs) / n, k * SDG ** 2 / n


def ideal(kept):
    w = [1.0 / PRIOR_SD ** 2]
    y = [PRIOR_MU / PRIOR_SD ** 2]
    for brand, label in kept:
        u, v = offset(brand)
        prec = 1.0 / (KEPT ** 2 + v)
        w.append(prec)
        y.append(prec * (nominal(label) + u))
    return math.fsum(y) / math.fsum(w), 1.0 / math.fsum(w)


def table(target, kept):
    mu, var = ideal(kept)
    u, v = offset(target)
    sd = math.sqrt(var + v + STYLE ** 2)
    rows = []
    for lab in LABELS:
        m = nominal(lab) + u - mu
        ps, pl = Z.cdf((-T - m) / sd), 1.0 - Z.cdf((T - m) / sd)
        rows.append((lab, m, ps, 1.0 - ps - pl, pl))
    return mu, math.sqrt(var), u, sd, rows


print("brand  n  shrunk offset  sd")
for b in sorted(MEAS):
    u, v = offset(b)
    print("%-6s %d  %+.3f         %.3f" % (b, len(MEAS[b]), u, math.sqrt(v)))
kept = [("Alder", "M"), ("Cedar", "L"), ("Elm", "L")]
mu, s, u, sd, rows = table("Birch", kept)
print("kept: Alder M, Cedar L, Elm L -> ideal %.2f cm (sd %.2f)" % (mu, s))
print("target Birch, offset %+.3f, total sd %.3f" % (u, sd))
for lab, m, ps, pf, pl in rows:
    if pf > 0.01:
        print("  %-3s mean error %+.2f  small %.4f  fit %.4f  large %.4f"
              % (lab, m, ps, pf, pl))
best = max(rows, key=lambda r: r[3])
print("recommend %s (P(fit) %.4f)" % (best[0], best[3]))

Single source and citation

Fan, Y. (2026). Anthropometric Principal-Component Sizing, Vanity-Size Drift and Cross-Border Return-Rate Minimisation (Version 1.0.0), Chapter 8: Closing the Loop: a Buyer's Direct-Message Size Recommender, a Return-Reason Code Table and a Pre-Registered Test. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 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}
}

Other chapter answers in this book