关于「实证闭环:实战工具与检查器规格」,《零客服界面与自助修正状态机》给出的定量结论是什么?

结论:第八章给出台账字段表、蒸馏小工具「工单根因聚类与自助修正状态机检查器」、按四件套分组的检查器规格与预注册检验:日入百单时检出三成下降每组需 1,089 单。书中以标注框区分已发表的证据、可运行的 Python 计算(22 段程序)、原典与作者假说。

数据口径:在合成的一个月里,可自助类别合计只占 0.869,退款、付款与杂项必须留给人;把完成率与新咨询的不确定性抽样进去,消灭率九成区间为 0.468 到 0.601,要达到 0.9 需要统一完成率 1.086,算术上做不到。 书中参数为示意,全部计算由书中程序生成;本书是非个人化经营建议,不构成法律意见。

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: 1089

contacts 335self-served 171sent to human 164
elimination rate 0.51095% interval 0.457 to 0.564
completion among eligible 0.606
week 1: contacts94elimination 0.500
week 2: contacts76elimination 0.579
week 3: contacts81elimination 0.519
week 4: contacts84elimination 0.452
checks:
C1 human reachable in <= 2 stepsgreenworst page 2 steps
C2 recompute more than oncered1 order(s)
C3 rebind before verificationgreen0 row(s)
C4 genuine users blocked by lockred0.0448 (7 locks)
C5 refund stopped by self-servicegreen0 row(s)
C6 rule change on n=600 per armredneeds 1,089 (p82)
human tickets per day 5.86; load 0.065 erlang
查看完整验算程序
# Chapter 8. p81: four weeks of the after-sale ledger and its checks
# 2,800 synthetic orders (100 a day). Each contact is a row: week,
# class, self-serviceable, finished by self-service, sent to a human,
# recompute count, rebind verified, lock hit (0.10 of not_received),
# genuine user blocked (0.4 of lock hits), refund stopped. Completion
# rates are the p21 mid-range. The checks and thresholds are the ones
# distilled into distill/tool.py. One order is given two recomputes on
# purpose, so that check C2 has something to catch.
import math
import random

NAMES = ["not_received", "wrong_input", "progress", "repeat_chase",
         "link_expired", "refund", "content_q", "payment", "other"]
W = [0.22, 0.15, 0.20, 0.12, 0.08, 0.06, 0.07, 0.05, 0.05]
MID = {"not_received": 0.75, "wrong_input": 0.70, "progress": 0.60,
       "repeat_chase": 0.675, "link_expired": 0.825, "content_q": 0.25}
STEPS_TO_HUMAN = {"order": 1, "delivery": 1, "resend": 2,
                  "recompute": 2, "progress": 1}


def wilson(k, n, z=1.96):
    p = k / n
    d = 1 + z * z / n
    c = p + z * z / (2 * n)
    h = z * math.sqrt(p * (1 - p) / n + z * z / (4 * n * n))
    return (c - h) / d, (c + h) / d


rng = random.Random(40181)
rows = []
for day in range(28):
    for _ in range(100):
        if rng.random() >= 0.122:
            continue
        k = rng.choices(range(9), W)[0]
        name = NAMES[k]
        elig = name in MID
        done = elig and rng.random() < MID[name]
        lock = name == "not_received" and rng.random() < 0.10
        rows.append({"week": day // 7 + 1, "cls": name, "elig": elig,
                     "self": done, "human": not done,
                     "recomputes": 1 if name == "wrong_input" else 0,
                     "verified": True, "lock": lock,
                     "false_block": lock and rng.random() < 0.4,
                     "refund_stopped": False})
rows[5]["recomputes"] = 2
n = len(rows)
done = 0
elig = 0
locks = 0
fb = 0
for r in rows:
    done += r["self"]
    elig += r["elig"]
    locks += r["lock"]
    fb += r["false_block"]
lo, hi = wilson(done, n)
print("contacts {}  self-served {}  sent to human {}".format(
    n, done, n - done))
print("elimination rate {:.3f}  95% interval {:.3f} to {:.3f}".format(
    done / n, lo, hi))
print("completion among eligible {:.3f}".format(done / elig))
for wk in range(1, 5):
    a = 0
    b = 0
    for r in rows:
        if r["week"] == wk:
            a += 1
            b += r["self"]
    print("  week {}: contacts {:>3}  elimination {:.3f}".format(
        wk, a, b / a))
print("checks:")
worst = max(STEPS_TO_HUMAN.values())
twice = 0
for r in rows:
    if r["recomputes"] > 1:
        twice += 1
unver = 0
stopped = 0
for r in rows:
    unver += not r["verified"]
    stopped += r["refund_stopped"]
not_rec = 0
for r in rows:
    not_rec += r["cls"] == "not_received"
fbr = fb / not_rec
checks = [
    ("C1 human reachable in <= 2 steps", "red" if worst > 2 else "green",
     "worst page {} steps".format(worst)),
    ("C2 recompute more than once", "red" if twice else "green",
     "{} order(s)".format(twice)),
    ("C3 rebind before verification", "red" if unver else "green",
     "{} row(s)".format(unver)),
    ("C4 genuine users blocked by lock",
     "red" if fbr > 0.02 else ("yellow" if fbr > 0.005 else "green"),
     "{:.4f} ({} locks)".format(fbr, locks)),
    ("C5 refund stopped by self-service", "red" if stopped else "green",
     "{} row(s)".format(stopped)),
    ("C6 rule change on n=600 per arm",
     "red" if 600 < 1089 else "green", "needs 1,089 (p82)"),
]
for name, flag, why in checks:
    print("  {:<35} {:<6} {}".format(name, flag, why))
a = (n - done) / 28 / 12 * 8 / 60
print("human tickets per day {:.2f}; load {:.3f} erlang".format(
    (n - done) / 28, a))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《日入百单零客服自愈界面、错填自助修正状态机与售后工单源头消灭母典》第 8 章「实证闭环:实战工具与检查器规格」(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-401 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026401,
  author    = {Fan, Yafu},
  title     = {Zero-Ticket Self-Repairing Order Interfaces, Self-Service Correction State Machines and Pre-Emptive Support Elimination},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论