关于「阶段唯一瓶颈:五环节漏斗、利特尔定律与瓶颈转移」,《瓶颈定焦与价值工时比排程》给出的定量结论是什么?
结论:四种情形的周产出都是 14.03 单,因为进来的就是这么多,四种情形都接得住。差别在流转时间与在制品。基线下,一张订单平均在链上停留 1.402 周,在制品 19.63 张,其中 12.79 张堵在承接门前。获客提速一半,流转时间降到 1.302 周,承接门前反而是 12.84 张;交付提速一半,流转时间降到 1.224 周,承接门前仍是 12.79 张。只有承接提速两成,流转时间才降到 0.673 周,在制品降到 9.42 张,承接门前只剩 2.75 张。
数据口径:史密斯 1956 年证明,单人面对一张清单,使加权延迟代价最小的次序是按价值除以耗时排序,而不是按价值排序;七件示意任务上,按重要性排比穷举最优多付 0.311。 图形学的视锥剔除与细节层次说明,固定帧预算下不该每件都精做:八件示意任务全部精做只做完三件,总价值只有最优安排的 0.783。
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: 0.673, 1.224, 1.302, 1.402, 12.79, 12.84
| case | thru/wk | flow wks | WIP | WIP before engage |
|---|---|---|---|---|
| baseline | 14.03 | 1.402 | 19.63 | 12.79 |
| acquire +50 % | 14.03 | 1.302 | 18.23 | 12.84 |
| deliver +50 % | 14.03 | 1.224 | 17.13 | 12.79 |
| engage +20 % | 14.03 | 0.673 | 9.42 | 2.75 |
| 52 weeks, work pushed in at stage-1 speed | ||||
| baseline passed acquire 1,054 passed engage 757 stuck 297 | ||||
| acquire +50 % passed acquire 1,596 passed engage 757 stuck 839 | ||||
查看完整验算程序
# Chapter 3. p32: speeding a non-bottleneck stage, simulated.
# A three-stage line (acquire -> engage -> deliver) with exponential
# work at rates 20, 15, 18 orders per week; orders released as a Poisson
# stream of 14 per week. Infinite buffers, first come first served:
# D_s,k = max(D_s-1,k, D_s,k-1) + S_s,k.
# Little's law (L = lam W) turns the mean flow time into mean WIP.
# Second block: work pushed in as fast as stage 1 can take it for 52
# weeks, showing where unfinished work piles up. Illustrative rates.
import math
import random
def line(rates, lam, n):
rng = random.Random(37632)
t, prev = 0.0, [0.0] * len(rates)
flow, wait2, last = [], [], 0.0
for _ in range(n):
t += rng.expovariate(lam)
d = t
for s, r in enumerate(rates):
start = max(d, prev[s])
if s == 1:
wait2.append(start - d)
d = start + rng.expovariate(r)
prev[s] = d
flow.append(d - t)
last = d
w = math.fsum(flow) / n
return n / last, w, lam * w, lam * math.fsum(wait2) / n
print("case thru/wk flow wks WIP WIP before engage")
cases = [("baseline", [20, 15, 18]), ("acquire +50 %", [30, 15, 18]),
("deliver +50 %", [20, 15, 27]), ("engage +20 %", [20, 18, 18])]
for name, r in cases:
th, w, L, q2 = line(r, 14.0, 200000)
print("%-17s %7.2f %8.3f %5.2f %6.2f" % (name, th, w, L, q2))
def pushed(rates, weeks):
rng = random.Random(37633)
prev = [0.0] * len(rates)
done1 = done2 = 0
t1 = 0.0
while True:
t1 += rng.expovariate(rates[0])
if t1 > weeks:
break
done1 += 1
start = max(t1, prev[1])
prev[1] = start + rng.expovariate(rates[1])
if prev[1] <= weeks:
done2 += 1
return done1, done2
print("52 weeks, work pushed in at stage-1 speed")
for name, r in cases[:2]:
a, b = pushed(r, 52.0)
print("{:<17} passed acquire {:>5,} passed engage {:>5,} "
"stuck {:>4,}".format(name, a, b, a - b))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《阶段瓶颈动态定焦、价值工时比优先级排序与认知视锥渲染加速母典》第 3 章「阶段唯一瓶颈:五环节漏斗、利特尔定律与瓶颈转移」(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-376 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026376,
author = {Fan, Yafu},
title = {Stage-Bottleneck Triage, Weighted Shortest-Processing-Time Scheduling and Level-of-Detail Execution Velocity},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}