关于「先做最重要的事为什么不对:价值工时比与相邻交换」,《瓶颈定焦与价值工时比排程》给出的定量结论是什么?

结论:k 等于零与 0.3 时,按比值排的次序几乎不变,期望代价约 292.8 点,比按重要性排省下 0.318 左右。k 等于 0.6 时,书库上架的计划耗时从 180 分钟被压到 93 分钟,它在比值排序里一下子跳到第二位,期望代价升到 316.3 点,比真实比值排序多付 0.080。即便如此,它仍比按重要性排好,差距 0.220。

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。

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: 0.080, 0.220, 0.3, 0.318, 0.6, 292.8

korder by w/planE costorder by wE costgap
0.0CAEBDFG292.8BCGDEAF385.80.318
0.3CAEBDGF292.7BCGDEAF385.80.318
0.6CBEADGF316.3BCGDEAF385.80.220
cost of the k = 0.6 bias against true WSEPT: 0.080
planned minutes at k = 0.6: 39 93 45 71 51 60 79
查看完整验算程序
# Chapter 1. p13: uncertain durations (WSEPT) and biased estimates.
# Actual minutes are log-normal with the planned mean (sigma = 0.6).
# Rothkopf (1966): ordering by w / E[p] minimises expected sum w C.
# Bias: the operator underestimates large tasks more, planned = true *
# (true / 60) ** (-k); k = 0 is unbiased. The ranking then uses the
# biased plan while costs accrue on true durations. Days and weights are
# the seven illustrative tasks of p11. 20,000 simulated days per row.
import math
import random
import statistics

TASKS = [(20, 6.0), (180, 20.0), (30, 15.0), (90, 9.0), (40, 8.0),
         (60, 5.0), (120, 10.0)]
SIG = 0.6


def simulate(order, rng, days=20000):
    mu_shift = -SIG * SIG / 2.0
    out = []
    for _ in range(days):
        t, terms = 0.0, []
        for i in order:
            t += TASKS[i][0] * math.exp(rng.gauss(mu_shift, SIG))
            terms.append(TASKS[i][1] * t / 60.0)
        out.append(math.fsum(terms))
    return statistics.fmean(out)


def plan(k):
    return [p * (p / 60.0) ** (-k) for p, _ in TASKS]


idx = range(len(TASKS))
by_w = sorted(idx, key=lambda i: (-TASKS[i][1], i))
print("k     order by w/plan   E cost   order by w   E cost   gap")
for k in (0.0, 0.3, 0.6):
    pl = plan(k)
    o = sorted(idx, key=lambda i: (-TASKS[i][1] / pl[i], i))
    c1 = simulate(o, random.Random(37613))
    c2 = simulate(by_w, random.Random(37613))
    lab = "".join("ABCDEFG"[i] for i in o)
    print("%.1f   %-16s %7.1f   %-10s %7.1f   %.3f" % (
        k, lab, c1, "BCGDEAF", c2, c2 / c1 - 1.0))
true_o = sorted(idx, key=lambda i: (-TASKS[i][1] / TASKS[i][0], i))
c0 = simulate(true_o, random.Random(37613))
o6 = sorted(idx, key=lambda i: (-TASKS[i][1] / plan(0.6)[i], i))
c6 = simulate(o6, random.Random(37613))
print("cost of the k = 0.6 bias against true WSEPT: %.3f" % (c6 / c0 - 1))
print("planned minutes at k = 0.6: " +
      " ".join("%d" % round(x) for x in plan(0.6)))

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 1 章「先做最重要的事为什么不对:价值工时比与相邻交换」(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}
}

本书其他章节结论