关于「实战工具与检查器规格:每日作战排程台账、排程器与预注册检验」,《瓶颈定焦与价值工时比排程》给出的定量结论是什么?

结论:第八章给出实战工具与检查器规格:每日作战排程台账字段表、蒸馏小工具「瓶颈定焦与价值工时比排程器」的参数与输出、红黄绿检查器、日周月节奏、预注册成对周检验与推翻条件;示意模拟中五步流程平均每天 79.74 点,排满上限的「最有价值先做」习惯只有 42.85 点。书中以标注框区分已发表的证据、可运行的 Python 计算(20 段程序)、原典与作者的假说。

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: 20, 42.85, 79.74

bottleneck: engage; planning budget 210 of 360 focus minutes
plan: K1 60mK2 80mK3 50mD1 20m
pull list: R1 15mA1 45m
policyplannedmean valuep10 valuedays all done
pipeline21079.7462.000.760
H1 value, fine36042.8540.000.220
H2 list, medium36035.658.400.091
查看完整验算程序
# Chapter 8. p81: the daily pipeline against two common habits.
# Pipeline: bottleneck (min stage capacity, p31) -> frustum classes
# (p41) -> WSPT order (p11) -> level-of-detail DP inside a planning
# budget = cap / (1 + buffer) (p51, p62) -> run the day.
# Habit H1: most valuable first, everything fine, plan the full cap.
# Habit H2: list order, everything medium, plan the full cap.
# A day: actual minutes = planned * 1.2 * exp(N(0, 0.6^2)); the focus
# cap of 360 minutes is a hard stop (unfinished work earns nothing).
# Value counts fully for IN tasks, 0.3 for DEFER, 0 for CULL. If the
# pipeline's plan ends early, it pulls the unplanned non-culled tasks
# at coarse level in w/p order (the buffer is not idle time).
# 4,000 simulated days, same draws for every policy. Illustrative.
import math
import random
import statistics

CAPS = {"acquire": 20.0, "engage": 15.0, "deliver": 18.0,
        "collect": 40.0, "repeat": 24.0}
# id, stage fed (None = no path to cash), value V at fine, medium min
TASKS = [("A1", "acquire", 35.0, 90), ("X1", None, 25.0, 60),
         ("K3", "engage", 22.0, 50), ("D1", "deliver", 18.0, 40),
         ("K1", "engage", 40.0, 60), ("R1", "repeat", 12.0, 30),
         ("X2", None, 10.0, 30), ("K2", "engage", 30.0, 40)]
CAP, BUF = 360, 0.676
CREDIT = {"IN": 1.0, "DEFER": 0.3, "CULL": 0.0}


def levels(v, m, cash):
    lo = 0.2 if cash else 0.45
    return [(m // 2, lo * v), (m, 0.8 * v), (2 * m, v)]


def classify(stage, neck):
    if stage is None:
        return "CULL"
    return "IN" if stage == neck else "DEFER"


def lod_plan(items, budget):
    cap = budget // 10
    best = [0.0] * (cap + 1)
    keep = []
    for _, _, opts, cr in items:
        nxt, ch = best[:], [0] * (cap + 1)
        for b in range(cap + 1):
            for lv, (m, v) in enumerate(opts, start=1):
                u = m // 10
                if u <= b and best[b - u] + v * cr > nxt[b]:
                    nxt[b], ch[b] = best[b - u] + v * cr, lv
        keep.append(ch)
        best = nxt
    out, b = [], cap
    for i in range(len(items) - 1, -1, -1):
        lv = keep[i][b]
        if lv:
            m, v = items[i][2][lv - 1]
            out.append((items[i][0], m, v * items[i][3]))
            b -= m // 10
    return out[::-1]


def protocol():
    neck = min(CAPS, key=lambda s: CAPS[s])
    items = []
    for tid, st, v, m in TASKS:
        c = classify(st, neck)
        if c != "CULL":
            items.append((tid, c, levels(v, m, c == "IN"), CREDIT[c]))
    items.sort(key=lambda it: -it[2][1][1] / it[2][1][0])
    plan = lod_plan(items, int(CAP / (1 + BUF)) // 10 * 10)
    done = {p[0] for p in plan}
    tail = [(t, o[0][0], o[0][1] * cr) for t, _, o, cr in items
            if t not in done]
    tail.sort(key=lambda p: -p[2] / p[1])
    return sorted(plan, key=lambda p: -p[2] / p[1]), tail, neck


def habit(order_key, lv):
    neck = min(CAPS, key=lambda s: CAPS[s])
    rows = []
    for tid, st, v, m in sorted(TASKS, key=order_key):
        c = classify(st, neck)
        mm, vv = levels(v, m, c == "IN")[lv]
        rows.append((tid, mm, vv * CREDIT[c]))
    plan, used = [], 0
    for r in rows:
        if used + r[1] <= CAP:
            plan.append(r)
            used += r[1]
    return plan


def run(plan, tail=(), days=4000):
    rng = random.Random(37681)
    vals, full = [], 0
    for _ in range(days):
        t, got, k = 0.0, [], 0
        for _, m, v in plan:
            t += m * 1.2 * math.exp(rng.gauss(0.0, 0.6))
            if t <= CAP:
                got.append(v)
                k += 1
        for _, m, v in tail:
            if t >= CAP:
                break
            t += m * 1.2 * math.exp(rng.gauss(0.0, 0.6))
            if t <= CAP:
                got.append(v)
        vals.append(math.fsum(got))
        if k == len(plan):
            full += 1
    q = statistics.quantiles(vals, n=10)
    return statistics.fmean(vals), q[0], full / days


plan, tail, neck = protocol()
print("bottleneck: %s; planning budget %d of %d focus minutes" % (
    neck, int(CAP / (1 + BUF)) // 10 * 10, CAP))
print("plan: " + "  ".join("%s %dm" % (t, m) for t, m, _ in plan))
print("pull list: " + "  ".join("%s %dm" % (t, m) for t, m, _ in tail))
print("policy          planned  mean value  p10 value  days all done")
for name, pl, tl in (("pipeline", plan, tail),
                     ("H1 value, fine", habit(lambda r: -r[2], 2), ()),
                     ("H2 list, medium", habit(lambda r: 0, 1), ())):
    mins = 0
    for _, m, _ in pl:
        mins += m
    mean, p10, full = run(pl, tl)
    print("%-15s %7d  %10.2f  %9.2f  %13.3f" % (name, mins, mean, p10, full))

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 8 章「实战工具与检查器规格:每日作战排程台账、排程器与预注册检验」(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