关于「任务有向无环图:卡恩拓扑排序、关键路径与蒙特卡洛工期」,《零传令兵调度母典》给出的定量结论是什么?

结论:第二章用卡恩排序查环、用关键路径法算浮动,确定性计划 17.4 小时在蒙特卡洛中按时完成的概率只有 0.0213。

数据口径:在四卷一批的示意模拟里,改用意图编译器加调度器之后,每批人工介入均值为 9.951 次,比人工传话的 58.4 次下降 0.830;但不可逆与对外动作必须由人签字,仅签字一项就占旧口径的 0.137,是永远降不到零的地板。 书中参数为示意,全部计算由书中程序生成;全书只写调度原理与公开流程形态,不含服务器、端口、账户、密钥与任何部署命令;本书是非个人化经营分析,不构成经营、法律或运维建议。

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: 0.0213, 17.4

deterministic plan 17.4 h
P50 23.3 h, P80 26.2 h, P95 28.9 h
P(length <= plan)0.0213
P(length <24 h) = 0.5848
criticality index of parallel tasks:
verify_refs0.1064
write_programs0.8937
write_chapters1.0000
write_meta0.0000
distill_kit0.0000
sign_shelf_publish0.4966
sign_external_register0.5033
查看完整验算程序
# Chapter 2. p23: Monte Carlo of the critical path under uncertainty.
# Model: same 15-node graph as p22. Agent tasks draw a triangular time
# (0.7 d, mode d, 2.0 d), right-skewed as overruns are. Each human gate
# waits triangular(0.5, 12, mode 4) hours. 20,000 seeded runs give the
# length quantiles, the chance of meeting a 24-hour deadline and the
# criticality index: how often each parallel task is on the longest
# path. The deterministic plan is beaten less often than one expects.
import random

DUR = {
    "claim_lock": 0.1, "read_spec": 0.5, "verify_refs": 2.0,
    "write_programs": 3.0, "write_chapters": 8.0, "write_meta": 1.0,
    "distill_kit": 1.5, "run_gate": 0.5, "write_receipt": 0.1,
    "build_book": 1.0, "sign_shelf_publish": 4.0,
    "sign_external_register": 4.0, "shelf_publish": 0.2,
    "external_register": 0.2, "ledger_append": 0.1,
}
ORDER = list(DUR)
PRED = {
    "verify_refs": ["claim_lock", "read_spec"],
    "write_programs": ["claim_lock", "read_spec"],
    "write_chapters": ["verify_refs", "write_programs"],
    "write_meta": ["verify_refs", "write_programs"],
    "distill_kit": ["verify_refs", "write_programs"],
    "run_gate": ["write_chapters", "write_meta", "distill_kit"],
    "write_receipt": ["run_gate"], "build_book": ["write_receipt"],
    "sign_shelf_publish": ["build_book"],
    "sign_external_register": ["build_book"],
    "shelf_publish": ["sign_shelf_publish"],
    "external_register": ["sign_external_register"],
    "ledger_append": ["shelf_publish", "external_register"],
}
N = 20000


def longest(dur):
    ef, arg = {}, {}
    for n in ORDER:
        ps = PRED.get(n, [])
        best = max(ps, key=lambda p: ef[p]) if ps else None
        ef[n] = (ef[best] if best else 0.0) + dur[n]
        arg[n] = best
    path, n = set(), "ledger_append"
    while n:
        path.add(n)
        n = arg[n]
    return ef["ledger_append"], path


if __name__ == "__main__":
    rng = random.Random(40723)
    lengths, hits = [], {n: 0 for n in ORDER}
    for _ in range(N):
        d = {}
        for n in ORDER:
            if n.startswith("sign_"):
                d[n] = rng.triangular(0.5, 12.0, 4.0)
            else:
                d[n] = rng.triangular(0.7 * DUR[n], 2.0 * DUR[n], DUR[n])
        t, path = longest(d)
        lengths.append(t)
        for n in path:
            hits[n] += 1
    lengths.sort()
    q = [lengths[int(p * N) - 1] for p in (0.5, 0.8, 0.95)]
    plan = longest(DUR)[0]
    print("deterministic plan {:.1f} h".format(plan))
    print("P50 {:.1f} h, P80 {:.1f} h, P95 {:.1f} h".format(*q))
    ok_plan = len([x for x in lengths if x <= plan]) / N
    ok_24 = len([x for x in lengths if x <= 24.0]) / N
    print("P(length <= plan) = {:.4f}".format(ok_plan))
    print("P(length <= 24 h) = {:.4f}".format(ok_24))
    print("criticality index of parallel tasks:")
    for n in ["verify_refs", "write_programs", "write_chapters",
              "write_meta", "distill_kit", "sign_shelf_publish",
              "sign_external_register"]:
        print("  {:<24} {:.4f}".format(n, hits[n] / N))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《总司令自然语言意图编译、有向无环任务自动拆解与零传令兵自治调度母典》第 2 章「任务有向无环图:卡恩拓扑排序、关键路径与蒙特卡洛工期」(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-407 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026407,
  author    = {Fan, Yafu},
  title     = {Commander Intent Compilation, Automated DAG Task Decomposition and Zero-Courier Multi-Agent Dispatch},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论