关于「实证闭环:把五件套的多评审裁决改为显式优先级,并预注册证伪方案」,《不存在完美裁决器》给出的定量结论是什么?

结论:第八章给出七步裁决流程、蒸馏小工具「评审聚合悖论检测器」与预注册证伪方案。书中以标注框区分已发表的证据、可运行的 Python 计算(21 段程序)、原典与作者的假说。

数据口径:多道检查意见不一时,经营者最常见的想法是:只要合成规则设计得足够聪明,总能得到一个公正、一致、无可挑剔的裁决。 阿罗 1950 年与 1951 年证明,选项不少于三个时,不存在同时满足无限制域、弱帕累托、无关选项独立与非独裁的排序合成规则。

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

decisions audited60
CYCLE8
SPOILER13
OVERRIDE23
SPLIT4
no flag34
first flagged decisions:
#2 BCA ABC CAB CYCLE,SPOILER,OVERRIDE
#4 BCA ABC CAB CYCLE,SPOILER,OVERRIDE
#5 ACDB DACB BCDA CYCLE,SPOILER,OVERRIDE
#6 ABC BAC BCA OVERRIDE
查看完整验算程序
# Chapter 8. p81: a quarterly audit of a multi-guard decision log.
# The log is synthetic (fixed seed): 60 decisions, each with 3 guards
# ranking 3 or 4 options uniformly at random, i.e. the worst case of
# chapter 2. For each decision the detector raises four flags:
#   CYCLE     no Condorcet winner
#   SPOILER   deleting a non-winning option changes the Borda winner
#   OVERRIDE  a majority prefers some option to the priority pick
#   SPLIT     Borda winner and Condorcet winner differ
# A real audit replaces the synthetic log with the kit's own records.
import itertools
import random

rng = random.Random(20261113)


def beats(prof, x, y):
    return 2 * len([o for o in prof if o.index(x) < o.index(y)]) > len(prof)


def condorcet(prof):
    opts = sorted(prof[0])
    for x in opts:
        if all(beats(prof, x, y) for y in opts if y != x):
            return x
    return None


def borda(prof):
    m = len(prof[0])
    s = {}
    for o in prof:
        for pos, c in enumerate(o):
            s[c] = s.get(c, 0) + m - 1 - pos
    return min(s, key=lambda c: (-s[c], c))


def flags(prof):
    out = []
    cw = condorcet(prof)
    bw = borda(prof)
    if cw is None:
        out.append("CYCLE")
    for d in prof[0]:
        if d == bw:
            continue
        sub = [tuple(c for c in o if c != d) for o in prof]
        if borda(sub) != bw:
            out.append("SPOILER")
            break
    lex = prof[0][0]
    if any(beats(prof, y, lex) for y in prof[0] if y != lex):
        out.append("OVERRIDE")
    if cw is not None and cw != bw:
        out.append("SPLIT")
    return out


if __name__ == "__main__":
    log = []
    for k in range(60):
        opts = "ABC" if k % 2 == 0 else "ABCD"
        orders = list(itertools.permutations(opts))
        log.append([rng.choice(orders) for _ in range(3)])
    count = {f: 0 for f in ("CYCLE", "SPOILER", "OVERRIDE", "SPLIT")}
    clean = 0
    for prof in log:
        fl = flags(prof)
        for f in fl:
            count[f] += 1
        if not fl:
            clean += 1
    print("decisions audited: {}".format(len(log)))
    for f in count:
        print("  {:<8} {:>3}".format(f, count[f]))
    print("  no flag  {:>3}".format(clean))
    print("first flagged decisions:")
    shown = 0
    for i, prof in enumerate(log):
        fl = flags(prof)
        if fl and shown < 4:
            shown += 1
            print("  #{:<2} {}  {}".format(
                i, " ".join("".join(o) for o in prof), ",".join(fl)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《阿罗不可能定理与多评审规则聚合、五件套拒绝码冲突时为何不存在完美裁决器母典》第 8 章「实证闭环:把五件套的多评审裁决改为显式优先级,并预注册证伪方案」(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-310 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026310,
  author    = {Fan, Yafu},
  title     = {Arrow's Impossibility Theorem and the Aggregation of Multiple Reviewers: Why No Perfect Arbiter Exists When Kit Refusal Codes Conflict},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论