关于「千条规则里的冲突,多半是同义冗余」,《冲突多半是冗余》给出的定量结论是什么?
结论:第一章正面陈述并量化命题:一千条示意规则里,标红配对中真矛盾只占 0.0392,裁决后 971 条是冗余,只剩 18 条约束性规则;越界规则占一成时真矛盾比例也只有 0.2667,但无解指标已遍及全部。序中修正立项蓝图:约翰逊 1974 年证明的贪心比值是调和数级,至多 ln n 加一倍;奎因 1952 年讨论的是素蕴含项而非集合覆盖;「冲突多半是冗余」是作者假说而非已发表定律。
数据口径:书中规则库与数值均为示意;全部计算由书中程序生成;本书属非个人化经营方法,不构成法律、合规或经营决策意见,规则的增删须由主事者人工放行。 第一章正面陈述并量化命题:一千条示意规则里,标红配对中真矛盾只占 0.0392,裁决后 971 条是冗余,只剩 18 条约束性规则;越界规则占一成时真矛盾比例也只有 0.2667,但无解指标已遍及全部。
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: 0.0392, 18, 971
| illustrative library | 1000 rules, 12 metrics, seed 20261113 |
| writer mix: same 0.550, variant 0.440, opposite side 0.010 | |
| same-metric pairs | 41,369 |
| duplicate | 22,041 share 0.5328 |
| subsumed | 18,423 share 0.4453 |
| compatible | 148 share 0.0036 |
| contradiction | 757 share 0.0183 |
| pairs flagged (numbers differ) | 19,328 |
| of which true contradictions share 0.0392 | |
| binding rules kept (one per side) | 18 |
| rules redundant once ruled | 971 |
| metrics infeasible 7 of 12; rules to delete for feasibility 11 | |
| audit snapshot 2026-11-13 (report quoted, not re-run): | |
| rule lines 1333 kit + 330 canon; comparable metrics 12 | |
| threshold conflicts 3 (high 0, medium 0, low 1, ruled 2) | |
| infeasible metrics 0; duplicates 20; near-synonyms 9 |
查看完整验算程序
# Chapter 1. p11: in a 1000-rule library most flagged pairs are redundancy
# Model: an illustrative library of machine rules, each a threshold
# (metric, op, value) with op in >=, <=, ==. Twelve metrics, each with
# one canonical value; every new rule restates the canonical rule
# (paraphrase), writes a stricter or looser variant in the same
# direction, or (rarely) writes a rule that crosses the opposite side.
# Writer mix 0.550 / 0.440 / 0.010 is illustrative; fixed seed.
# A naive checker flags every pair on one metric whose numbers differ.
# Exact rational comparison with fractions.Fraction throughout.
# The last block quotes counts from the read-only audit report of the
# production pruner (2026-11-13 snapshot), typed in, not re-run here.
import random
from collections import Counter
from fractions import Fraction as F
from itertools import combinations
SEED = 20261113
N_RULES = 1000
P_OPP = F(1, 100)
P_VAR = F(44, 100)
METRICS = [
("tap_px", "lower", F(44), F(4)),
("font_px", "lower", F(16), F(1)),
("contrast", "lower", F(9, 2), F(1, 2)),
("refund_days", "lower", F(7), F(7)),
("followups", "upper", F(2), F(1)),
("msgs_day", "upper", F(1), F(1)),
("pushes_week", "upper", F(4), F(1)),
("discount_pct", "upper", F(30), F(5)),
("quiet_start", "upper", F(21), F(1)),
("settle_day", "exact", F(5), F(1)),
("launch_hour", "exact", F(10), F(1)),
("resets", "exact", F(0), F(1)),
]
def make_rule(rng, p_opp=P_OPP, p_var=P_VAR):
name, kind, c, step = METRICS[rng.randrange(len(METRICS))]
u = F(rng.randrange(1000), 1000)
if kind == "exact":
if u < p_opp:
return (name, "==", c + step * rng.choice([1, 2, 3]))
return (name, "==", c)
op = ">=" if kind == "lower" else "<="
if u < p_opp:
k = rng.choice([1, 2])
if kind == "lower":
return (name, "<=", c - step * k)
return (name, ">=", c + step * k)
if u < p_opp + p_var:
k = rng.choice([-2, -1, 1, 2])
return (name, op, max(F(0), c + step * k))
return (name, op, c)
def pair_kind(a, b):
(oa, va), (ob, vb) = sorted([a, b])
if (oa, va) == (ob, vb):
return "duplicate"
if oa == ob and oa != "==":
return "subsumed"
if oa == ob:
return "contradiction"
if (oa, ob) == ("<=", ">="):
return "contradiction" if vb > va else "compatible"
if oa == "<=": # ("<=", "==")
return "contradiction" if vb > va else "subsumed"
return "contradiction" if va < vb else "subsumed" # ("==", ">=")
def classify(rules):
by = {}
for name, op, v in rules:
by.setdefault(name, Counter())[(op, v)] += 1
tally = Counter()
for name in sorted(by):
keys = sorted(by[name].items())
for (k, n) in keys:
tally["duplicate"] += n * (n - 1) // 2
for (ka, na), (kb, nb) in combinations(keys, 2):
tally[pair_kind(ka, kb)] += na * nb
return by, tally
def min_delete(cnt):
vals = sorted({v for (_, v) in cnt})
best = None
for t in vals:
cost = 0
for (op, v), n in cnt.items():
if (op == ">=" and v > t) or (op == "<=" and v < t):
cost += n
if op == "==" and v != t:
cost += n
best = cost if best is None else min(best, cost)
return best
def kernel_size(cnt):
ops = {op for (op, _) in cnt}
return len(ops & {">=", "<="}) + (1 if "==" in ops else 0)
rng = random.Random(20261113)
rules = [make_rule(rng) for _ in range(N_RULES)]
by, tally = classify(rules)
pairs = tally["duplicate"] + tally["subsumed"] + tally["compatible"] \
+ tally["contradiction"]
flagged = pairs - tally["duplicate"]
kern = 0
dele = 0
bad = 0
for name in sorted(by):
kern += kernel_size(by[name])
d = min_delete(by[name])
dele += d
bad += 1 if d else 0
print("illustrative library: %d rules, %d metrics, seed %d"
% (N_RULES, len(METRICS), SEED))
print("writer mix: same %.3f, variant %.3f, opposite side %.3f"
% (1 - P_OPP - P_VAR, P_VAR, P_OPP))
print("same-metric pairs %7s" % format(pairs, ","))
for k in ("duplicate", "subsumed", "compatible", "contradiction"):
print(" %-14s %7s share %.4f"
% (k, format(tally[k], ","), tally[k] / pairs))
print("pairs flagged (numbers differ) %7s" % format(flagged, ","))
print(" of which true contradictions share %.4f"
% (tally["contradiction"] / flagged))
print("binding rules kept (one per side) %7d" % kern)
print("rules redundant once ruled %7d" % (N_RULES - kern - dele))
print("metrics infeasible %d of %d; rules to delete for feasibility %d"
% (bad, len(METRICS), dele))
print("audit snapshot 2026-11-13 (report quoted, not re-run):")
print(" rule lines 1333 kit + 330 canon; comparable metrics 12")
print(" threshold conflicts 3 (high 0, medium 0, low 1, ruled 2)")
print(" infeasible metrics 0; duplicates 20; near-synonyms 9")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《跨卷规则消解与公理归约、千条机器规则阈值冲突检测与最小原语集母典》第 1 章「千条规则里的冲突,多半是同义冗余」(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-258 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026258,
author = {Fan, Yafu},
title = {Cross-Volume Rule Resolution and Axiom Reduction: Detecting Threshold Conflicts among a Thousand Machine Rules and Finding a Minimal Primitive Set},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}