关于「四步剃刀:因果链条、数值量级、失效边界与反例」,《解释深度剃刀与校准》给出的定量结论是什么?
结论:四步剃刀给每条判断打四个分:链条、量级、边界、反例,加权后 0.60 及格。合成数据里,确信度最高的两条判断恰好都不及格。费米拆分只在因子误差小且互不相关时有用,否则可能比直觉更差。链条越长,结论越弱,验证应当先挑最弱的环。
数据口径:书中参数为示意,数据全部合成,全部计算由书中程序生成。 第一章在合成数据里量出装置类自评下降 1.283 分,事实类下降区间从 -0.263 到 0.063,跨过零,并给出剃刀与检索的分流规则。
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: 0.60
id conf chain mag bound counter score verdict flags A1 0.90 0.667 0.50 0.333 0.00 0.425 fail HICONF,NOCX A2 0.70 0.750 1.00 0.667 1.00 0.846 pass - A3 0.80 0.375 0.00 0.000 0.50 0.231 fail NONUM A4 0.60 1.000 1.00 1.000 1.00 1.000 pass - A5 0.95 0.500 0.00 0.333 0.00 0.242 fail HICONF,NONUM,NOCX A1: new channel repeat rate beats old A2: price cut lifts weekly units A3: FAQ table cuts repeat questions A4: parallel drafting shortens cycle A5: old readers accept bundle price
查看完整验算程序
# Chapter 3. p31: the four-step razor score of a written assumption.
# Step 1 chain: c = min(nodes, 4) / 4 * testable_links / links.
# Step 2 magnitude: m = 0 (adjective only), 0.5 (a point number),
# 1 (a range with a unit).
# Step 3 boundary: b = min(conditions, 3) / 3.
# Step 4 counterexample: x = 0 none, 0.5 stated but not observable,
# 1 at least one observable counterexample.
# Score S = 0.35 c + 0.25 m + 0.20 b + 0.20 x; pass at S >= 0.60.
# Weights and the pass line are illustrative and must be re-fitted on the
# reader's own ledger (chapter 8). Assumptions below are synthetic.
import math
W = (0.35, 0.25, 0.20, 0.20)
PASS = 0.60
def razor(nodes, testable, mag, cond, counter):
links = max(1, nodes - 1)
c = min(nodes, 4) / 4.0 * min(testable, links) / links
if nodes < 2:
c = 0.0
b = min(cond, 3) / 3.0
parts = (c, mag, b, counter)
return parts, math.fsum(w * p for w, p in zip(W, parts))
ITEMS = [ # id, label, conf, nodes, testable links, mag, cond, counter
("A1", "new channel repeat rate beats old", 0.90, 4, 2, 0.5, 1, 0.0),
("A2", "price cut lifts weekly units", 0.70, 3, 2, 1.0, 2, 1.0),
("A3", "FAQ table cuts repeat questions", 0.80, 3, 1, 0.0, 0, 0.5),
("A4", "parallel drafting shortens cycle", 0.60, 5, 4, 1.0, 3, 1.0),
("A5", "old readers accept bundle price", 0.95, 2, 1, 0.0, 1, 0.0),
]
print("id conf chain mag bound counter score verdict flags")
for iid, label, conf, nodes, t, mag, cond, cx in ITEMS:
(c, m, b, x), s = razor(nodes, t, mag, cond, cx)
flags = []
if conf >= 0.9 and s < PASS:
flags.append("HICONF")
if m == 0.0:
flags.append("NONUM")
if x == 0.0:
flags.append("NOCX")
verdict = "pass" if s >= PASS else "fail"
print("{} {:.2f} {:.3f} {:.2f} {:.3f} {:.2f} {:.3f} {:<7} {}"
.format(iid, conf, c, m, b, x, s, verdict, ",".join(flags) or "-"))
for iid, label, *_ in ITEMS:
print("{}: {}".format(iid, label))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《解释深度错觉破除、认知边界校准与真知假知剥离剃刀母典》第 3 章「四步剃刀:因果链条、数值量级、失效边界与反例」(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-396 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026396,
author = {Fan, Yafu},
title = {Shattering the Illusion of Explanatory Depth, Epistemic Calibration and the Razor Separating True Knowledge from Assumed Familiarity},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}