关于「凭据卫生:格式校验与吊销名单」,《单兵零信任母典》给出的定量结论是什么?
结论:第三章说明熵判别分不开十六进制密钥与提交哈希,格式加校验码之扫描,月报警只有 1.5 条;吊销名单用布隆过滤器压缩。
数据口径:全书只写防守,不写任何攻击、入侵、绕过或规避检测手法;书中主机、域名、令牌全为合成,参数为示意,全部计算由书中程序生成;本书为非个人化经营分析,不构成安全审计或法律意见,任何生产变更须经人工复核放行。 书中以标注框区分已发表之证据、可运行之 Python 计算(20 段程序)、原典与作者之假说。
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: 1.5
| rule tpr fpr alerts precision | |
| entropy t | 3.5 0.9745 0.799250 159,852.9 0.000018 |
| entropy t | 4.0 0.5000 0.010000 2,001.5 0.000749 |
| format F | 0.5000 0.000000 1.5 0.999998 |
| format F, all keys A | 1.0000 0.000000 3.0 0.999999 |
| missed per 1,000 leaked secrets, published detection shares: | |
| TruffleHog, Search found 0.2524, missed | 747.6 |
| TruffleHog, BigQuery found 0.2939, missed | 706.1 |
| regex only, Search found 0.1926, missed | 807.4 |
| regex only, BigQuery found 0.2229, missed | 777.1 |
| findable after push in a median of 20 s: scan before push |
核心公式
\[H = -\sum_i \frac{c_i}{n}\log_2\frac{c_i}{n}.\]
查看完整验算程序
# Chapter 3. p33: base rates decide whether a secret scanner is usable.
# A one-person shop scans N = 200,000 long random-looking strings a month
# (hashes, ids, blobs); s = 3 of them are real secrets (illustrative).
# The strings are half git-style hashes, a quarter identifiers and a
# quarter paths; real secrets are half A and half B style. Flag shares
# come from p31 (t = 3.5: hashes 0.9510, identifiers 0.4985, paths 0.7965,
# B keys 0.9490; t = 4.0: paths 0.0400, B keys 0). Format F with checksum
# flags A tokens only. Alerts = N fpr + s tpr and
# precision = s tpr / alerts (Bayes). Published detection shares from
# Meli et al. (2019): TruffleHog found 0.25236 of their Search secrets
# and 0.2939 of BigQuery secrets; with its regexes only, 0.19263 and
# 0.2229. Their own distinct-format regexes: 133,934 of 134,887 unique
# Search candidates were valid, an accuracy of 99.29 percent.
import math
N, S = 200000, 3
def scan(tpr, fpr):
alerts = math.fsum([N * fpr, S * tpr])
return alerts, (S * tpr / alerts if alerts else 0.0)
if __name__ == "__main__":
print("rule tpr fpr alerts precision")
fpr35 = math.fsum([0.5 * 0.9510, 0.25 * 0.4985, 0.25 * 0.7965])
rules = [("entropy t=3.5", math.fsum([0.5, 0.5 * 0.9490]), fpr35),
("entropy t=4.0", 0.5, 0.25 * 0.0400),
("format F", 0.5, 1.0 / 62 ** 6),
("format F, all keys A", 1.0, 1.0 / 62 ** 6)]
for name, tpr, fpr in rules:
a, prec = scan(tpr, fpr)
print("{:<24} {:>6.4f} {:>8.6f} {:>10,.1f} {:>9.6f}".format(
name, tpr, fpr, a, prec))
print("missed per 1,000 leaked secrets, published detection shares:")
for name, share in (("TruffleHog, Search", 0.25236),
("TruffleHog, BigQuery", 0.2939),
("regex only, Search", 0.19263),
("regex only, BigQuery", 0.2229)):
print(" {:<22} found {:.4f}, missed {:>6.1f}".format(
name, share, 1000.0 * (1.0 - share)))
print("findable after push in a median of 20 s: scan before push")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《单兵硅基帝国零信任网安架构、源站隧道隐身与反爬虫防勒索母典》第 3 章「凭据卫生:格式校验与吊销名单」(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-336 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026336,
author = {Fan, Yafu},
title = {Zero-Trust Cybersecurity for a One-Commander Silicon Empire: Origin Cloaking, Secret Hygiene and Anti-Scraper Defense},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}