关于「结构同态:衍生品必须是母书论证图的同态像」,《衍生品的出处与保真》给出的定量结论是什么?
结论:倒置和翻转,结构检查每次都查得出,哈希一次也查不出,因为论断的文字没变。随机添加一条边,结构检查查出 0.9620;没查出的部分里,有 0.0160 是碰巧加上了一条母书里本来就有的边,这不算走样。两个论断连同编号互换位置,结构检查查出 0.9802,剩下的是那些互换的论断恰好不在衍生品画出的任何一条边上,等于什么也没改。数字被改动一位,结构检查一次也查不出,哈希每次都查出。两种检查合起来,除了删边以外,每一种改动的检出率都在 0.9620 以上。删边两者都不报,因为删边是允许的,它由覆盖率去报告。
数据口径:书中参数为示意,代价用记账点数而非价格;全部计算由书中程序生成。 书中以标注框区分已发表的证据、可运行的 Python 计算(19 段程序)、原典与作者的假说。
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: 0.0160, 0.9620, 0.9802
| edit | structure | hash | either | harmless |
|---|---|---|---|---|
| drop | 0.0000 | 0.0000 | 0.0000 | 0.0000 |
| reverse | 1.0000 | 0.0000 | 1.0000 | 0.0000 |
| flip | 1.0000 | 0.0000 | 1.0000 | 0.0000 |
| add | 0.9620 | 0.0000 | 0.9620 | 0.0160 |
| swap | 0.9802 | 0.0000 | 0.9802 | 0.0000 |
| digit | 0.0000 | 1.0000 | 1.0000 | 0.0000 |
| drop is omission: no check flags it; coverage reports it |
查看完整验算程序
# Chapter 4. p43: structure checks and hashes catch different edits.
# A derivative keeps a random 5 of the 8 master edges and all 8 claim
# texts (each with a provenance hash by anchor id). One random edit is
# applied, then two checks run: the structure check of p41 and the hash
# check (claim text under each id against the master hash).
# Edit types: drop an edge; reverse an edge; flip a label; add an edge
# between two random claims; swap two claims with their ids (positions
# exchange); slip a digit inside one claim text.
# Illustrative; fixed-seed Monte Carlo, 4000 trials per edit type.
import hashlib
import random
MASTER = [("C1", "C2", "supports"), ("C2", "C3", "precedes"),
("C3", "C4", "supports"), ("C4", "C2", "refines"),
("C5", "C6", "supports"), ("C2", "C8", "supports"),
("C7", "C8", "contradicts"), ("C1", "C5", "precedes")]
MSET = set(MASTER)
LABELS = ["supports", "precedes", "refines", "contradicts"]
IDS = ["C%d" % i for i in range(1, 9)]
TEXT = {c: "claim %s holds at 0.%d5" % (c, i) for i, c in
enumerate(IDS, 1)}
HASH = {c: hashlib.sha256(TEXT[c].encode()).hexdigest() for c in IDS}
TRIALS = 4000
def edit(rng, kind, edges, text):
if kind == "drop":
edges.pop(rng.randrange(len(edges)))
elif kind == "reverse":
i = rng.randrange(len(edges))
u, v, lab = edges[i]
edges[i] = (v, u, lab)
elif kind == "flip":
i = rng.randrange(len(edges))
u, v, lab = edges[i]
edges[i] = (u, v, rng.choice([x for x in LABELS if x != lab]))
elif kind == "add":
u, v = rng.sample(IDS, 2)
edges.append((u, v, rng.choice(LABELS)))
elif kind == "swap":
a, b = rng.sample(IDS, 2)
sw = {a: b, b: a}
edges[:] = [(sw.get(u, u), sw.get(v, v), lab)
for u, v, lab in edges]
elif kind == "digit":
c = rng.choice(IDS)
text[c] = text[c].replace("5", "6")
rng = random.Random(20261113)
print("%-8s %9s %7s %7s %9s" % ("edit", "structure", "hash", "either",
"harmless"))
for kind in ("drop", "reverse", "flip", "add", "swap", "digit"):
st = hs = ei = harmless = 0
for _ in range(TRIALS):
edges = rng.sample(MASTER, 5)
before = set(edges)
text = dict(TEXT)
edit(rng, kind, edges, text)
s_bad = any(e not in MSET for e in edges)
h_bad = any(hashlib.sha256(text[c].encode()).hexdigest()
!= HASH[c] for c in IDS)
if set(edges) <= MSET and set(edges) != before and \
kind != "drop":
harmless += 1
st += s_bad
hs += h_bad
ei += s_bad or h_bad
print("%-8s %9.4f %7.4f %7.4f %9.4f" % (
kind, st / TRIALS, hs / TRIALS, ei / TRIALS, harmless / TRIALS))
print("drop is omission: no check flags it; coverage reports it")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《母书多模态衍生编译与版权链、单部专著到播客投影片题库的结构同态流水线母典》第 4 章「结构同态:衍生品必须是母书论证图的同态像」(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-268 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026268,
author = {Fan, Yafu},
title = {Multimodal Derivative Compilation and Provenance Chains: A Structure-Preserving Pipeline from One Monograph to Podcasts, Slides and Quiz Banks},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}