关于「实证闭环:素材疲劳与混剪台账、实战工具与检查器规格」,《素材疲劳与正交混剪》给出的定量结论是什么?

结论:第八章给出素材疲劳与混剪台账、渲染命名流水线、蒸馏小工具「素材疲劳半衰期与正交混剪排程试算器」、按四件套分组的检查器规格与预注册检验。素材只用自有或已授权镜头,零伪造证言与前后对比,AI 合成画面按平台现行规则如实标注。书中以标注框区分已发表的证据、可运行的 Python 计算(22 段程序)、原典与作者的假说。

数据口径:买量圈流行的说法是:扩量期素材衰退得快,因为观众重复曝光使点击率按可估计的半衰期衰减;把短视频拆成开头钩子、痛点、证据、行动召唤四个模块,用正交表组合,十几个镜头就能自动合成上百条低相关的测试素材,突破产能瓶颈。 书中参数为示意,全部计算由书中程序生成。

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

idgenes freq caphlevelrulesmanifest
c01 00003.154.2 GREEN-9a6ba7f8962f
c02 01124.354.0 YELLOWF16deba194c7fb
c03 02215.643.9 REDF17cb5bd6aa68a
c04 10112.432.1 REDF2830e42da4c8a
c05 11202.054.4 REDF75bace3f826fe
c06 12022.254.1 REDF46a585524cb98
c07 20222.554.6 REDF5bc5aff4b2d92
c08 21011.854.0 REDF3,F6184d7b92a6ce
查看完整验算程序
# Chapter 8. p81: red / yellow / green rules on a creative ledger
# Eight illustrative ledger rows (no real account). Rules:
#  F1 frequency above cap -> red; at 0.8 of cap or more -> yellow
#     cap = 1 + floor(h * log2(c0 V / m)), c0 0.012, V 2.0, m 0.012
#  F2 half-life upper bound below 0.7 x last month -> red;
#     point estimate below 0.85 x last month -> yellow
#  F3 licence missing or expired -> red; 14 days or less -> yellow
#  F4 AI-synthetic footage without platform label -> red
#  F5 "winner" declared without FDR correction -> red
#  F6 proof shot not verified as real (or not marked illustrative) -> red
#  F7 manifest hash differs from the registered hash -> red
import hashlib
import math

C0, V, M = 0.012, 2.0, 0.012


def cap(h):
    return 1 + math.floor(h * math.log2(C0 * V / M))


def manifest(r):
    keys = ("id", "genes", "lic", "ai", "label", "proof")
    text = "|".join(str(r[k]) for k in keys)
    return hashlib.sha256(text.encode("ascii")).hexdigest()[:12]


ROWS = [
    dict(id="c01", genes="0000", freq=3.1, h=4.2, lo=3.6, hi=4.9, prev=4.0,
         lic=120, ai=False, label=False, win=False, fdr=False, proof=True),
    dict(id="c02", genes="0112", freq=4.3, h=4.0, lo=3.5, hi=4.6, prev=4.1,
         lic=60, ai=False, label=False, win=False, fdr=False, proof=True),
    dict(id="c03", genes="0221", freq=5.6, h=3.9, lo=3.3, hi=4.4, prev=4.0,
         lic=90, ai=False, label=False, win=False, fdr=False, proof=True),
    dict(id="c04", genes="1011", freq=2.4, h=2.1, lo=1.7, hi=2.6, prev=4.0,
         lic=90, ai=False, label=False, win=False, fdr=False, proof=True),
    dict(id="c05", genes="1120", freq=2.0, h=4.4, lo=3.8, hi=5.0, prev=4.3,
         lic=9, ai=False, label=False, win=False, fdr=False, proof=True),
    dict(id="c06", genes="1202", freq=2.2, h=4.1, lo=3.4, hi=4.8, prev=4.0,
         lic=200, ai=True, label=False, win=False, fdr=False, proof=True),
    dict(id="c07", genes="2022", freq=2.5, h=4.6, lo=4.0, hi=5.3, prev=4.4,
         lic=150, ai=False, label=False, win=True, fdr=False, proof=True),
    dict(id="c08", genes="2101", freq=1.8, h=4.0, lo=3.2, hi=4.9, prev=3.9,
         lic=None, ai=True, label=True, win=False, fdr=False, proof=False),
]
REG = {r["id"]: manifest(r) for r in ROWS}
ROWS[4]["lic"] = 45          # licence edited after registration


def judge(r):
    red, yel = [], []
    c = cap(r["h"])
    if r["freq"] > c:
        red.append("F1")
    elif r["freq"] >= 0.8 * c:
        yel.append("F1")
    if r["hi"] < 0.7 * r["prev"]:
        red.append("F2")
    elif r["h"] < 0.85 * r["prev"]:
        yel.append("F2")
    if r["lic"] is None or r["lic"] < 0:
        red.append("F3")
    elif r["lic"] <= 14:
        yel.append("F3")
    if r["ai"] and not r["label"]:
        red.append("F4")
    if r["win"] and not r["fdr"]:
        red.append("F5")
    if not r["proof"]:
        red.append("F6")
    if manifest(r) != REG[r["id"]]:
        red.append("F7")
    level = "RED" if red else ("YELLOW" if yel else "GREEN")
    return c, level, red + yel


print("id  genes freq cap  h    level   rules      manifest")
for r in ROWS:
    c, level, why = judge(r)
    print("{} {} {:>4.1f} {:>3} {:>4.1f} {:<7} {:<10} {}".format(
        r["id"], r["genes"], r["freq"], c, r["h"], level,
        ",".join(why) or "-", manifest(r)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《投放素材频次疲劳半衰期、分镜基因正交重组与百条广告全自动混剪母典》第 8 章「实证闭环:素材疲劳与混剪台账、实战工具与检查器规格」(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-400 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026400,
  author    = {Fan, Yafu},
  title     = {Ad-Creative Fatigue Half-Lives, Orthogonal Storyboard Recombination and Automated Permutation Rendering},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论