15.1 和 15.2 的失败，伤的是钱——LTV 掉几个点、ROAS 跌破回本线，你还能改、还能救。15.3 的失败伤的是命：资质缺一项、文案踩一条红线、防沉迷漏放一次，轻则提审被打回、重则上线后被下架整改，一款已经跑起来的游戏一夜归零。这类失败最不像技术问题，却最不该靠「运气」——所以本节把它做成可跑的体检脚本。本站那篇独立开发者的手游广告失败复盘 里也提到素材与文案踩坑的段落，那是事后叙述；本节把「会不会踩坑」变成五个体检项，各跑一遍「健康配置 vs 失败配置」的对照，用缺陷量说话。
运营与合规的失败，为什么该做成「体检」
合规失败有一个反直觉的特点：它在事故前不可见，在事故后不可逆。你不会因为「暂时没被查到」就安全，也不会因为「这次过了审」就一劳永逸——平台规则、内容红线、防沉迷要求都在变。对抗这种不可见性，最务实的办法是把每一条要求翻译成一条可判定的规则，再让脚本对「健康配置」和「失败配置」各跑一遍，看缺陷量差多少。这不是替代法务判断，而是在提审前先把机器能查的部分查干净：真实资质要求、真实审核结果与真实合规判定需真实业务环境与专业法务判断，命令行不可复现（X）。
本节五个体检项覆盖合规与运营两侧：资质覆盖率（缺一项就提审失败）、内容红线（分级扫描 block / review）、防沉迷闸（时段 / 时长 / 消费三闸的漏放）、活动节奏空档（一年内最大空档周数）、奖励通胀（活动奖励相对常规产出的比例）。
S 层：五个体检项，健康 vs 失败对照（S 层）
下面这段是 S 层体检脚本，读 config/l15_ops_compliance.json，对五个体检项各跑一次「健康配置 vs 失败配置」，产出 out/l15_ops_compliance.{json,csv,svg}。每个体检项输出一个缺陷量（资质缺项数 / 红线 block 条数 / 防沉迷漏放次数 / 空档周数 / 通胀超限档），失败配置只要缺陷量 > 0 即判 FAIL。
#!/usr/bin/env python3
# file: sim/l15_ops_compliance.py
# tier: S
# verify: /opt/homebrew/bin/python3 sim/l15_ops_compliance.py --config config/ --out out/
"""运营与合规的失败模式体检：读 config/l15_ops_compliance.json。
五个体检项，各自跑「健康配置 vs 失败配置」对照，每个输出一个「缺陷量」：
1) 资质覆盖率 required = must（含 hasPaid 时的 must-if-paid）；缺陷 = 提审前缺项数。
2) 内容红线 分级扫描：任一 block 命中即 block，否则 review，最后 pass；缺陷 = block 条数。
3) 防沉迷 时段 / 时长 / 消费三闸；缺陷 = 该被拦却放行的次数（漏放）。
4) 活动节奏 一年内节点最大空档周数；缺陷 = max(0, 最大空档 - 目标空档)。
5) 奖励通胀 活动奖励 / 一周常规产出；缺陷 = max(0, 通胀‰ - 1000) // 100。
清单条目、红线词库、防沉迷规则、节点排期与奖励预算均为示例归纳；真实资质要求、真实审核结果与
真实合规判定需真实业务环境与专业法务判断，命令行不可复现（X）。全整数、地板除，无浮点。
产出 out/l15_ops_compliance.{json,csv,svg}。
"""
from __future__ import annotations
import argparse, csv, json, os, sys, xml.etree.ElementTree as ET
RANK = {"pass": 0, "review": 1, "block": 2}
GROUP_COLOR = {"合规": "#dc2626", "运营": "#f59e0b"}
def required_ids(lic: dict) -> list[str]:
return [it["id"] for it in lic["items"]
if it["level"] == "must" or (it["level"] == "must-if-paid" and lic["hasPaid"])]
def coverage(lic: dict, holdings: list[str]) -> dict:
req, held_set = required_ids(lic), set(holdings)
held = [i for i in req if i in held_set]
return {"required": len(req), "held": len(held),
"coveragePermille": len(held) * 1000 // len(req),
"missing": [i for i in req if i not in held_set]}
def scan_text(text: str, rules: list[dict]) -> list[dict]:
hay = text.lower()
return [{"ruleId": r["id"], "severity": r["severity"], "pattern": p}
for r in rules for p in r["patterns"] if hay.find(p.lower()) >= 0]
def decide(hits: list[dict]) -> str:
top = "pass"
for h in hits:
if RANK[h["severity"]] > RANK[top]:
top = h["severity"]
return top
def redline_defect(redline: dict, texts: list[dict]) -> dict:
rows = [{"id": t["id"], "kind": t["kind"], "decision": decide(scan_text(t["text"], redline["rules"]))}
for t in texts]
blocks = [r["id"] for r in rows if r["decision"] == "block"]
return {"blocks": blocks, "reviews": [r["id"] for r in rows if r["decision"] == "review"],
"blockCount": len(blocks)}
def gate_blocked(profile: dict, at: dict) -> bool:
"""返回 True 表示该次被防沉迷拦下（合规）；False 表示放行。"""
if at["dayKind"] not in profile["allowedDays"]:
return True
if not (profile["windowStart"] <= at["clock"] < profile["windowEnd"]):
return True
if at["durationMin"] > profile["dailyMinCap"]:
return True
return at["spendFen"] > profile["singleSpendCapFen"]
def antiediction_defect(cfg: dict, gate_on: bool) -> dict:
leaked = [at["id"] for at in cfg["attempts"]
if not (gate_blocked(cfg["profile"], at) if gate_on else False)]
return {"leakCount": len(leaked), "leaked": leaked}
def max_gap_weeks(nodes: list[int], year_weeks: int) -> int:
if not nodes:
return year_weeks
pts = [0] + sorted(nodes) + [year_weeks]
return max(pts[i + 1] - pts[i] for i in range(len(pts) - 1))
def inflation_permille(reward_fen: int, daily_faucet: int, event_days: int) -> int:
return reward_fen * 1000 // (daily_faucet * event_days)
def svg(path: str, checks: list[dict]) -> int:
"""纯 Python SVG：每个体检项一根横条 = 失败配置的缺陷量。"""
W, PAD, ROW = 760, 220, 42
H = 78 + ROW * len(checks)
r = ET.Element("svg", {"xmlns": "http://www.w3.org/2000/svg", "width": str(W),
"height": str(H), "viewBox": f"0 0 {W} {H}"})
ET.SubElement(r, "rect", {"width": str(W), "height": str(H), "fill": "#ffffff"})
for x, y, s, sz, fill in [(16, 26, "《合成小岛》运营 / 合规失败体检：失败配置缺陷量", 15, "#111827"),
(16, 46, "缺陷量口径：资质缺项 / 红线 block 条数 / 防沉迷漏放次数 / 空档周数 / 通胀超限档", 12, "#6b7280")]:
ET.SubElement(r, "text", {"x": str(x), "y": str(y), "font-size": str(sz), "fill": fill}).text = s
span = W - PAD - 120
hi = max(max(c["failureDefect"] for c in checks), 1)
for i, c in enumerate(checks):
y, w = 66 + i * ROW, span * c["failureDefect"] // hi
ET.SubElement(r, "text", {"x": "16", "y": str(y + 22), "font-size": "12",
"fill": "#374151"}).text = c["name"]
ET.SubElement(r, "rect", {"x": str(PAD), "y": str(y + 6), "width": str(w), "height": "22",
"rx": "4", "fill": GROUP_COLOR.get(c["group"], "#6b7280")})
ET.SubElement(r, "text", {"x": str(PAD + w + 8), "y": str(y + 22), "font-size": "12",
"fill": "#6b7280"}).text = f'{c["failureDefect"]} · {c["verdict"]}'
os.makedirs(os.path.dirname(path), exist_ok=True)
ET.ElementTree(r).write(path, encoding="utf-8", xml_declaration=True)
return os.path.getsize(path)
def build_checks(cfg: dict) -> list[dict]:
lic, red, cad, inf = cfg["license"], cfg["redline"], cfg["cadence"], cfg["eventInflation"]
hc, fc = coverage(lic, lic["healthyHoldings"]), coverage(lic, lic["failureHoldings"])
hr, fr = redline_defect(red, red["healthyTexts"]), redline_defect(red, red["failureTexts"])
ha, fa = antiediction_defect(cfg["antiediction"], True), antiediction_defect(cfg["antiediction"], False)
hg, fg = max_gap_weeks(cad["healthyNodes"], cad["yearWeeks"]), max_gap_weeks(cad["failureNodes"], cad["yearWeeks"])
hd, fd = max(0, hg - cad["targetGapWeeks"]), max(0, fg - cad["targetGapWeeks"])
hi = inflation_permille(inf["healthyEventRewardFen"], inf["dailyFaucetFen"], inf["eventDays"])
fi = inflation_permille(inf["failureEventRewardFen"], inf["dailyFaucetFen"], inf["eventDays"])
return [
{"id": "license", "group": "合规", "name": "资质覆盖率",
"healthy": f'{hc["coveragePermille"]}‰', "failure": f'{fc["coveragePermille"]}‰',
"healthyDefect": hc["required"] - hc["held"], "failureDefect": fc["required"] - fc["held"],
"verdict": "PASS" if fc["coveragePermille"] == 1000 else "FAIL",
"detail": f'提审前缺 {fc["required"] - fc["held"]} 项：{",".join(fc["missing"])}'},
{"id": "redline", "group": "合规", "name": "内容红线",
"healthy": f'{hr["blockCount"]} block', "failure": f'{fr["blockCount"]} block',
"healthyDefect": hr["blockCount"], "failureDefect": fr["blockCount"],
"verdict": "PASS" if fr["blockCount"] == 0 else "FAIL",
"detail": f'命中 block 的文案：{",".join(fr["blocks"]) or "无"}；review：{",".join(fr["reviews"]) or "无"}'},
{"id": "antiediction", "group": "合规", "name": "防沉迷闸",
"healthy": f'{ha["leakCount"]} 漏放', "failure": f'{fa["leakCount"]} 漏放',
"healthyDefect": ha["leakCount"], "failureDefect": fa["leakCount"],
"verdict": "PASS" if fa["leakCount"] == 0 else "FAIL",
"detail": f'未接入防沉迷时放行的违规尝试：{",".join(fa["leaked"])}'},
{"id": "cadence", "group": "运营", "name": "活动节奏空档",
"healthy": f'{hg} 周', "failure": f'{fg} 周',
"healthyDefect": hd, "failureDefect": fd, "verdict": "PASS" if fd == 0 else "FAIL",
"detail": f'最大空档 {fg} 周 > 目标 {cad["targetGapWeeks"]} 周，超出 {fd} 周'},
{"id": "inflation", "group": "运营", "name": "活动奖励通胀",
"healthy": f'{hi}‰', "failure": f'{fi}‰',
"healthyDefect": max(0, hi - 1000) // 100, "failureDefect": max(0, fi - 1000) // 100,
"verdict": "PASS" if fi <= 1000 else "FAIL",
"detail": f'活动奖励 = 一周常规产出的 {fi}‰，超过 1000‰ 即奖励通胀'}]
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--config", default="config/")
ap.add_argument("--out", default="out/")
a = ap.parse_args()
with open(os.path.join(a.config, "l15_ops_compliance.json"), encoding="utf-8") as f:
cfg = json.load(f)
checks = build_checks(cfg)
fails = [c for c in checks if c["verdict"] == "FAIL"]
os.makedirs(a.out, exist_ok=True)
sb = svg(os.path.join(a.out, "l15_ops_compliance.svg"), checks)
report = {"version": cfg["version"], "game": cfg["game"], "observedAt": cfg["observedAt"],
"note": cfg["note"], "checkCount": len(checks), "checks": checks,
"failCount": len(fails), "failIds": [c["id"] for c in fails], "svgBytes": sb}
with open(os.path.join(a.out, "l15_ops_compliance.json"), "w", encoding="utf-8") as f:
json.dump(report, f, ensure_ascii=False, indent=2)
cols = ("id", "group", "name", "healthy", "failure", "healthyDefect", "failureDefect", "verdict", "detail")
with open(os.path.join(a.out, "l15_ops_compliance.csv"), "w", encoding="utf-8", newline="") as f:
w = csv.writer(f)
w.writerow(cols)
for c in checks:
w.writerow([c[k] for k in cols])
for c in checks:
print(f'[{c["verdict"]}] {c["group"]} {c["name"]:<8} 健康 {c["healthy"]:<10} '
f'失败 {c["failure"]:<10} 缺陷 {c["healthyDefect"]}→{c["failureDefect"]}')
print(f'体检结论：{len(fails)}/{len(checks)} 项 FAIL {report["failIds"]}')
return 0
if __name__ == "__main__":
sys.exit(main())
真跑输出（同机）：
[FAIL] 合规 资质覆盖率 健康 1000‰ 失败 250‰ 缺陷 0→6
[FAIL] 合规 内容红线 健康 0 block 失败 2 block 缺陷 0→2
[FAIL] 合规 防沉迷闸 健康 0 漏放 失败 3 漏放 缺陷 0→3
[FAIL] 运营 活动节奏空档 健康 6 周 失败 26 周 缺陷 0→20
[FAIL] 运营 活动奖励通胀 健康 285‰ 失败 2142‰ 缺陷 0→11
体检结论：5/5 项 FAIL ['license', 'redline', 'antiediction', 'cadence', 'inflation']
D 层：五组检查的输入清单（D 层）
脚本的输入是一份 D 层配置：九项资质清单（标 must / must-if-paid / should）、八条内容红线（分 block / review 两级）、一份防沉迷画像与三条违规尝试、一张 52 周节点排期表、以及活动奖励预算。这些条目都是示例归纳，真实平台清单会因品类与地区而异——但结构是通用的：把「要求」写成条目，把「违规样本」写成输入，让脚本替你数缺陷。
{
"version": 1,
"game": "merge-island",
"observedAt": "2026-10",
"note": "运营与合规的失败模式体检（D 层）。把「资质覆盖率 / 内容红线 / 防沉迷 / 活动节奏 / 奖励通胀」五个体检项，各自跑一遍「健康配置 vs 失败配置」的对照。清单条目、红线词库、防沉迷规则、节点排期与奖励预算均为示例归纳；真实资质要求、真实审核结果与真实合规判定需真实业务环境与专业法务判断，命令行不可复现（X）。全整数、禁浮点。",
"license": {
"hasPaid": true,
"items": [
{ "id": "entity-license", "name": "主体营业执照", "level": "must" },
{ "id": "soft-copyright", "name": "软件著作权登记", "level": "must" },
{ "id": "icp-filing", "name": "ICP 备案", "level": "must" },
{ "id": "game-approval", "name": "游戏版号（含内购时）", "level": "must-if-paid" },
{ "id": "category-permit", "name": "平台类目资质", "level": "must" },
{ "id": "privacy-policy", "name": "隐私政策与用户协议", "level": "must" },
{ "id": "realname-antiaddiction", "name": "实名与防沉迷接入", "level": "must" },
{ "id": "pay-qualify", "name": "支付 / 结算资质", "level": "must-if-paid" },
{ "id": "minor-protection-report", "name": "未成年人保护专项报告", "level": "should" }
],
"healthyHoldings": ["entity-license", "soft-copyright", "icp-filing", "game-approval",
"category-permit", "privacy-policy", "realname-antiaddiction", "pay-qualify"],
"failureHoldings": ["entity-license", "soft-copyright"]
},
"redline": {
"rules": [
{ "id": "r-political", "category": "涉政", "severity": "block", "patterns": ["领导人", "敏感事件"] },
{ "id": "r-porn", "category": "涉黄", "severity": "block", "patterns": ["色情", "裸聊"] },
{ "id": "r-violence", "category": "涉暴", "severity": "block", "patterns": ["血腥", "恐怖袭击"] },
{ "id": "r-gambling", "category": "赌博", "severity": "block", "patterns": ["现金红包", "押注", "真钱"] },
{ "id": "r-ip", "category": "侵权", "severity": "block", "patterns": ["宝可梦", "马里奥", "迪士尼"] },
{ "id": "r-superstition", "category": "迷信", "severity": "review", "patterns": ["转运", "开光"] },
{ "id": "r-induce-share", "category": "诱导分享", "severity": "review", "patterns": ["分享给", "转发到"] },
{ "id": "r-false-ad", "category": "虚假宣传", "severity": "review", "patterns": ["稳赚不赔", "一键暴富"] }
],
"healthyTexts": [
{ "id": "h1", "kind": "游戏名", "text": "合成小岛" },
{ "id": "h2", "kind": "买量文案", "text": "拖动相同岛屿即可合成更大的岛屿" },
{ "id": "h3", "kind": "活动文案", "text": "登录领取限定装扮" }
],
"failureTexts": [
{ "id": "f1", "kind": "游戏名", "text": "合成小岛：一键暴富，稳赚不赔" },
{ "id": "f2", "kind": "素材标题", "text": "官方授权宝可梦合成玩法" },
{ "id": "f3", "kind": "活动文案", "text": "幸运大转盘，赢取现金红包和转运符" }
]
},
"antiediction": {
"profile": { "band": "8-16 周岁", "allowedDays": ["fri", "sat", "sun"],
"windowStart": "20:00", "windowEnd": "21:00", "dailyMinCap": 60,
"singleSpendCapFen": 5000, "monthlySpendCapFen": 20000 },
"attempts": [
{ "id": "a-school-night", "clock": "20:30", "dayKind": "school", "durationMin": 10, "spendFen": 0 },
{ "id": "a-over-daily", "clock": "20:30", "dayKind": "sat", "durationMin": 120, "spendFen": 0 },
{ "id": "a-over-spend", "clock": "20:30", "dayKind": "sat", "durationMin": 5, "spendFen": 6000 }
]
},
"cadence": {
"yearWeeks": 52,
"targetGapWeeks": 6,
"healthyNodes": [2, 8, 14, 20, 26, 32, 38, 44, 50],
"failureNodes": [2, 3, 4, 5, 6, 24, 25, 26]
},
"eventInflation": {
"dailyFaucetFen": 200,
"eventDays": 7,
"healthyEventRewardFen": 400,
"failureEventRewardFen": 3000
}
}
五个体检项分别意味着什么
资质覆盖率：健康配置八项齐全、覆盖率 1000‰；失败配置只持有营业执照与软著两项，覆盖率掉到 250‰，提审前缺六项。这不是「慢慢补」的事——缺 game-approval（版号）会让含内购的版本直接无法上架，缺 realname-antiaddiction（实名与防沉迷）会被平台强制下架。
内容红线：失败配置三条文案里两条命中 block——「一键暴富、稳赚不赔」踩虚假宣传、「官方授权宝可梦」踩 IP 侵权；还有一条命中 review（现金红包 + 转运符）。红线扫描的价值在于分级：block 是必须改掉的硬伤，review 是提审前人工复核的灰区。
防沉迷闸：健康配置下三条违规尝试（工作日登录、超时长、超单笔消费）全被拦；一旦未接入防沉迷，三次全部放行——3 漏放。防沉迷不是「可选优化」，是提审硬门槛，漏放一次就是合规事故。
活动节奏空档：健康排期九场活动把最大空档压到 6 周（正好等于目标）；失败排期前五场挤在第 2–6 周、之后一路空到第 24 周——最大空档 26 周，玩家有半年没新内容，直接掉线。
奖励通胀：健康活动的奖励等于一周常规产出的 285‰（约三成，合理）；失败活动是 2142‰（两周多），奖励通胀 11 档——玩家拿活动奖励就能跳过成长曲线，长期产出与付费动机被一起摧毁。
本章小结
三节下来，失败被拆成了三层：玩法 / 数值 / 变现（15.1，伤钱）、买量 / 留存（15.2，伤钱且难救）、运营 / 合规（15.3，要命）。它们的共同点是——都能被写成参数、喂进模型、跑出坏数字。避坑的正确姿势从来不是背一份「不要做什么」的清单，而是把每条失败做成一个可复现的对照：改一个数，跑一遍，看它掉几个点。这样你在真正投放、真正提审之前，就已经见过它翻车的样子。
下一章是全书的收口：把三卷的所有模型装配成一份可交付的上线方案，跑一次全链路仿真验收。
阅读导航：上一节：15.2 买量与留存的典型失败 · 下一节：16.1 装配完整上线方案 。
继续阅读
探索更多技术文章
浏览归档，发现更多关于系统设计、工具链和工程实践的内容。