O!Hole 经济与活动调整 · 第二版

2026-08-19 校准窗口 08-07~08-14｜周产出 119.97M、周消耗 96.81M

一次改版内完成两层调整：全量活动层（公会锦标赛 × 星星收集周末轮替，全量更新、不进任何 A/B）与玩家级经济 A/B（削减整包只进实验组）。

全量更新 · 不进 A/B

全量活动层
公会锦标赛（新）× 星星收集（现状不动）：周末隔周轮替；周一~周四公会宝箱完全不动

玩家级 A/B

A ＝新活动基线＋当前经济配置
B ＝同基线＋削减包＋水晶冲刺＋搏击改版＋结算广告×2＋新手保护

实验组净削减

−16.2~−17.3%

−19.49M ~ −20.74M / 周

新玩家 D0–6

−12.3~−13.4%

早期新手 −2.3% ｜进阶新手 −18.4~−20.1%

老玩家 D7+

−18.6~−19.6%

承担削减主体

静态产耗比

1.02~1.04

消耗不变口径；动态值待 A/B 验证

金币总账（瀑布）

红＝削减通道；绿＝B 版本内回补；深蓝＝实验组净额。活动层（−0.21~−0.40pp）为全服共同变化，不进此瀑布。

调整主项一览

位置	修改内容
关卡结算	星级基础 20/25/30 → 15/20/25（全段统一，难度倍率与关内收集不动）
火山寻宝	改纯门票制：取消每日免费票与每日恢复；剧情章节券 21 → 10 张；兑换维持 6
金币塔	首次达标数 3 → 6（3,333 → 1,667）＋连胜要求 7 → 10 场
每日任务	高档 −40%，基础档不动
夺宝活动	中段 −25%，顶档 7,000/8,000 不动
剧情章节奖励	第 4 章起 −25%；前 3 章 ×2（120/100/180 → 240/200/360；重合处以翻倍为准）
通行证	免费轨金币 −30%，付费轨与末奖不动
结算广告	固定加成 +40/36/32 → （星级基础 × 难度倍率）× 2
广告复活	每日广告复活 1 → 3 次；当日第 2、3 次仅在金币不够当次复活价时出现；广告复活不推进 900/1800/3600 阶梯
新手保护	签到组 1 +40% ＋前 3 章剧情章节 ×2

同版本给玩家的

三个新活动
水晶冲刺（每周 60 分钟无限体力＋卡包，零金币）；新搏击俱乐部（金币奖励 51→102，+90%）；公会锦标赛（新团队玩法，周末档）。

规则升级
结算广告改翻倍：高难关看广告收益比现在更高（困难 3★ 92→100、Boss 3★ 122→150），且第一次做到「打得好的人拿得多」——现行规则是反的（1★ 加 40、3★ 只加 32）。广告复活每日 1 → 3 次，当日第 2、3 次仅在金币不够时出现，且不推进金币复活阶梯。

以下改动均只在实验组配置生效，对照组维持现网值。

关卡结算−5.06M

星级基础 20/25/30 → 15/20/25（全关卡统一，不分段，仅实验组生效）；难度倍率（困难×2/Boss×3）与关内收集金币不动 → 通道 −16.5%。

配置落点：Level.xlsm / LevelReward：星级基础 1★15／2★20／3★25，仅实验组配置生效（对照组维持 20/25/30）；Global 关卡倍率「2;3」不动

火山寻宝−7.13M

改纯门票制：取消每日免费票与每日恢复，入场完全依赖剧情章节券；章节券 21 → 10 张；银币兑换维持 6 不动。该通道原有 71.7% 的局来自每日免费票，取消后产出降至约 1.17M。

配置落点：Challenge.xlsm / ChallengeDefine：每日免费票与每日恢复置 0，exchange 维持 6；Build.xlsm 奖励列的火山券（道具 119）21 张 → 10 张。上线前置：add_cur 不封顶与跨自然日不重置（>24h 判断）两个缺陷须先修，且对照组同样生效——修复前 1.17M 只是稳态模型值。

金币塔−3.51M ~ −4.76M

首次达标数 3 → 6（首玩奖励 3,333 → 1,667）＋连胜要求 7 → 10 场。奖池 10,000 不动，最终存活人数目标 8~12 人。该项为区间估算——金币塔只有发奖埋点、没有参与埋点，连胜门槛抬高后的淘汰效应无法实测。

配置落点：ActivityBattleRoyal.xlsm：first_reward 3 → 6（机器人达标配置联动）；连胜场次要求 7 → 10

每日任务·高档−3.15M

高档 −40%；基础档一分不动。实测该档池 7.88M，其中 20.3% 落在 L1–30 段——新玩家并非完全够不着。

配置落点：Task.xlsm / DailyTaskReward：高档＝101≥400 的 6 行——id8 420→250、id14 500→300、id15 800→480、id33 500→300、id43 400→240、id44 700→420；基础档 8 行不动

夺宝活动·中段−2.14M

中段 −25%，顶档 7,000/8,000 与前段甜档原样保留。实测中段池 8.56M。

配置落点：TreasureHunt.xlsm / TreasureHuntReward：1001：档2 100 保留、档7 200→150、档11 300→225、档16 500→375、档25 7,000 不动；1003：档2 100 保留、档7 200→150、档11 500→375、档17 1,000→750、档25 8,000 不动

剧情章节奖励净 +0.88M

前 3 章 ×2（Build 现基数 120/100/180 → 240/200/360，+1.38M）作新手补偿；第 4 章起 −25%（−0.50M）。两段有重合，以翻倍为准——第 3 章对应 L34–54，整章都在 L31 之上，这部分只翻倍、不叠加 ×0.75。前 3 章是本轮唯一加码的回补项：实测池 1.38M（按各章面额计数，120 以章 4 起的 L59 为上限、180 以章 9 起的 L184 为上限），其中 83.3% 直接落到 D0–6 新玩家，是所有可用回补工具里效率最高的一个。

配置落点：Build.xlsm 奖励列：章 1–3（对应 L4–54）金币 120/100/180 → 240/200/360；章 4 起（L59 及以上）×0.75。章节与关卡对照：章1＝L4/9/14、章2＝L19/24/29、章3＝L34/39/44/49/54。分支已实装 ×2。（LevelChest 为已废弃的旧功能，不在本次范围）

通行证·免费轨−0.83M

免费轨金币 −30%；付费轨与最终大奖不动。实测该通道在 L1–30 段产出为 0，不触及早期新手。

配置落点：BattlePass.xlsm（涂鸦 id=1，31 档）：免费轨金币 6 档 100→70、档30 200→140、档31 300→210；末奖 3,000 不动

结算广告 ×2（abtest 分支已实装）−0.53M

规则：看广告＝（星级基础 × 难度倍率）× 2。实装路径是服务端开关，不是配表：LevelMechanism.ads_reward_type 对照组 1、实验组 2；服务端读到 2 时把结算金币并入广告奖励，客户端结算面板显示「×2」。不新增 LevelAds 行、不切 Difficulty.reward_ads 指针。关内收集不参与翻倍（实装与此一致）；免广告仅免结算插屏，激励广告保留——双倍必须实际观看获得。广告价值向高难度倾斜，修正「星级越高加得越少」的反向曲线：困难 3★ 100（现 92）、Boss 3★ 150（现 122）都比现在更高。待核一项：金币 ×2 buff 生效时，广告是否按已 buff 金额再翻倍（×4）——取决于服务端并入 ads_items 时取的是 buff 前还是 buff 后的值。

难度	1★	2★	3★	现行（固定加成后）
通常 ×1 → 广告后	30	40	50	60 / 61 / 62
困难 ×2 → 广告后	60	80	100	80 / 86 / 92
Boss ×3 → 广告后	90	120	150	100 / 111 / 122

上表为看广告后的总额，由双倍规则随结算基础自动得出，不需要逐档填表。

配置落点：LevelMechanism.xlsm：ads_reward_type = 1（对照组）／ads_reward_type_11b = 2（实验组）；服务端 Stage:GetAdsRewardTypeAbOpen == 2 时金币并入 ads_items；客户端提交 f101fa69（结算按钮显示 ×2）。广告按钮覆盖面维持现状不扩大。基准：实测单次额外奖励均值 38.0 → 新值 24.4。

广告复活扩展（abtest 分支已实装）0（不产金币）

现状有两层限制：Global.revive_ad = [1,0,0]（本局第一次复活可看广告）＋ Ads 表 id 101 的每日上限 1 次。实装改三处：① 每日上限 1 → 3（ads_limit_times_11b）；② 新增 revive_ad_check_gold = [0,1,1]，按当日已看广告复活次数取值——第 1 次不查金币，第 2、3 次仅当金币不够当次复活价时才显示广告按钮，够钱的直接走金币；③ revive_times_11b = 0，广告复活独立计数、不推进 900/1800/3600 阶梯。这样有钱的玩家不会用广告替代金币复活——复活占全部金币消耗的 93.4%，是最不能被蚕食的口。增量硬上限＝每人每日 +2 次，实际增量待 A/B 读数。

配置落点：Ads.xlsm id 101：ads_limit_times 1 → ads_limit_times_11b 3；Global.xlsm 新增 revive_ad_check_gold [0,1,1]、revive_times_11b 0；revive_ad [1,0,0] 与 revive_gold [900,1800,3600] 不动。服务端广告复活与金币复活分计（提交 83a061c2、f7253795、2859bac7）。

新手保护+3.07M

签到组 1 +40%（+1.69M）＋前 3 章剧情章节 ×2（+1.38M）。口径提醒：签到组 1 ＝前 7 次签到，按签到次数判定、与关卡和账号天数都无关；索引走完 21 行会循环回组 1，所以老玩家也会再吃到加码。实测（用索引 21 独有的 250 金币当探针，每周 1,081 人走完一轮）：组 1 总池 4.22M ＝新玩家 3.20M ＋老玩家 1.03M，故加码 76% 到新玩家、24%（0.41M）漏给老玩家——占老玩家承受额的 2.8%，可接受；要做到 100% 精准需按 create_days 门控，属服务端开发，本次不做。对照：前 3 章效率 83.3%，但基础池仅 1.38M，撑不起全部回补量。

配置落点：Sign.xlsm 组 1（7 行，现金币 200/100/100/200/0/150/200）整体 ×1.4；Build.xlsm 前 3 章金币 ×2

配置附表（全部改动行）

表	配置项	当前值	新值	状态
Level / LevelReward	星级基础 1★/2★/3★	20 / 25 / 30	15 / 20 / 25	已定
Global	关卡倍率（困难;Boss）	2;3	不变	已定
Challenge / ChallengeDefine	每日免费票 / 每日恢复	开启	关闭（纯门票制）	已定
Challenge / ChallengeDefine	exchange	6	不变	已定
Build	火山券（道具 119）	21	10	已定
Challenge / Challenge_GS	add_cur 封顶＋跨日重置	缺陷	修复（两组共同）	上线前置
ActivityBattleRoyal	first_reward	3	6	已定
ActivityBattleRoyal	连胜场次要求	7	10	已定
Task / DailyTaskReward	高档＝101≥400 共 6 行	420/500/800/500/400/700	250/300/480/300/240/420	已定
TreasureHunt / Reward	1001/1003 中段金币	200/300/500 ＋ 200/500/1000	150/225/375 ＋ 150/375/750	已定
TreasureHunt / Reward	顶档与前段甜档	7,000 / 8,000 / 100	不变	已定
Build	章 1–3 金币（L4–54）	120 / 100 / 180	240 / 200 / 360	已实装
Build	章 4 起金币（L59+）	现值	×0.75（与前 3 章不重叠）	已定
BattlePass (id=1)	免费轨金币（8 档）	100×6 / 200 / 300	70×6 / 140 / 210	已定
LevelMechanism	ads_reward_type / ads_reward_type_11b（结算广告 ×2）	1 / —	1 / 2	已实装
Ads (id 101)	广告复活每日上限 ads_limit_times / _11b	1 / —	1 / 3	已实装
Global	revive_ad_check_gold（当日第 2、3 次需金币不足）	—	[0, 1, 1]	已实装
Global	revive_times_11b（广告复活不推进阶梯）	—	0	已实装
Global	revive_ad / revive_gold	[1,0,0] / [900,1800,3600]	不变	已定
Sign	组 1（7 行金币）	200/100/100/200/0/150/200	整体 ×1.4	已定
Schedule	星星 5001 间隔	7D	14D	已定
Schedule	锦标赛排期行	—	新增 14D、错 7 天	新增

状态说明：已实装＝已在 feature/dev_abtest_20260818 分支核到实装（tip f1528b5，2026-08-21 拉取）；已定＝数值定稿、待进分支；上线前置＝两组共同修复；新增＝新功能配置。

水晶冲刺＋新搏击俱乐部（只进实验组，随经济 A/B 判定）

水晶冲刺（周一）
普通/高级各 30min、各发一次 114×30（无限体力）；失败即终止；奖励零金币：普通胜＝银卡包＋超闪＋加时，高级胜＝金卡包＋变大＋重力＋超闪＋指针；全胜漏斗 11.4%。数值照 0723 测试稿，零改动。

新搏击俱乐部（周三）
8 阶段、目标分 18–47 冻结；仅 2/5/8 胜发奖：100＋铜包＋加时 / 150＋金包＋超闪＋重力 / 300＋红宝石包＋变大＋超闪×2＋重力；金币 EV 51→102（+90% 明示加码）。替换旧搏击周三槽位。

以下两项为全量更新，不进 A/B：公会锦标赛与星星收集的周末轮替对全服统一生效，是 A/B 两个版本的共同基线。

公会锦标赛

组队与匹配：房间在某公会首次产生积分时创建，创建时固定加入 10 个 AI 公会；真人公会在首次产生积分时进入房间（此时校验解锁成员数 ≥10、读取公会总星数），每房至多 20 个公会，后到者进入新房间（最后一个房间可能不满员）。按公会总星数四档（0–1 万 / 1–3 万 / 3–8 万 / 8 万+）选用对应 AI 行为参数，决定名次难度与预算；AI 参数开赛前冻结，期间不作调整。

代币与计分：关卡内生成 3 种「头盔」收集物（每关合计 3~7 个），吸入即得代币；代币推进个人阶段节点，开箱获得活动积分（add_score），积分计入公会总分——公会排名由开箱积分驱动，不是通关数。

个人阶段奖励内容

节点	阶段内代币	活动累计代币	公会积分	奖励内容
1-1	10	10	10	铜卡包×1
1-2	20	20	15	超级闪电×1
1-3	40	40	25	无限体力 15 分钟
1-4	100	100	40	金卡包×1＋称号「锦标骑士」
2-1	60	160	20	加时闹钟×1
2-2	80	180	30	金币 50
2-3	100	200	40	变大×1
2-4	250	350	80	红宝石卡包×1＋称号「锦标大骑士」
循环	每 100	350+100n	30	变大×1（固定；服务端循环奖励为单一配置，不支持轮换）

奖励分配原则：金币集中于公会排名奖励；个人阶段以道具与卡包为主；公会内贡献奖励为单一道具。阶段道具均为 ×1，×2 仅用于公会内贡献奖励（与 RM 实录一致）；限时道具仅 1-3 一处；无限超闪／无限加时不投放。卡包按稀有度随关键节点递进：铜（入门）→ 金（阶段一完成）→ 红宝石（阶段二完成）。

公会排名奖励（组内前 3，有效贡献成员同档同额）与公会内贡献奖励（前 3，可叠加）

榜单	名次	奖励内容
公会排名	1	金币 2,000＋金卡包×1
	2	金币 1,200
	3	金币 600
公会内贡献	1	加时闹钟×2
	2	超级闪电×2
	3	重力增幅×2

公会排名奖励以金币为主，仅第 1 名附金卡包×1（夺冠仪式感）；第 2/3 名为纯金币，卡包等稀有资源不进入其余全员发放口径。公会内贡献奖励为单一道具、按道具价值分档。

预算：金币 ≈68 万/期（公会排名 ~62 万＋个人阶段 ~6 万）≈ 34 万/周 ≈ +0.28pp，走廊内。治理规则：AI 参数开赛前冻结、期间不调；某期超出预算走廊时，自下一期调整奖励数额。上线前输出预算假设表（各星数档真人夺榜率、各名次有效成员数、节点到达人数）供复算；道具与卡包产出并入周度道具管线跟踪。卡包：个人阶段三处——铜（1-1）／金（1-4，周均 ~1,150，上线首期盯收集册集齐速度）／红宝石（2-4，周均 ~325）；另公会榜第 1 名附金卡包（随真人夺榜人数，量小）；第 2/3 名、内部榜与循环零包。限时道具仅 1-3 无限体力 15min 一处。

排期：周五 08:00 → 周一 08:00（72H）、双周一期，与星星收集错开 7 天（星星排期 5001 间隔 7D→14D）；排期行随版本配置核对。

星星收集：本次不修改

奖励（1,500/750/400/300＋红宝石/金/银/铜包）与匹配均维持现状——轮替本身已是变化项，同期修改奖励会使两类周末无法对比。当前线上数据：冠军占获奖用户 ~74%、每期 ~657 名冠军、结算触达 ~3,500 人；冠军每期 70% 轮换、玩家年龄中位 16 天（后续待追踪问题）。

轮替说明：轮替的目的：为玩家提供有新鲜感的周末内容，并分别观察两个活动对玩家的影响，作为后续活动编排的依据（不涉及活动去留）。观察方法：主比较用两活动共同的 48H 核心窗口（周六 08:00~周一 08:00），按每可曝光玩家/每活跃玩家标准化，相邻两周配对；前 2 组配对做运行验收，之后持续跟踪两类周末在活跃、留存、变现与投放上的差异。

周末轮替排期

周	周中（不变）	周末（轮替）
第1 周	公会宝箱（周一 08:00 ~ 周五 08:00，照常）	公会锦标赛（周五 08:00 ~ 周一 08:00）
第2 周	公会宝箱（周一 08:00 ~ 周五 08:00，照常）	星星收集（周六 08:00 起 48H）
第3 周	公会宝箱（周一 08:00 ~ 周五 08:00，照常）	公会锦标赛（周五 08:00 ~ 周一 08:00）
第4 周	公会宝箱（周一 08:00 ~ 周五 08:00，照常）	星星收集（周六 08:00 起 48H）

夺宝周末场与赛马每个周末照常在跑——非公会玩家的周末内容不受影响；锦标赛周末对未入会玩家显示入会引导。

① 进阶新手（本轮风险最集中的一格）
口径：注册 0–6 天且已推进到 L31+。这批人净变化 −18.4%~−20.1%，与老玩家同档甚至更深——因为火山、金币塔、通行证对他们已全部解锁。盯：该人群的 D7 留存、余额 <600 占比、人均主线通关数。为什么单列：新玩家 D0–6 整体 −12.3~−13.4% 是个平均数，内部差异极大——仍在 L1–30 的早期新手只有 −2.3%，风险几乎全在这一格。

需要重点关注的情况：该人群余额 <600 占比超过 45%，或 D7 留存相对对照组掉超过 3pp → 触发阀 1：火山章节券由 10 张回调至 14 张（约回补 1.5M）。

② 新玩家整体与早期新手
D1 日终低余额占比（约 600 为参考线）；失败时买不起复活（<900）的占比；新手保护的触达与使用率；金币塔首玩奖励 3,333→1,667 后的首日行为。口径说明：约 600＝历史留存危险参考线；900＝首次复活价格；900/1800/3600 为同局递增价。

需要重点关注的情况：实验组的低余额或买不起复活的占比相对 A 明显上行；或保护机制触达率低（该拿到广告复活的人没拿到）。

③ 留存与深度
D1/D3/D7（新玩家与老玩家分开看）；人均主线通关数；会话时长与日局数。

需要重点关注的情况：新玩家 D1/D3 连续数日低于对照组，或人均通关下降集中在 L31–100 段（削减主落区）——优先核查新手保护触达率与该段的金币/道具缺口。

④ 变现
D7 LTV（广告与 IAP 分列）；金币复活购买次数与金币礼包销量；分难度的广告观看率——广告新规的专属信号（通常关观看率可能下降、高难关应上升，总量是关键）。

需要重点关注的情况：总广告观看量或广告 LTV 下滑而高难关未见抬升；或复活购买不升反降（削减未转化为付费）。

⑤ 资源与摩擦
低金币失败后直接结束会话的比例；道具缺货率与手持中位——加时闹钟是第一观察位（星星双周化后净减最多、且投放本就低于消耗）；指针/冻结在平衡线附近；金币复活是否随道具缺货同步上升（转嫁信号）。

需要重点关注的情况：加时缺货率抬头、或金币复活随缺货同步上升——道具紧缩在转嫁成金币支出。

⑥ 活动侧
冲刺/搏击的参与率、完成率与奖励流水；锦标赛阶段/节点到达率、真人公会进组内前 3 的比例与达标成员数（预算敏感位，由各星数档 AI 参数控制）、公会内前三贡献集中度；星星周 vs 锦标赛周的周末对比（48H 核心窗口：通关强度、参与、投放额、次周回访）。

需要重点关注的情况：排名奖支出跑出预算走廊（真人夺榜率远超模拟）、或阶段 1 到达率过低（头盔产出/门槛需调）；轮替对比连续偏向一边（为后续活动编排提供依据）。

O!Hole · 经济与活动调整方案 · 第二版 · 2026-08-19

O!Hole — Economy & LiveOps Adjustment · V2

2026-08-19 Calibration window 08-07~08-14 | weekly output 119.97M, sink 96.81M

One release, two layers: a global LiveOps layer (Guild Tournament × Star Collection weekend rotation — shipped to everyone, not part of any A/B) and a player-level economy A/B (the reduction package goes to the experiment arm only).

GLOBAL RELEASE · NOT IN A/B

Global LiveOps layer
Guild Tournament (new) × Star Collection (unchanged): alternate weekends; Mon–Thu Guild Chest untouched

PLAYER-LEVEL A/B

A ＝ new LiveOps baseline + current economy
B ＝ same baseline + reduction package + Crystal Rush + new Fight Club + settlement-ad ×2 + newbie protection

Experiment-arm net cut

−16.2~−17.3%

−19.49M ~ −20.74M per week

New players D0–6

−12.3~−13.4%

early −2.3% | advanced −18.4~−20.1%

Veterans D7+

−18.6~−19.6%

absorbs the bulk of the cut

Static output/sink

1.02~1.04

sink held constant; dynamic value pending the A/B

Coin ledger (waterfall)

Red = reduction channels; green = give-backs inside the experiment arm; navy = experiment-arm net. The global LiveOps layer (−0.21~−0.40pp) applies to everyone and is excluded here.

Adjustments at a glance

Where	Changes
Level settlement	star base 20/25/30 → 15/20/25 (uniform; multipliers & pickups unchanged)
Volcano Hunt	switch to ticket-only: daily free ticket and refill removed; story-chapter tickets 21 → 10; exchange stays at 6
Gold Tower	first qualifiers 3 → 6 (3,333 → 1,667) + win streak 7 → 10
Daily Tasks	high tiers −40%, base tiers untouched
Treasure Hunt	mid tiers −25%, top prizes kept
Story Chapter Reward	−25% from chapter 4; first 3 chapters ×2 (120/100/180 → 240/200/360; the doubling wins where they overlap)
Pass	free-track coins −30%; paid track & final prize untouched
Settlement ad	fixed bonus +40/36/32 → (star base × difficulty multiplier) × 2
Ad revive	daily ad revives 1 → 3; the day's 2nd and 3rd appear only when coins fall short of that revive's price; ad revives do not advance the 900/1800/3600 ladder
Newbie protection	check-in group 1 +40% + first 3 story chapters ×2

What players get in the same release

Three new events
Crystal Rush (60 min unlimited energy + packs weekly, zero coins); new Fight Club (coin rewards 51→102, +90%); Guild Tournament (new team mode, weekend slot).

Rule upgrades
The settlement ad becomes a clean double: harder levels now pay more than today (Hard 3★ 92→100, Boss 3★ 122→150), and for the first time playing well pays better — today's rule is inverted (1★ gets +40, 3★ only +32). Ad revives go from 1 to 3 per day; the day's 2nd and 3rd appear only when coins fall short, and they do not advance the coin-revive ladder.

All changes below apply to the experiment-arm config only; the control arm keeps live values.

Level settlement−5.06M

Star base 20/25/30 → 15/20/25, uniform across all levels, experiment arm only; difficulty multipliers (Hard ×2 / Boss ×3) and in-level pickups unchanged → channel −16.5%.

Config location：Level.xlsm / LevelReward: star base 15/20/25, experiment-arm config only (control keeps 20/25/30); Global multipliers unchanged

Volcano Hunt−7.13M

Switch to ticket-only: the daily free ticket and refill are removed, entry depends entirely on story-chapter tickets; chapter tickets 21 → 10; silver exchange stays at 6. 71.7% of runs came from the daily free ticket, so output falls to about 1.17M.

Config location：Challenge.xlsm / ChallengeDefine: daily free ticket and refill set to 0, exchange stays 6; Build.xlsm reward column volcano tickets (item 119) 21 → 10. Launch prerequisite: the uncapped add_cur and the failure to reset across calendar days (the >24h check) must be fixed first, and the fix applies to the control arm too — until then 1.17M is only a steady-state model value.

Gold Tower−3.51M ~ −4.76M

First qualifiers 3 → 6 (first reward 3,333 → 1,667) plus win streak 7 → 10. Prize pool 10,000 unchanged; target 8–12 final survivors. This is a range estimate — the tower has payout telemetry but no participation telemetry, so the attrition effect of the higher streak cannot be measured directly.

Config location：ActivityBattleRoyal.xlsm: first_reward 3 → 6 (bot config follows); required win streak 7 → 10

Daily Tasks high tiers−3.15M

High tiers −40%; base tiers fully untouched. The measured high-tier pool is 7.88M, of which 20.3% lands in the L1–30 band — new players are not fully insulated from it.

Config location：Task.xlsm / DailyTaskReward: high = the 6 rows with 101 ≥ 400 — id8 420→250, id14 500→300, id15 800→480, id33 500→300, id43 400→240, id44 700→420; the 8 base rows untouched

Treasure Hunt mid tiers−2.14M

Mid tiers −25%; top prizes 7,000/8,000 and the early sweetener tier are kept. Measured mid-tier pool 8.56M.

Config location：TreasureHunt.xlsm / TreasureHuntReward: 1001: tier2 100 kept, tier7 200→150, tier11 300→225, tier16 500→375, tier25 7,000 kept; 1003: tier2 100 kept, tier7 200→150, tier11 500→375, tier17 1,000→750, tier25 8,000 kept

Story Chapter Rewardnet +0.88M

First 3 chapters ×2 (current Build bases 120/100/180 → 240/200/360, +1.38M) as newbie compensation; −25% from chapter 4 onward (−0.50M). The two ranges overlap, and the doubling wins — chapter 3 spans L34–54, entirely above L31, so it is only doubled and never also cut. The first-3-chapter boost is the only give-back increased this round: a measured pool of 1.38M (counted by each chapter's face value, capping 120 below chapter 4 at L59 and 180 below chapter 9 at L184), of which 83.3% reaches D0–6 new players — the most efficient instrument available.

Config location：Build.xlsm reward column: chapters 1–3 (levels 4–54) coins 120/100/180 → 240/200/360; chapter 4 onward (L59+) ×0.75. Chapter-to-level map: ch1 = L4/9/14, ch2 = L19/24/29, ch3 = L34/39/44/49/54. The ×2 is already on the branch. (LevelChest is deprecated legacy and out of scope.)

Pass free track−0.83M

Free-track coins −30%; paid track and final prize untouched. Measured output in the L1–30 band is zero, so early new players are unaffected.

Config location：BattlePass.xlsm (Doodle id=1, 31 tiers): free-track coins — six tiers 100→70, tier30 200→140, tier31 300→210; final 3,000 untouched

Settlement ad ×2 (implemented on the abtest branch)−0.53M

Rule: watching the ad = (star base × difficulty multiplier) × 2. It is implemented as a server-side switch, not a table: LevelMechanism.ads_reward_type is 1 for control and 2 for the experiment arm; when the server reads 2 it merges the settlement coins into the ad reward, and the result panel shows ‘×2’. No new LevelAds row and no Difficulty.reward_ads pointer change. In-level pickups are not doubled (matching the implementation); No-Ads only removes the settlement interstitial — the double must be earned by actually watching. Ad value now scales with difficulty, fixing the inverted star curve: Hard 3★ 100 (vs 92 today) and Boss 3★ 150 (vs 122) are both higher than now. One open check: when the coin ×2 buff is active, does the ad double the already-buffed amount (×4)? It depends on whether the server merges pre- or post-buff coins into ads_items.

Difficulty	1★	2★	3★	Current (fixed bonus)
Normal ×1 → after ad	30	40	50	60 / 61 / 62
Hard ×2 → after ad	60	80	100	80 / 86 / 92
Boss ×3 → after ad	90	120	150	100 / 111 / 122

Totals after the ad; the doubling rule derives them from the settlement base automatically — no per-tier table entry needed.

Config location：LevelMechanism.xlsm: ads_reward_type = 1 (control) / ads_reward_type_11b = 2 (experiment); server Stage:GetAdsRewardTypeAbOpen == 2 merges coins into ads_items; client commit f101fa69 (result button shows ×2). Ad-button coverage stays as it is. Baseline: measured mean extra reward 38.0 → new 24.4.

Ad revive extension (implemented on the abtest branch)0 (creates no coins)

Today there are two layers: Global.revive_ad = [1,0,0] (the first revive in a run may use an ad) plus a daily cap of 1 on Ads id 101. The implementation changes three things: ① daily cap 1 → 3 (ads_limit_times_11b); ② new revive_ad_check_gold = [0,1,1], indexed by today's ad-revive count — the first needs no coin check, the 2nd and 3rd show the ad button only when coins fall short of that revive's price, otherwise the coin button is used; ③ revive_times_11b = 0, so ad revives are counted separately and do not advance the 900/1800/3600 ladder. Paying players therefore do not swap coin revives for ads — revives are 93.4% of all coin consumption and must not be eroded. Hard ceiling on the increment = +2 per player per day; the actual lift is for the A/B to read.

Config location：Ads.xlsm id 101: ads_limit_times 1 → ads_limit_times_11b 3; Global.xlsm adds revive_ad_check_gold [0,1,1] and revive_times_11b 0; revive_ad [1,0,0] and revive_gold [900,1800,3600] unchanged. Server counts ad revives separately from coin revives (commits 83a061c2, f7253795, 2859bac7).

Newbie protection+3.07M

Check-in group 1 +40% (+1.69M) plus first-3-story-chapters ×2 (+1.38M). Caliber note: group 1 = the first seven check-ins, keyed on check-in count — not on level and not on account age; the index loops back to 1 after 21 rows, so veterans reach the boost again. Measured (using the 250-coin reward unique to index 21 as a probe: 1,081 players complete a cycle each week): the group-1 pool is 4.22M = 3.20M new players + 1.03M veterans, so the boost lands 76% on new players and 24% (0.41M) on veterans — 2.8% of what veterans absorb, which is acceptable. Making it exact would require gating on create_days, i.e. server work, which is out of scope this round. For comparison, the chapter boost is 83.3% efficient but its base pool is only 1.38M and cannot carry the whole give-back.

Config location：Sign.xlsm group 1 (7 rows, currently 200/100/100/200/0/150/200) ×1.4; Build.xlsm first 3 chapters ×2

Config appendix (all touched rows)

Table	Item	Current	New	Status
Level / LevelReward	star base 1★/2★/3★	20 / 25 / 30	15 / 20 / 25	final
Global	difficulty multipliers	2;3	unchanged	final
Challenge / ChallengeDefine	daily free ticket / refill	on	off (ticket-only)	final
Challenge / ChallengeDefine	exchange	6	unchanged	final
Build	volcano tickets (item 119)	21	10	final
Challenge / Challenge_GS	add_cur cap + cross-day reset	defective	fixed (both arms)	prerequisite
ActivityBattleRoyal	first_reward	3	6	final
ActivityBattleRoyal	required win streak	7	10	final
Task / DailyTaskReward	high = six rows with 101≥400	420/500/800/500/400/700	250/300/480/300/240/420	final
TreasureHunt / Reward	mid coin tiers	200/300/500 ＋ 200/500/1000	150/225/375 ＋ 150/375/750	final
TreasureHunt / Reward	top prizes & early sweetener	7,000 / 8,000 / 100	unchanged	final
Build	chapters 1–3 (L4–54)	120 / 100 / 180	240 / 200 / 360	on branch
Build	chapter 4 onward (L59+)	current	×0.75 (no overlap with ch1–3)	final
BattlePass (id=1)	free-track coins (8 tiers)	100×6 / 200 / 300	70×6 / 140 / 210	final
LevelMechanism	ads_reward_type / ads_reward_type_11b (settlement ad ×2)	1 / —	1 / 2	on branch
Ads (id 101)	ad-revive daily cap ads_limit_times / _11b	1 / —	1 / 3	on branch
Global	revive_ad_check_gold (day's 2nd/3rd need coins short)	—	[0, 1, 1]	on branch
Global	revive_times_11b (ad revives don't advance ladder)	—	0	on branch
Global	revive_ad / revive_gold	[1,0,0] / [900,1800,3600]	unchanged	final
Sign	group 1 (7 coin rows)	200/100/100/200/0/150/200	×1.4 overall	final
Schedule	Star 5001 interval	7D	14D	final
Schedule	Tournament schedule row	—	new, 14D offset 7 days	new

Status: on branch = verified implemented on feature/dev_abtest_20260818 (tip f1528b5, fetched 2026-08-21); final = value locked, pending branch; prerequisite = fixed in both arms; new = new feature config.

Crystal Rush + new Fight Club (experiment arm only, judged with the economy A/B)

Crystal Rush (Monday)
Normal/Advanced 30min each, each grants one 30-min unlimited-energy buff; failure ends the run; zero coins: Normal win = Silver pack + Super Lightning + Extra Time; Advanced win = Gold pack + Grow + Gravity + Super Lightning + Compass; full-clear funnel 11.4%. Per the 07-23 spec, unchanged.

New Fight Club (Wednesday)
8 stages, target scores 18–47 frozen; rewards only at 2/5/8 wins: 100 + Bronze pack + Extra Time / 150 + Gold pack + Super Lightning + Gravity / 300 + Ruby pack + Grow + Super Lightning ×2 + Gravity; coin EV 51→102 (deliberate +90%). Replaces the old Wednesday Fight Club.

The two items below are global releases — not in the A/B：The Guild Tournament × Star Collection weekend rotation ships to the entire player base and forms the shared baseline of both A/B versions.

Guild Tournament

Grouping & matching：A room is created when a guild first scores, seeded with 10 AI guilds; real guilds join a room on their first score (unlocked-member count ≥10 and total guild stars are read at that moment), up to 20 guilds per room, later arrivals open a new room (the last room may be short). Four star tiers (0–10k / 10–30k / 30–80k / 80k+) select the AI behavior profile, which sets rank difficulty and budget; AI parameters freeze before start and are not tuned mid-cycle.

Tokens & scoring：Levels spawn three 'helmet' collectibles (3–7 per level in total); swallowing them grants tokens; tokens advance personal stage nodes, and opening a stage chest grants activity score (add_score) that feeds the guild total — guild ranking is driven by chest score, not raw clears.

Personal stage rewards

Node	Tokens (in stage)	Tokens (cumulative)	Guild score	Reward
1-1	10	10	10	Bronze pack ×1
1-2	20	20	15	Super Lightning ×1
1-3	40	40	25	Unlimited energy, 15 min
1-4	100	100	40	Gold pack ×1 + title Knight
2-1	60	160	20	Extra Time Clock ×1
2-2	80	180	30	50 coins
2-3	100	200	40	Grow ×1
2-4	250	350	80	Ruby pack ×1 + title Grand Knight
Loop	per 100	350+100n	30	Grow ×1 (fixed; the server repeats a single loop config — no rotation)

Reward allocation: coins concentrate on guild ranking rewards; personal stages carry items and card packs; intra-guild contribution rewards are single items. Stage items are all ×1 — ×2 is reserved for intra-guild contribution rewards (matching RM footage); one timed item only (node 1-3); no unlimited-lightning/extra-time buffs. Packs step up in rarity at key nodes: Bronze (entry) → Gold (stage 1 complete) → Ruby (stage 2 complete).

Guild ranking rewards (top 3, flat per qualifying member) & intra-guild contribution rewards (top 3, stackable)

Board	Rank	Reward
Guild ranking	1	2,000 coins + Gold pack ×1
	2	1,200 coins
	3	600 coins
Intra-guild contribution	1	Extra Time Clock ×2
	2	Super Lightning ×2
	3	Gravity Boost ×2

Guild ranking rewards are coin-centric; only rank 1 adds one Gold pack (a championship flourish) — ranks 2/3 stay coins-only, keeping scarce resources out of the remaining full-guild payouts. Intra-guild rewards are single items, tiered by item value.

Budget: coins ≈680k per cycle (guild ranking ~620k + stages ~60k) ≈ 340k/week ≈ +0.28pp, inside the corridor. Governance: AI parameters freeze before start and are not tuned mid-cycle; if a cycle overruns the corridor, reward amounts adjust from the next cycle. Before launch, publish the budget assumption table (real-guild placement rate per star tier, qualifying members per rank, node reach) for independent recalculation; item and pack output joins the weekly item pipeline. Packs: personal stages — Bronze (1-1) / Gold (1-4, ~1,150/week, watch album completion pace in cycle one) / Ruby (2-4, ~325/week); plus one Gold pack for guild rank 1 (scales with real-guild wins, small volume); ranks 2/3, the intra board and the loop stay pack-free. Timed items: a single 15-min energy buff at 1-3.

Schedule：Fri 08:00 → Mon 08:00 (72h), biweekly, offset 7 days from Star Collection (Star schedule 5001 interval 7D→14D); schedule rows verified with the release config.

Star Collection: no changes this round

Rewards (1,500/750/400/300 + Ruby/Gold/Silver/Bronze packs) and matching stay as-is — the rotation itself is the change; altering rewards at the same time would make the two weekend types incomparable. Current live data: champions ≈74% of prize winners, ~657 champions and ~3,500 claimants per cycle; 70% rotate each cycle, median account age 16 days (a known issue to be tracked and addressed later).

Rotation notes：Purpose of the rotation: keep weekends fresh and observe each event's impact on players separately, informing future scheduling (not a keep-or-retire decision). Method: primary comparison on the shared 48h core window (Sat 08:00–Mon 08:00), normalized per exposable/active player, adjacent weeks paired; the first 2 pairs validate operations, then track differences in activity, retention, monetization and payout across the two weekend types.

Weekend rotation schedule

Week	Weekdays (unchanged)	Weekend (rotates)
Week 1	Guild Chest (Mon 08:00–Fri 08:00, unchanged)	Guild Tournament（Fri 08:00–Mon 08:00）
Week 2	Guild Chest (Mon 08:00–Fri 08:00, unchanged)	Star Collection（Sat 08:00, 48h）
Week 3	Guild Chest (Mon 08:00–Fri 08:00, unchanged)	Guild Tournament（Fri 08:00–Mon 08:00）
Week 4	Guild Chest (Mon 08:00–Fri 08:00, unchanged)	Star Collection（Sat 08:00, 48h）

Weekend Treasure Hunt and Horse Racing keep running every weekend — non-guild players lose nothing; on tournament weekends, non-members see a join-guild prompt.

① Advanced new players (where this round's risk concentrates)
Caliber: registered 0–6 days and already past L31. Net change for this group is −18.4% to −20.1% — as deep as veterans or deeper, because Volcano, Gold Tower and the Pass are all unlocked for them. Watch: D7 retention, share below 600 balance, mainline clears per user. Why it is broken out: the D0–6 figure of −12.3~−13.4% is an average hiding a wide spread — early new players still in L1–30 only take −2.3%, so nearly all the risk sits in this one cell.

Situations that call for attention：If this group's share below 600 exceeds 45%, or its D7 retention falls more than 3pp versus control → trigger valve 1: restore volcano chapter tickets from 10 to 14 (about 1.5M back).

② New players overall & the early cohort
Share of D1 end-of-day low balances (~600 as reference line); share failing with no revive budget (<900); reach & usage of newbie protection; day-one behavior after the Gold Tower first-reward change 3,333→1,667. Calibers: ~600 = historical retention risk reference; 900 = first revive price; 900/1800/3600 = escalating in-run prices.

Situations that call for attention：Low-balance or can't-afford-revive shares in the experiment arm clearly trending above A; or protection reach is low (players who should get the ad revive are not getting it).

③ Retention & depth
D1/D3/D7 (new vs. veteran players read separately); mainline clears per user; session length and runs per day.

Situations that call for attention：New-player D1/D3 below the control arm for several consecutive days, or clears-per-user dropping mainly in L31–100 (where the cuts land) — first audit newbie-protection reach and the coin/item gap in that band.

④ Monetization
D7 LTV (ad and IAP split); coin-revive purchases and coin pack sales; ad view rate by difficulty — the ad rule's own signal (normal levels may dip, harder levels should rise; total volume is what matters).

Situations that call for attention：Total ad views or ad LTV dropping without the expected lift on harder levels; or revive purchases falling instead of rising (cuts not converting to payment).

⑤ Resources & friction
Share of sessions ending right after a broke-fail; item stockout rate and median holdings — Extra Time Clock is the first watch position (largest net decrease after the Star biweekly change, and already supplied below consumption); Compass/Freeze near balance; whether coin revives rise in step with item stockouts (a transfer signal).

Situations that call for attention：Extra Time stockouts ticking up, or coin revives rising with stockouts — item tightening transferring into coin spend.

⑥ Events
Crystal Rush / Fight Club participation, completion and reward flow; Tournament stage/node reach rates, share of real guilds placing top-3 with qualifying member counts (budget-sensitive, controlled by per-tier AI profiles), intra-guild top-3 concentration; Star weekend vs. Tournament weekend on the 48h core window (clear intensity, participation, payout, next-week return).

Situations that call for attention：Rank-reward spend escaping the budget corridor (real-guild placement far above simulation), or stage-1 reach too low (helmet spawn/thresholds need tuning); rotation comparison consistently one-sided (input for future event scheduling).

O!Hole · Economy & LiveOps Adjustment · V2 · 2026-08-19