一句话结论：游戏的付费高度集中在深层关卡的高投入玩家身上——65% 的充值发生在第 100 关之后，35% 在 250 关之后。深层关即便通关率只有 60–70%，玩家的“进入下一关率”仍高达 95–99%——他们不会因为难而流失，而是选择付费复活继续。所以正确方向不是“全局加难”，而是“早期易、深层难”的杠铃结构：前期做易以扩漏斗、保留存，把难度集中加在深层去变现。数据也表明难度直接驱动留存：6/8 降难后 D1/D3/D7 全面上升，6/17 加难又几乎全部回吐——全局加难有留存代价，且代价主要在早期新手段。Bottom line: monetization is heavily concentrated among high-investment players deep in the game — 65% of spend happens after level 100, 35% after level 250. Even where deep levels have only 60–70% clear rates, players' "reach-next-level" rate stays 95–99% — they don't churn on difficulty; they pay to revive and continue. So the right move is not "harden everything" but a barbell: easy early, hard deep — keep the early game easy to grow the funnel and protect retention, and concentrate difficulty deep to monetize. The data also shows difficulty directly drives retention: after the 6/8 easing D1/D3/D7 all rose, and the 6/17 hardening gave nearly all of it back — broad hardening has a retention cost, mostly in the early game.
65%
充值来自 100 关以后of spend is after L100
≈21%
收入来自前 3 名付费玩家（深层高度集中）of revenue from the top 3 payers (deep, top-heavy)
95–99%
深层关“进入下一关率”（难但不流失）deep-level reach-next (hard but no churn)
50–62%
中深段失败→付费复活率fail→paid-revive rate, mid/deep
+6.5pt
6/8 降难后 D1 跃升（难度直接驱动留存）D1 jump after the 6/8 easing (difficulty drives retention)
1钱在哪里赚：付费集中在深层关卡Where the money is: payment concentrates deep
把每一笔充值（event_id=205，排除内部/QA 后合计 $16,162）按付费所在关（last_id）分档，占比随深度显著上升：Bucketing every purchase (event_id=205, $16,162 total excl. internal/QA) by the level at purchase (last_id), the share rises sharply with depth:
1–25 关L1–2512.2%
25–50 关L25–5010.4%
50–100 关L50–10012.0%
100–150 关L100–15012.3%
150–250 关L150–25018.3%
250 关以上L250+34.9%
充值金额占比（分母＝排除内部/QA 后总额 $16,162）。Share of spend (denominator = $16,162 excl. internal/QA).
洞察：付费靠的是把玩家推到深层——L100 之后贡献约 65% 收入，深层玩家价值远高于浅层。但深层收入高度集中在头部：前 3 名付费玩家就占全站约 21% 收入，且集中在最深层——把这 3 名头部玩家剔除后，L100+ 仍占约 57%（方向稳健），但最深层 L250+ 从约 35% 降到约 22%。所以“钱在深层”成立，但最深段主要靠少数头部玩家撑起、集中而脆弱，并非宽厚的付费基本盘——判断深层加难幅度时应以“剔除头部”的稳健口径为准。Insight: revenue comes from pushing players deep — levels past L100 contribute about 65% of revenue, and deep players are worth far more than shallow ones. But deep revenue is highly top-heavy: the top 3 payers alone are ~21% of all revenue, concentrated in the deepest levels — excluding those 3, L100+ is still ~57% (the direction holds), but the deepest tier L250+ drops from ~35% to ~22%. So "money is deep" holds, yet the deepest segment rests on a few top payers — concentrated and fragile, not a broad paying base; size any deep-hardening against the "top-payers-excluded" figures.
2为什么“难”能变成“付费”而不是“流失”Why difficulty converts to payment, not churn
变现闭环是：难度 → 失败 → 付费复活（金币/看广告）→ 继续。一关能否变现，取决于失败后玩家选择复活还是退出。按深度看：The loop is difficulty → fail → paid revive (coins/ad) → continue. Whether a level monetizes depends on whether players revive or quit after failing. By depth:
关卡段Segment
关数n
次数通关率Attempt pass
失败→复活率Fail→revive
复活成功率Revive succ
进入下一关率Reach-next
浅层 1–49Shallow 1–49
4
80.0%
44.6%
85.8%
89.1%
中段 50–99Mid 50–99
10
82.6%
62.4%
92.2%
95.0%
深段 100–149Deep 100–149
21
84.7%
57.3%
93.3%
97.4%
最深 150+Deepest 150+
61
82.7%
50.6%
93.1%
98.6%
Hard/Boss/Normal 关（排除休息关），到达≥40。覆盖说明：本数据集只含被改动的“跟踪关”（L29–234），前 49 关仅覆盖 L29/34/44/49（4 关），故“浅层”一行代表性有限。Hard/Boss/Normal (Rest excluded), reached≥40. Coverage: the dataset only covers the changed "tracked" levels (L29–234); below L50 only L29/34/44/49 (4 levels), so the "shallow" row is not representative.
“愿意复活”的甜区在中深段。中段（50–99）复活率最高 62.4%，深段（100–149）57.3%，且这两段进入下一关率仍有 95–97%——玩家既愿为难度买单、又几乎不流失。最深段（150+）承载最多收入、复活率却回落到 50.6%，说明最深处还有把复活率拉高的空间，而它恰是流失风险最低的一段（98.6%）。The "willing-to-revive" sweet spot is mid/deep. Mid (50–99) peaks at 62.4% revive-take, deep (100–149) 57.3%, and both keep reach-next at 95–97% — players pay for difficulty and barely churn. The deepest (150+) carries the most revenue yet revive-take falls to 50.6%, so there's headroom to raise revive-take deepest, exactly where churn risk is lowest (98.6%).
按关卡类型：难度容忍度不同By level type: different tolerance for difficulty
类型Type
关数n
通关率Pass
失败→复活率Fail→revive
复活成功率Revive succ
进入下一关率Reach-next
休息 RestRest
21
99.8%
—
—
99.4%
普通 NormalNormal
65
90.0%
55.4%
95.3%
99.0%
困难 HardHard
26
71.3%
50.4%
88.2%
94.6%
Boss
6
60.5%
39.5%
84.7%
95.0%
反直觉但关键：Boss 关最难（通关率 60.5%），但失败→复活率最低（只有 39.5%）。Boss 的失败常是“被压制、没机会”，玩家觉得复活也没用就直接退；普通/困难关的失败更像“差一点”，玩家愿意复活。结论：把 Boss 做得更难不会带来复活收入，只会劝退。变现要押在普通/困难关的“差一点”上，Boss 只当节奏/门槛。Counter-intuitive but key: Boss levels are hardest (60.5% pass) yet have the lowest fail→revive (only 39.5%). Boss failures feel "overwhelmed, no chance", so players quit rather than revive; Normal/Hard failures feel "so close", so players revive. So making Boss harder brings no revive revenue — only churn. Monetize on the "so-close" of Normal/Hard levels; keep Boss as pacing/gates.
把失败分成撞菌型（碰撞死）和超时型（倒计时耗尽），对比复活行为：Splitting failures into collision (hit a bacterium) and timeout (clock ran out), and comparing revive behavior:
撞菌型 · 复活率Collision · revive55.7%
超时型 · 复活率Timeout · revive43.0%
撞菌死比超时死多带来约 +13 个点的付费复活率（55.7% vs 43.0%）。撞到细菌更像“运气差一点”，玩家更愿意再来一次；超时更像“我需要更强/更多时间”，复活意愿低。所以加难优先用撞菌型压力，而不是砍时间。Collision failures yield ~+13 points more paid revive (55.7% vs 43.0%). A collision feels like "unlucky, one more try"; a timeout feels like "I need more skill/time", lowering revive intent. So prefer collision-type pressure when hardening, not cutting time.
4难度确实直接驱动留存：6 月两次方向相反的难度改动Difficulty directly drives retention: two opposite June changes
6 月有两次方向相反的难度改动：6 月 8 日降低难度、6 月 17 日增加难度。用按注册日的分群留存（platform=All）对比三段。注意：这几次都不是 A/B 测试，同期买量/用户质量本身也在波动，因此对比会受用户质量波动影响、结论并非百分之百确定，仅作参考方向。June had two opposite difficulty changes: eased on Jun 8, hardened on Jun 17. Comparing cohort retention by registration date (platform=All) across three windows. Note: none of these were A/B tests; acquisition/user quality also fluctuated, so the comparison is confounded by user-quality shifts and the conclusion is not fully certain — directional only.
注册群Cohort
D1
D3
D7
D1 进度(到达关 P50)D1 stage P50
A · 基线 05-24~06-07（6/8 前）A · Baseline 05-24~06-07 (pre-6/8)
41.0%
21.5%
12.6%
20.5
B · 6/8 降难后 06-09~16B · After 6/8 easing 06-09~16
47.5%
27.6%
16.9%
25.0
C · 6/17 加难后 06-18~29C · After 6/17 hardening 06-18~29
41.2%
22.4%
13.9%
21.9
难度直接、强烈地影响留存。6/8 降难后 D1/D3/D7 与进度全面跃升（+6.5 / +6.1 / +4.3 个点，D1 留存者多走约 4.5 关）；6/17 加难又几乎全部回吐。所以“全局加难”有明确留存代价，代价主要在早期新手段。但深层关加难几乎不流失（进下一关 95–99%）。→ 杠铃结构：早期做易以扩漏斗、保留存，把难度集中加在深层去变现。Difficulty moves retention directly and strongly. After the 6/8 easing, D1/D3/D7 and progression all jumped (+6.5 / +6.1 / +4.3 pts; D1-retained users reached ~4.5 levels further); the 6/17 hardening gave nearly all of it back. So broad hardening has a real retention cost, mostly in the early game. Yet deep-level hardening barely churns (reach-next 95–99%). → Barbell: keep early easy to grow the funnel and protect retention; concentrate difficulty deep to monetize.
这三关“进入下一关率”明显偏低、难度已越过劝退线，是主漏斗上的漏点。下面讲清它们带来多少收入（含金币）、卡掉多少人、能接受付费降多少、该改到什么程度。These three have clearly low reach-next rates — difficulty past the churn line, leaking on the main funnel. Below: how much revenue they bring (incl. coins), how many players they lose, how much revenue drop is acceptable, and how far to ease them.
关卡Level
到达Reached
卡死流失Lost
流失率Loss%
关上真金($)Cash($)
所在段金币消耗Coin spend(band)
通关率(全挑战)Pass(all-attempts)
进下一关(现→目标)Reach-next(now→target)
L29
2883
718
24.9%
$185
0.69M
67.8%
73.9% → ≥88%*
L89
930
193
20.8%
$70
1.97M
73.2%
78.1% → ≥90%
L119
730
139
19.0%
$202
1.86M
58.7%
81.2% → ≥92%
合计Total
—
~1,050
—
$457 (2.8%)
~4.5M
通关率＝通关次数 / 挑战次数(20001，含未结算/中途退出)，比“仅按结算”更能反映真实卡点。金币消耗＝gold-flow 按 10 关段，会高估到单关。*L29 见 ③。Pass = clears / challenge-starts (20001, incl. unsettled/mid-quit) — a truer picture of the wall than settlements-only. Coin spend = gold-flow by 10-level band, over-attributes to a single level. *L29 see ③.
① 这三关带来的收入很少① These walls bring little revenue
关上真金充值合计 $457（占全站 2.8%）。它们卡住的复活主要花金币（软货币），而金币基本是玩家免费获得的，并非直接收入来源。所以这些墙吃的主要是免费金币——是留存/参与的摩擦，不是变现来源；真正的增量收入 ≈ 真金 $457，远低于 $1,000（<6% 收入）。Cash purchases at these walls total $457 (2.8% of revenue). The revives they cause mostly spend coins (soft currency), which players earn largely for free — not a direct revenue source. So these walls mostly consume free coins — a friction/engagement device, not a monetization source; their true incremental revenue ≈ the $457 cash, well under $1,000 (<6% of revenue).
② 但它们卡掉的人很多，卡在通往付费区的路口② But they lose many players — right before the paying zone
三关合计卡死约 1,050 人（到达但不进下一关）。而付费集中在深层：到达 L100 的 844 人贡献 65% 收入（$10,570），每个深层到达者 ~$12.5；从 L30 走到 L100 的通过率是 39%。这些墙把上千人挡在深水区之外。若易化、回收其中一部分流量，估算下游深层收入增量约 $1,000–3,000（打折后仍 $800+）。The three lose ~1,050 players combined (reach but don't advance). Payment is deep: the 844 players who reach L100 generate 65% of revenue ($10,570), ~$12.5 per deep-reacher; flow-through L30→L100 is 39%. These walls keep over a thousand players out of the deep zone. Easing them to recover some of that flow yields an estimated downstream deep-revenue gain of ~$1,000–3,000 ($800+ even discounted).
能接受付费降低多少？可以接受这三关的直接付费全部让出（~$457 / 2.8%）；实际预计只掉 ~$150–230（~1.4%），而净效应预计总收入上升（下游 > 关上）。一句话：让出很小、换回可观。How much revenue drop is acceptable? We can accept giving up all direct payment at these walls (~$457 / 2.8%); the expected actual drop is ~$150–230 (~1.4%), and the net effect should be higher total revenue (downstream > at-wall). In short: give up little, recover a lot.
③ 该改到什么程度（三关不一样）③ How far to ease (differs by level)
L119（首选）：真难（通关率 58.7%），流失最可能由难度造成、回收率最高，且在深层、回收的人直接进最高 ARPU 段。适度降低难度，把进下一关率抬到 ≥92%。L119 (top pick): genuinely hard (58.7% pass), loss most likely difficulty-driven with highest recovery, and deep — recovered players enter the highest-ARPU zone. Ease moderately; lift reach-next to ≥92%.
L29：到达 2883 是超大早期流量，25% 流失里相当一部分是早期“游客”自然流失、非难度——易化能改善但封顶有限，进下一关率目标现实设 ≥88%，别指望拉到 95%。L29: reached by 2883 — huge early traffic; much of the 25% loss is natural early "tourist" churn, not difficulty — easing helps but has a ceiling, so target reach-next ≥88% realistically, not 95%.
6行动建议（只讲大方向）Recommendations (direction only)
杠铃结构：前期（约前 49 关）与休息关保持易——它们几乎不贡献付费、却是新手留存最脆弱处。Barbell: keep the early game (~first 49 levels) and Rest levels easy — they barely monetize yet are where new-user retention is most fragile.
把难度集中加在深层（约 50–200、尤其 150+）：这里承载大部分收入、玩家愿意付费复活、几乎不流失，是变现主战场。Concentrate difficulty deep (~50–200, especially 150+): the bulk of revenue, players pay to revive, and churn is minimal — the monetization battleground.
回调早期/中期的劝退墙（L29 / L89 / L119）：救留存 + 把更多人送进深层付费区，净效应双赢。Roll back the early/mid walls (L29 / L89 / L119): recover retention + push more players into the deep paying zone — a win-win.
Boss 关不再加难：Boss 失败→复活率最低（被压制就退），加难只劝退不变现。Don't harden Boss levels: they have the lowest fail→revive (players quit when overwhelmed) — hardening only churns, doesn't monetize.
具体加多少、用什么手法（细菌 / 时间等）以《51-250 关难度提升设计方案》为准，本报告不展开。Exact magnitude and method (bacteria / time etc.) follow the Levels 51-250 Difficulty Increase Plan; not detailed here.
7护栏（观察指标）Guardrails (metrics to watch)
进入下一关率：加难后单关尽量保持在约 93% 以上；明显偏低说明卡点太硬、在漏人，需回调。Reach-next rate: after hardening, keep each level around 93%+; clearly lower means the wall is too hard and leaking — roll back.
复活成功率：尽量别太低，让“复活花钱有用”。不过它受关卡结构影响、不容易精确保证，作参考观察即可。Revive success rate: keep it from getting too low so "paying to revive pays off." But it depends on level structure and is hard to guarantee precisely — treat it as a reference metric.