执行摘要
① 全段铺开提失败：休息 ~7%（半数轻加）、普通 26% 起步→29%、困难 60%（人均尝试护栏 ≤2.3–2.5）、超难 85% 真墙 → 整体 33.4% 起步、最终 35.1%（≥2×），无单关立墙。② 分两版：先"细菌版"（低流失，中小图），再"时间版"（大图缩时）。③ 细菌结构大改：当前 大细菌占一半（48%）却没威胁；改为中细菌为主（~70%）、大细菌减为偶发障碍（~20%）、小细菌只在超难关用手摆。④ 提密同时后移"连败自动降级"阈值，否则被抵消。
1对标与打法
Jelly Hole · 手摆细菌
大图把可见细菌摆在路线/转向/贪心点，玩家"怪自己、下次能躲"，愿重试 → 留存好。
All in Hole · 更短时间
多靠压缩限时制造节奏压力；炸弹按难度档分固定区间、按进度分段递增，不是逐关随机。
第一版 · 先上细菌为主 + 大图选择性叠时间
全段铺开加可躲避细菌（中小图）；并对"大图 + 当前时间宽松"的子集叠加时间压力补齐到 2×。
第二版 · 后上扩大时间压力（学 All in Hole）
第一版留存确认后，把时间压力扩到更多大图，进一步逼近翻倍。
为什么第一版就要叠一点时间只靠细菌全段铺开，实测只能到 1.7–1.9×，达不到"翻倍"。缺口要靠时间补，但必须有纪律地叠：只在面积 >3500 且当前"耗时/限时 <86%"（时间有余量）的关缩时间——大图细菌本就被稀释，时间是它天然的杠杆，且有余量才不会立刻进流失区。耗时/限时 ≥86% 的关一律不叠时间，留第二版。
2现状与平衡点
2.1 当前线上表现：困难 / 超难是主压力点
困难 36% / 超难 45% 的单次失败率已经明显制造失败，但最终未通关（流失）率只有 3.5% / 4.8%——玩家反复失败仍能过去，这层"安全垫"就是我们提失败的空间；但垫子不厚，加压要小步。
细菌已是失败主因（67%），超时 32%。把失败往可躲避细菌方向迁移是顺势而为，且超时型失败更易导致流失。
普通关吃 57% 流量、贡献 40% 失败——整体失败率的大头在普通关，所以只动困难关搬不动整体。
密度 × 时间 风险矩阵：先细菌后时间的依据
时间不紧 <86%
时间偏紧 ≥86%
低密 <0.22
17.9%
通关 98.4% · 尝试 1.20
有余量
41.8%
通关 95.7% · 尝试 1.64
靠缩时间
目标密 0.22–0.30
32.4%
通关 97.2% · 尝试 1.44
★甜点区
53.9%
通关 93.5% · 尝试 2.03
流失区
高密 ≥0.30
38.8%
通关 96.0% · 尝试 1.57
边际递减
40.6%
通关 93.9% · 尝试 1.58
双压叠加
加细菌密度（纵向）失败升而通关率几乎不动 → 第一版主用；缩时间（横向）通关率掉到 93.5% = 流失 → 第二版、只用在有余量的大图。
2.3 地图大小 × 道具（计分物件）密度 → 难度（热力图）
口径说明：此处"道具密度" = 计分物件数 ÷ 面积，不含细菌（与第 4 节作为调参旋钮的"细菌密度"是两个不同指标）。单元格为次数通关率（越低越难），下方为爆炸率。大图 + 低物件密度是最毒象限（65% 过、爆炸率 29%）：能吃的东西稀 → 玩家跑得远 → 暴露在细菌里久 → 爆炸率高。第一版细菌要避免把关卡推进这一格；第二版的时间压力恰好适合放在大图。
3全段铺开：各类型失败率目标
| 类型 | 关数 | 流量占比 | 当前失败 | 目标失败 | 手段 |
|---|
| 休息关 | 37 | 16% | 1.6% | ~6–8% | 半数加 1–2 个明显中细菌，留半数纯喘息（喘息阀门不侵蚀） |
| 普通关 | 123 | 57% | 12.2% | 26% 起步 → 29% | 约 1/4–1/3 做尖峰抬底；早期保护（档A/B）先于加压上线 |
| 困难关 | 32 | 21% | 36.3% | ~60% | 细菌+手摆为主；护栏放宽为人均尝试 ≤2.3–2.5（与 60% 自洽） |
| 超难关 | 8 | 5% | 45.3% | ~85% | 做真墙（不碰 90），必要时叠小幅时间压力 |
| 整体 | 200 | 100% | 17.4% | 33.4% → 35.1% | 起步安全锚 ≈1.9× → 最终锚 ≥2× |
为什么不能只动困难关普通关占 57% 流量；只动困难+超难（26% 流量）要让整体翻倍，它们得做到 104% 失败率（不可能）。要动整体，就必须动全段。
锚点口径
最终锚：休息 ~7 / 普通 29 / 困难 60 / 超难 85 → 整体 ≈35.1%（≥2×）；起步安全锚：普通先走 26% → 整体 ≈33.4%，留存稳定后逼近最终锚。口径说明：① 休息关取 6–8%，保住喘息阀门价值（AIH 放松关恒零细菌）；② 困难关 60% 对应人均尝试约 2.5 次，护栏同步定为 ≤2.3–2.5 次；③ 超难 85 为锚，90 为不可触碰的上限。翻倍缺口仍在普通关（57% 流量）——超难即使 100% 失败整体也只到 30.6%。
4细菌档位与大中小比例设定
4.1 大中小比例：当前 vs 目标
当前（整体）— 大细菌占了一半
目标·本版起步 — 小细菌 3–5%（同步开始动态尝试）
目标·动态验证后 — 小细菌可提到 10–15%
小（超难/手摆）中（主力威胁）大（偶发障碍）
| 类型 | 当前 小/中/大 | 目标 小/中/大 |
|---|
| 休息 | 0 / 0 / 0 | 0 / 100 / 0 半数关 1–2 中 |
| 普通 | 0 / 50 / 50 | 0 / 75 / 25 |
| 困难 | 15 / 44 / 41 | 0* / 75 / 25 |
| 超难 | 13 / 46 / 40 | 20 / 65 / 15 |
*困难关小细菌仅用手摆，不用程序生成。比例为百分比。
不同尺寸细菌的处理原则
| 尺寸 | 现状问题 | 处理方式 |
|---|
| 小细菌 | 太小看不见，玩家被"暗杀"，10% 比例偏高、体验负面 | 本版起：比例先压到 3–5%、只在超难关手摆（位置可见），普通/困难关不用程序小细菌；并同步开始简单动态尝试——让小细菌动起来：原地弹跳、或沿固定路线弹跳巡逻。运动自然吸引视线，固定巡逻路线可学习、可预判（"下次能躲"）；避免瞬移或随机游走（不可预判 = 恶意感）。动态验证后比例可提到 10–15%（超难为主） |
| 中细菌 | — | 作为主力威胁，占比拉到 ~70%；每关增量主要靠加中细菌 |
| 大细菌 | 个头大但没威胁，更像障碍物，会遮挡小目标物 | 大幅减量，每关封顶 1–2 个、仅作偶发障碍；想提难加中细菌而非大细菌（大细菌太多反而毫无威胁） |
4.2 档位划分（按单关总细菌数）
| 档位 | 当前均值（A–D段） | 目标总数 | 构成（小/中/大） | 说明 |
|---|
| 档0 · 休息 | 0 | 0–2 | 0 / 1–2 / 0 | 半数休息关加 1–2 中细菌，半数保持 0 |
| 档1 · 普通 | 5 / 6 / 7 / 8 | 3–5 | 0 / 3–4 / 1 | 总数持平略降；威胁来自大→中替换（当前大细菌占一半） |
| 档2 · 普通尖峰 | （无独立档，同普通） | 5–7 | 0 / 4–5 / 1–2 | 约 1/4 普通关用此档制造节奏 |
| 档3 · 困难 | 6–7.5 / 7.5 / 8.5 / 9.5 | 7–10 | 0* / 5–7 / 2 | 总数基本持平；*小仅手摆；程序小细菌清零、大封顶 2 |
| 档4 · 超难 | （无独立档，同困难） | 10–14 | 2–3 / 7–8 / 2 | 唯一有真数量增量的档；小集中此处且手摆可见，大封顶 2 |
当前均值 = 现行虫组配置的段均值（51–100 / 101–150 / 151–200 / 201–250）。整体意图：档1–3 原始数量几乎不变，难度来自结构替换（大→中）+ 手摆位 + 动态细菌；档4 承担峰值，才有真数量增量。
4.3 阶段分布（51–250 分四阶段，对齐引擎原生 50 关分段）
| 阶段 | 关段 | 普通 | 普通尖峰 | 困难 | 超难 |
|---|
| 阶段 A | 51 – 100 | 4 | 6 | 7 | 10 |
| 阶段 B | 101 – 150 | 4–5 | 6 | 8 | 11–12 |
| 阶段 C | 151 – 200 | 5 | 7 | 8–9 | 12–13 |
| 阶段 D | 201 – 250 | 5 | 7 | 9 | 13–14 |
与"连败自动降级"协同引擎按本关连败次数自动降细菌等级兜底留存（连败 2 次起降）。第一版要让失败真正变多，需把降级触发从 2;4;6;8 适度后移，否则加的细菌被很快抵消。具体阈值不再一刀切——见下节按上下文分档。
4.4 动态细菌：给"加细菌"增加设计维度（第一批做简单动态）
第一批引入两种简单动态细菌——动态细菌当前已有、但数量偏少——在数量之外提升威胁的丰富度；复杂的动态机制（弹跳巡逻、追踪等）留到第二步。
① 高空掉落细菌
细菌放在塔顶 / 物件堆顶。玩家吞掉或碰塌下方支撑物后，细菌掉落，需要躲避。危险全程可见——塔是玩家自己拆的，砸下来"怪自己"。
② 宝箱细菌
细菌藏在宝箱/容器里，打开（吞开）瞬间弹出，需要躲避。开箱是玩家的主动选择——想要宝箱里的奖励、结果贪出细菌，归因于己且"赌"的张力更强。
设计规则（保持"公平死亡"）
① 危险要感知：放在上方的细菌要让玩家能看到；宝箱细菌用外观做区分（带标记的箱型=可能藏细菌），玩家学会后是"知情的赌"。② 摆放位置：贪心点、高分物件旁、回收路线交叉处——和手摆细菌同一套位置逻辑。③ 用量：普通尖峰关/困难关每关 1–2 处，计入该关总细菌档位预算（不额外超档）。④ 节奏：第一批与静态手摆一起上，并同步开始动态小细菌的简单尝试（原地弹跳、固定路线弹跳巡逻）——小细菌体型太小、静态容易被忽视，动起来才会被注意到；追踪、复杂路线联动等复杂动态放第二步。
4.5 失败缓解机制：分阶段 × 按失败原因路由 × 双杠杆
参考 AIH 的设计，在 AIH 的基础上增加两条我们自己的规则：① 按失败原因路由杠杆（超时死 → 降经验门槛；细菌死 → 动细菌侧，且分梯度）；② 按阶段与身份定时点（新手期全降、过新手期后时间降难先来、细菌降难滞后一拍）。
为什么要按失败原因路由
我们 67% 的失败是细菌致死、32% 是超时。被细菌炸死的玩家，只给他折扣所需经验救不了他——他不是分不够，是被炸死的；反过来超时玩家降细菌也动错了杠杆。所以：超时失败 → 经验/所需分折扣（+必要时补时）；细菌失败 → 细菌侧梯度降难。
规则一 · 分阶段时点表（何时开始降、降什么）：
| 阶段 × 身份 | 超时失败 | 细菌失败 |
|---|
| 新手期（≤50关）· 不分身份 | 第 2 次挑战即全降：细菌降一档（配置层）+ 所需经验 ×0.8，失败 2 次后 ×0.6 并再降细菌。依据：免费玩家第 2 次挫败流失率最高、第 1 次最低——救援必须赶在第 2 次失败之前；付费玩家新手期同样全保（前期漏斗是最贵的流失）。 |
| 过新手期 · 付费玩家 | 第 2 次挑战即 经验 ×0.85，随失败次数按档位曲线加深（×0.75 → ×0.65） | 第 2 次不动细菌（只给路线提示/预警增强）；第 3 次起进细菌梯度（见规则二）——细菌一减难度骤降，滞后一拍保住挑战感，也给复活/道具留出转化窗口 |
| 过新手期 · 免费玩家 | 第 2 次挑战即 经验 ×0.8（比付费更狠一点） | 第 2 次即进细菌梯度第 2 级（降配置细菌一档）——免费玩家连败 2 次就是流失峰值，等不起第 3 次 |
规则二 · 细菌降难四级梯度（越靠前的级别越不可察觉）：
| 梯度 | 动作 | 为什么放这个位置 |
|---|
| L1 提示级 | 路线提示 / 细菌预警圈增强，现已有局内复活提示 | 零难度损失，先帮玩家"看见"死因 |
| L2 配置级 | 降随机生成（非手摆）细菌一档 | 随机生成的细菌本来每把都不一样，降一档玩家察觉不到——降难被藏起来（先降非手摆） |
| L3 动态级 | 掉落 / 宝箱细菌停止触发 | 动态威胁消失可被察觉但可解释（"这把没开箱"），仍保留静态布局 |
| L4 手摆级 | 减少手摆细菌（极端连败才用） | 手摆是固定布局、玩家记得住，一删立刻穿帮——放最后；深段超难关永不进 L4，用经验折扣兜底 |
引擎现成支持：virus_adjust 降的正是配置生成组（L2）、stripSceneVirus 才动手摆（L4）——只需把阈值和触发条件按上表重接，不需要新系统。
规则三 · 助攻档二维表（进度段 × 关卡档位 T0–T4 → 档，决定折扣幅度与梯度推进速度；51–250 用 A–E 五档即可覆盖，F 档预留给 251 关及以后的更深段）：
| 进度段 ＼ 关卡档位 | T0 休息 | T1 普通 | T2 尖峰 | T3 困难 | T4 超难 |
|---|
| 新手（≤50关） | A | A | A | A | A |
| 中段（51–150） | A | B | B | B | C |
| 深段（151–250） | B | C | C（预留可拆 D） | D | E（真墙：永不清细菌，经验兜底） |
填细、用粗：列坐标按 T 档拆到五列，表内相邻列先给同字母（如中段 T1/T2/T3 同为 B）。这样尖峰关与普通关的助攻可以独立调——一旦数据显示尖峰关连败流失高于普通关，把深段 T2 列拆为 D 即可，关卡配置零改动。
门控与原则① 经验折扣按黑洞成长门控生效（≥3 级，开局不放水）；细菌调整在开局生效。② 免费/付费的差异只在时点（免费早一拍）与幅度（免费略深），结构同一套表，工程上是一个开关两组参数。③ 保护优先级总序：提示 > 经验折扣 > 补时 > 降配置细菌 > 停动态 > 减手摆（手摆最后）。④ 再难也保留最终放水出路；AB 重点关注"连败 2 次免费玩家"留存与"失败→复活/道具转化率"（付费滞后一拍正是转化窗口）。
4.6 杠杆倾向：什么关调细菌，什么关调时间，什么时候用经验值
三个杠杆（细菌、时间、经验值/所需分）不是三选一，按关卡情形给倾向：
| 关卡情形 | 首选杠杆 | 理由 |
|---|
| 中小图（≤3500）+ 时间宽松（<86%） | 加细菌（静态手摆 + 动态炸弹） | 主战场：小图里细菌密度直接咬人，失败归因清晰 |
| 大图（>3500）+ 时间宽松 | 缩时间 | 大图细菌被稀释、加了也不毒；时间才咬人（AIH 大图同样靠时间） |
| 时间已紧（≥86%） | 都不加 · 换形态 | 时间压力转成可见炸弹；失败后靠所需分折扣兜底 |
| 超难关（真墙） | 细菌高档 + 手摆小细菌 | 失败兜底用所需分折扣、不清细菌 |
| 休息 / 普通轻压关 | 少量中细菌 | 只做轻微紧张感，不动时间 |
| 新手段（≤50，原则） | 只降不加或微调 | 前期漏斗是真漏点（关卡质量报告：L1–15/L25 流失的下游 LTV 最贵），首败即救 |
5全 200 关动作总表（可筛选）
5.1 三区判断模型（先分区，再定动作）
健康加压区
玩家最终通关率 ≥95%、耗时/限时 <86%、合计密度未超 0.30 → 可加细菌/手摆。
转形态区
通关率 ≥95% 但时间偏紧或密度偏高 → 不加总量，只做路线手摆 / 随机转固定 / 大图交第二版时间。
复查区
通关率 <95% 或耗时/限时 ≥100% → 不加压，先复查目标分/路线/时间。下调数值需样本量达标（见下）。
难度判断用 4 个指标：合计密度、手摆数量、耗时/限时、最终玩家通关率。只看配置密度会误判——例如 L59 配置为 0 但手摆有 6，L164 配置密度已高、即使无手摆也不宜再加量。
样本量门槛（通关率结论的前提）
当前窗口单关挑战玩家数中位仅 ~164、深段只有 74–102：n≈100 时 95% 通关率的置信区间宽约 ±6pp，无法区分 88% 与 94%（如 L209 的"88.2%"背后只是 12 名玩家未过）。因此：单关挑战玩家 ≥500 且观测通关率 ≤90%，才可判定显著低于 95% 护栏并下调数值；n < 500 的关只标"复查"（扩数据窗口继续观察），不据此改配置。
5.2 第一轮困难关施工清单
轻量补手摆
8
L69/79/84/109/119/134/184/189
已有手摆·只优化位置
8
L54/59/64/99/104/114/139/144
手摆口径不笼统按"困难关 +2~3 手摆"执行。40 个困难/超难里，9 关已有 5–9 个手摆（如 L59 配置0但手摆6、L104 手摆8），这些只做位置优化不加量；真正第一轮新增手摆只有 9 关（L94 优先 + 8 关轻量）。普通关约 1/4 做尖峰、休息关半数轻加，全段铺开的量不变。
5.3 逐关明细（全 200 关）
按类型 / 动作筛选，或搜关号、主题。细菌列：配置=slot1中大均值，手摆=场景周目1手摆数，合计=两者之和。
| 关 | 名称 | 类型 | 面积 | 配置 | 手摆 | 合计 | 次数通关 | 玩家通关 | 人均 | 动作 |
|---|
6监控红线
上线红线
- 单关 耗时/限时 ≥ 86%：第一版不加细菌（留第二版）。
- 大图（>4500）细菌密度 ≤ 0.23。
- 玩家最终通关率：整体 ≥95%，困难/超难单关 ≥90–92% → 低于即复查。
- 人均尝试护栏（分类型）：普通 ≤1.35、困难 ≤2.3–2.5（随 60% 目标放宽）、超难峰值 ≤2.5–2.8。
- 普通关时间型失败占比不升高（升=缩时误伤）；失败↑但复活/道具消耗不升 = 保护过早或挫败过强。
- 每关留可见安全线，自动降级不可关死。
观测指标
- 整体失败率 17%→33.4%（起步安全锚）→35.1%（最终锚 ≥2×）；困难 36%→~60%；超难 45%→~85%。
- 玩家最终通关率 ≥96%；连败 3+ 次玩家关卡级留存不低于对照。
- 复活/道具使用率、金币消耗、续命付费上升。
Executive summary
① Raise failures across all tiers: Rest ~7% (light additions on half), Normal 26% at launch → 29%, Hard 60% (attempts guardrail ≤2.3–2.5), Super-hard 85% true walls → overall 33.4% at launch, 35.1% final (≥2×), with no single-level blocker. ② Two releases: the "bacteria release" first (low churn risk, small/mid maps), then the "time release" (shorter timers on big maps). ③ Rebuild the size mix: large bacteria currently take half (48%) yet pose no threat; shift to mid-size as the main threat (~70%), large reduced to occasional obstacles (~20%), small only hand-placed in super-hard levels. ④ While densifying, postpone the auto-relief thresholds — otherwise the added bacteria are quickly cancelled out.
1Benchmarks & approach
Jelly Hole · hand-placed bacteria
On big maps, visible bacteria sit on routes, turns and greed spots. Players blame themselves — "next run I can dodge it" — and retry, so retention holds.
All in Hole · shorter timers
Pressure mainly from compressed time limits; bombs follow fixed bands per difficulty tier, stepping up by progression segment — not per-level randomness.
Release 1 · firstBacteria-led + selective time on big maps
Roll out dodgeable bacteria across all tiers (small/mid maps); stack time pressure on the "big map + loose timer" subset to close the gap to 2×.
Release 2 · laterExtend time pressure
Once Release-1 retention is confirmed, extend shorter timers to more big maps and push closer to 2×.
Why Release 1 already stacks some timeBacteria alone across all tiers only reaches ~1.7–1.9×, short of "double". The gap is closed with time — with discipline: only levels with area >3500 AND current clear-time/limit <86% (time headroom) get shorter timers. Bacteria are diluted on big maps anyway; time is the natural lever there, and headroom keeps players out of the churn zone. Levels at ≥86% never get time stacked in Release 1.
2Current state & the balance point
2.1 Live performance: hard / super-hard are the main pressure points
Hard 36% / super-hard 45% single-attempt failure already creates failures, yet final non-clear (churn) is only 3.5% / 4.8% — players keep failing but still get through. That cushion is our room to raise failures; it is thin, so step carefully.
Bacteria already cause 67% of failures, timeouts 32%. Shifting failures toward dodgeable bacteria goes with the grain — timeout failures are the churn-prone kind.
Normal levels take 57% of traffic and contribute 40% of failures — the overall failure rate lives in normal levels; touching only hard levels cannot move it.
Density × time risk matrix: why bacteria first, time later
Time loose <86%
Time tight ≥86%
Low density <0.22
17.9%
Pass 98.4% · 1.20 tries
Headroom
41.8%
Pass 95.7% · 1.64 tries
Time-squeezed
Target 0.22–0.30
32.4%
Pass 97.2% · 1.44 tries
★ Sweet spot
53.9%
Pass 93.5% · 2.03 tries
Churn zone
High ≥0.30
38.8%
Pass 96.0% · 1.57 tries
Diminishing
40.6%
Pass 93.9% · 1.58 tries
Double pressure
Raising bacteria density (vertical) lifts failures while final pass barely moves → the main Release-1 lever. Cutting time (horizontal) drops final pass to 93.5% = churn → Release 2 only, on big maps with headroom.
2.3 Map size × item (scorable object) density → difficulty (heatmap)
Definition: "item density" here = scorable objects ÷ area, bacteria excluded (a different metric from the bacteria-density tuning knob in Section 4). Cells show per-attempt clear rate (lower = harder); the small line is the explosion rate. Big map + low item density is the most toxic quadrant (65% clear, 29% explosions): sparse edibles → long travel → long exposure to bacteria. Release-1 bacteria must not push levels into that cell; Release-2 time pressure fits big maps well.
Low
Mid
High
Small map
Mid map
Big map
3All-tier rollout: failure-rate targets by type
| Type | Levels | Traffic | Current | Target | Approach |
|---|
| Rest | 37 | 16% | 1.6% | ~6–8% | Half get 1–2 clearly visible mid bacteria; the other half stay pure breathers (don't erode the relief valve) |
| Normal | 123 | 57% | 12.2% | 26% launch → 29% | About 1/4–1/3 become spikes to lift the base; early relief (tiers A/B) ships before the pressure |
| Hard | 32 | 21% | 36.3% | ~60% | Bacteria + hand placement as main levers; attempts guardrail relaxed to ≤2.3–2.5 (consistent with 60%) |
| Super-hard | 8 | 5% | 45.3% | ~85% | True walls (never touch 90); small time pressure where needed |
| Overall | 200 | 100% | 17.4% | 33.4% → 35.1% | Launch anchor ≈1.9× → final anchor ≥2× |
Why hard levels alone can't deliverNormal levels carry 57% of traffic. Doubling overall through hard+super alone (26% of traffic) would require a 104% failure rate — impossible. To move the whole, every tier must move.
Anchor rationale
Final anchors: Rest ~7 / Normal 29 / Hard 60 / Super 85 → overall ≈35.1% (≥2×); launch anchor: Normal starts at 26% → ≈33.4%, approaching the final anchor once retention holds. Notes: ① Rest at 6–8% preserves the breather value (AIH keeps relax levels at zero bombs); ② Hard 60% implies ~2.5 attempts per clear, so the guardrail is set to ≤2.3–2.5 accordingly; ③ 85 is the super-hard anchor and 90 a hard ceiling. The doubling gap sits in normal levels (57% traffic) — even 100% failure on super-hard only reaches 30.6% overall.
4Bacteria grades & size mix
4.1 Size mix: current vs target
Current (overall) — large bacteria take half
Target · at launch — small 3–5% (dynamic trials start now)
Target · after dynamics validate — small up to 10–15%
Small (super-hard / hand-placed)Mid (main threat)Large (occasional obstacle)
| Type | Current S/M/L | Target S/M/L |
|---|
| Rest | 0 / 0 / 0 | 0 / 100 / 0 1–2 mid on half |
| Normal | 0 / 50 / 50 | 0 / 75 / 25 |
| Hard | 15 / 44 / 41 | 0* / 75 / 25 |
| Super | 13 / 46 / 40 | 20 / 65 / 15 |
*Small bacteria in hard levels are hand-placed only, never procedurally spawned. Values are percentages.
Treatment by size
| Size | Current problem | Treatment |
|---|
| Small | Too small to see — players get "assassinated"; a 10% share is too high, experience is negative | From this version: cap at 3–5%, hand-placed in super-hard only (visible spots); no procedural small bacteria in normal/hard. Start simple dynamic trials at the same time — make them move: bounce in place, or patrol a fixed bouncing route. Motion draws the eye; fixed routes are learnable ("I can dodge it next time"); no teleporting or random wandering (unpredictable = unfair). After dynamics validate, share may rise to 10–15% (mostly super-hard) |
| Mid | — | The main threat: push the share to ~70%; per-level increases come mainly from mid bacteria |
| Large | Big but harmless — acts as an obstacle and hides small targets | Cut hard: cap at 1–2 per level as occasional obstacles; to add difficulty add mid, not large (too many large = no threat at all) |
4.2 Grades (by total bacteria per level)
| Grade | Current avg (stages A–D) | Target total | Mix (S/M/L) | Notes |
|---|
| T0 · Rest | 0 | 0–2 | 0 / 1–2 / 0 | Half of rest levels get 1–2 mid; the rest stay at 0 |
| T1 · Normal | 5 / 6 / 7 / 8 | 3–5 | 0 / 3–4 / 1 | Raw counts flat-to-slightly-down; the threat comes from the large→mid swap (large is half of today's mix) |
| T2 · Spike | (no own grade today — same as Normal) | 5–7 | 0 / 4–5 / 1–2 | ~1/4 of normal levels use this grade for rhythm |
| T3 · Hard | 6–7.5 / 7.5 / 8.5 / 9.5 | 7–10 | 0* / 5–7 / 2 | Counts roughly flat; *small hand-placed only; procedural smalls removed, large capped at 2 |
| T4 · Super | (no own grade today — shares Hard) | 10–14 | 2–3 / 7–8 / 2 | The only grade with a real count increase; small concentrated here and visibly hand-placed; large capped at 2 |
Current avg = per-stage means of the live bacteria-group configs (51–100 / 101–150 / 151–200 / 201–250). Intent: raw counts barely change on T1–T3 — difficulty comes from the large→mid swap, hand placement and dynamic bacteria; T4 carries the peak and is where counts genuinely rise.
4.3 Stage distribution (four stages aligned to the engine's native 50-level blocks)
| Stage | Levels | Normal | Spike | Hard | Super |
|---|
| Stage A | 51 – 100 | 4 | 6 | 7 | 10 |
| Stage B | 101 – 150 | 4–5 | 6 | 8 | 11–12 |
| Stage C | 151 – 200 | 5 | 7 | 8–9 | 12–13 |
| Stage D | 201 – 250 | 5 | 7 | 9 | 13–14 |
Coordinate with auto-reliefThe engine already lowers the bacteria group after consecutive fails on a level (from the 2nd fail). For Release 1 to genuinely raise failures, the triggers must move later than 2;4;6;8, or the added bacteria are quickly cancelled. Thresholds are no longer one-size-fits-all — see the context tiers below.
4.4 Dynamic bacteria: a new design dimension (simple dynamics in batch one)
Batch one introduces two simple dynamic bacteria — dynamic bacteria exist today, but in small numbers — adding threat richness beyond raw counts; complex mechanics (patrols with tracking, route interplay) wait for step two.
① Drop-from-height bacteria
Bacteria sit on towers / object stacks. When the player eats or topples the support below, the bacteria falls and must be dodged. The danger is visible throughout — the player dismantled the tower, so the hit feels "my fault".
② Chest bacteria
Bacteria hide inside chests/containers and pop out the moment the chest is opened (swallowed). Opening is the player's own choice — they wanted the loot and greeded out a bacteria; self-attributed, with stronger "gamble" tension.
Design rules (keep deaths fair)
① Danger must be perceivable: bacteria placed up high must be visible to the player; chest bacteria are distinguished by looks (marked chest = may contain bacteria) — an informed gamble once learned. ② Placement: greed spots, next to high-value objects, route crossings — the same logic as hand-placed bacteria. ③ Quantity: 1–2 per spike/hard level, counted inside the level's bacteria-grade budget (no overflow). ④ Cadence: batch one ships together with static hand placement, and simple dynamic small-bacteria trials start at the same time (bounce in place, fixed bouncing patrol) — small bacteria are too tiny to notice when static; motion makes them visible. Tracking and complex route interplay wait for step two.
4.5 Failure relief: staged × routed by failure cause × two levers
Built on AIH's design, with two rules of our own added: ① route the lever by failure cause (timeout → lower the XP requirement; bacteria death → act on the bacteria side, in a gradient); ② timing by stage and identity (full relief in the newbie phase; afterwards time relief comes first, bacteria relief lags one beat).
Why route by failure cause
67% of our failures are bacteria deaths, 32% timeouts. A player killed by bacteria is not short of score — an XP discount won't save him; conversely, removing bacteria doesn't help a timeout. So: timeout → XP/required-score discount (+extra time if needed); bacteria death → the bacteria-side gradient.
Rule 1 · Staged timing table (when relief starts, and what it touches):
| Stage × identity | Timeout failure | Bacteria death |
|---|
| Newbie (≤L50) · everyone | Full relief from the 2nd attempt: bacteria down one grade (config layer) + required XP ×0.8; after 2 fails ×0.6 plus further bacteria cuts. Basis: free players churn most at the 2nd frustration and least at the 1st — relief must land before the 2nd failure; paying players get the same full cover here (early-funnel churn costs the most). |
| Past newbie · paying | From the 2nd attempt XP ×0.85, deepening along the tier curve (×0.75 → ×0.65) | 2nd attempt: bacteria untouched (route hints / stronger warnings only); from the 3rd attempt enter the bacteria gradient (Rule 2) — cutting bacteria drops difficulty sharply; lagging one beat protects the challenge and leaves a revive/item conversion window |
| Past newbie · free | From the 2nd attempt XP ×0.8 (slightly stronger than paying) | 2nd attempt enters gradient level 2 directly (config bacteria down one grade) — two straight losses is the churn peak for free players; a 3rd can't be afforded |
Rule 2 · Four-step bacteria-relief gradient (earlier steps are less perceptible):
| Step | Action | Why this position |
|---|
| L1 Hints | Route hints / stronger bacteria warning ring; in-run revive hints already exist | Zero difficulty loss; first help players see what killed them |
| L2 Config | Lower randomly-generated (non-hand-placed) bacteria one grade | Random spawns differ run to run anyway — one grade down is imperceptible; the relief stays hidden (non-hand-placed first) |
| L3 Dynamics | Drop / chest bacteria stop triggering | Noticeable but self-explaining ("didn't open the chest this run"); the static layout stays intact |
| L4 Hand-placed | Remove hand-placed bacteria (extreme streaks only) | Fixed layouts are memorized — deletion is instantly obvious; last resort. Deep super-hard never enters L4; XP discounts are the floor |
The engine already supports this: virus_adjust lowers exactly the config-generated group (L2), and stripSceneVirus is what touches hand-placed (L4) — only thresholds and triggers need rewiring; no new system.
Rule 3 · Assist-tier matrix (progress segment × level grade T0–T4 → tier, which sets discount depth and gradient speed; five tiers A–E cover 51–250, F is reserved for levels 251+):
| Segment ＼ Level grade | T0 Rest | T1 Normal | T2 Spike | T3 Hard | T4 Super |
|---|
| Newbie (≤L50) | A | A | A | A | A |
| Mid (51–150) | A | B | B | B | C |
| Deep (151–250) | B | C | C (may split to D) | D | E (true wall: bacteria never cleared, XP floor) |
Fill fine, use coarse: columns are split by T grade while adjacent columns start with the same letter (mid-stage T1/T2/T3 are all B). Spike and normal assists can then be tuned independently — if data shows spike-level loss streaks churn worse than normal, split the deep-stage T2 column to D with zero change to level configs.
Gates & principles① XP discounts activate via the hole-growth gate (≥ level 3 — no free ride at level start); bacteria adjustments apply at level start. ② Free vs paying differ only in timing (free one beat earlier) and depth (slightly deeper) — same table, one switch and two parameter sets. ③ Relief priority: hints > XP discount > extra time > config bacteria down > dynamics off > hand-placed cut (last). ④ Always keep a final relief path; A/B focuses on retention of free players at 2 consecutive fails and the failure→revive/item conversion (the paying lag is exactly the conversion window).
4.6 Lever preference: which levels get bacteria, which get time, when XP
The three levers (bacteria, time, XP/required score) are not either-or; preference by situation:
| Situation | Preferred lever | Why |
|---|
| Small–mid map (≤3500) + loose time (<86%) | Add bacteria (static hand-placed + dynamic) | Main battlefield: density bites on small maps and attribution stays clear |
| Big map (>3500) + loose time | Shorten time | Bacteria get diluted on big maps — time is what bites (AIH does the same) |
| Time already tight (≥86%) | Add neither · change form | Convert time pressure into visible bacteria; XP discounts as the failure floor |
| Super-hard (true walls) | High bacteria grade + hand-placed small | Failure floor via required-score discounts; bacteria never cleared |
| Rest / light normal | A few mid bacteria | Mild tension only; timers untouched |
| Newbie (≤50, principle) | Reduce or fine-tune only | The early funnel is the real leak (level-quality report: churn at L1–15/L25 costs the most downstream LTV); rescue on the first fail |
5All 200 levels · action table (filterable)
5.1 Three-zone model (zone first, then action)
Healthy pressure
Final player pass ≥95%, time/limit <86%, combined density ≤0.30 → may add bacteria / hand placement.
Change form
Pass ≥95% but time tight or density high → no net additions; route hand-placement / random→fixed / big maps wait for Release-2 time.
Recheck
Pass <95% or time/limit ≥100% → no pressure; recheck target score / routes / time first. Numeric nerfs require the sample threshold (below).
Judge difficulty on 4 indicators: combined density, hand-placed count, time/limit, final player pass. Config density alone misleads — e.g. L59 has 0 config but 6 hand-placed; L164 already runs high config density and shouldn't take more even without hand placement.
Sample threshold (precondition for pass-rate conclusions)
Median challengers per level in the current window is only ~164, deep levels 74–102: at n≈100 the 95%-pass confidence interval is about ±6pp wide — 88% cannot be told apart from 94% (L209's "88.2%" is just 12 players not passing). Therefore: only with ≥500 challengers AND observed pass ≤90% may a level be judged significantly below the 95% guardrail and nerfed; below 500, mark it "recheck" (extend the window), don't change configs.
5.2 First-round hard-level work list
Priority hand-placed
1
L94 (true zero-bacteria)
Light hand-placed
8
L69/79/84/109/119/134/184/189
Convert · no net add
2
L129 / L164
Has hand-placed · reposition
8
L54/59/64/99/104/114/139/144
Change form / wait R2
7
Big maps → time
Recheck · no pressure
14
Weak metrics, sample pending
Hand-placed baselineDon't blanket "+2–3 per hard level". Of the 40 hard/super levels, 9 already carry 5–9 hand-placed bacteria (e.g. L59: config 0 but 6 hand-placed; L104: 8) — those get repositioning only; genuinely new hand placement covers just 9 levels (L94 first + 8 light). Normal spikes (~1/4) and light rest additions stay — the all-tier volume is unchanged.
5.3 Per-level detail (all 200)
Filter by type/action, or search level number / theme. Bacteria columns: Config = slot-1 mid+large mean; Hand = playthrough-1 scene placements; Total = sum.
| L | Name | Type | Area | Config | Hand | Total | Attempt pass | Player pass | Avg tries | Action |
|---|
6Guardrails & monitoring
Launch guardrails
- Levels at time/limit ≥86%: no bacteria added in Release 1 (reserved for Release 2).
- Big maps (>4500): bacteria density ≤0.23.
- Final player pass: overall ≥95%, hard/super per level ≥90–92% → recheck below that.
- Attempts guardrails by type: normal ≤1.35, hard ≤2.3–2.5 (relaxed with the 60% target), super peak ≤2.5–2.8.
- Normal-level timeout-failure share must not rise (rise = time cuts misfiring); failures up but revive/item spend flat = relief too early or frustration too harsh.
- Keep a visible safe path in every level; auto-relief must never be fully disabled.
Metrics to watch
- Overall failure 17%→33.4% (launch anchor)→35.1% (final ≥2×); hard 36%→~60%; super 45%→~85%.
- Final player pass ≥96%; retention of 3+ loss-streak players no worse than control.
- Revive/item usage, coin spend and continue-payments trending up.