# Negative Cases ## Scientific Overclaim Bad output: ```text 体验浓度公式已经证明能提升留存，因此 D1 必然上升。 ``` Required correction: mark `theory_status: design_hypothesis`, and state that real telemetry is required. ## Frequency First Bad output: ```text 玩家看不懂，所以先把事件频率提高一倍。 ``` Required correction: diagnose `CLP` first. Do not add events before reducing noise. ## Boredom Means Add Events Bad output: ```text 玩家说无聊，所以当前刺激太低，B 组把怪物、弹窗和任务频率都提高。 ``` Required correction: first output `optimal_stimulation_fit` and `boredom_type`. Boredom can be `under_stimulation`, `over_stimulation`, `habituation`, `low_agency`, `low_meaning`, `mixed`, or `unknown`. ## False Density Bad output: ```text 给更多红点、弹窗和弱奖励，让玩家每分钟看到更多东西。 ``` Required correction: weak prompts without meaningful decisions or salient feedback do not improve ED. ## Multi-Lever Confound Bad output: ```text B 组同时改奖励、UI、剧情、战斗难度、镜头和商业化。 ``` Required correction: each variant keeps one primary lever and a rollback path. ## FEP Overclaim Bad output: ```text 根据自由能原理，这个方案已经从神经科学上证明会让玩家进入心流。 ``` Required correction: FEP is a design metaphor here. Mark `theory_status: design_hypothesis` and validate with telemetry or playtest evidence. ## Novelty Amount Fallacy Bad output: ```text SDT 需要 novelty，所以我们把新规则、随机事件和信息层都加倍。 ``` Required correction: SDT novelty must be treated as optimal, learnable novelty. Output `optimal_novelty_fit`; if novelty is random, overwhelming, or not attributable, treat it as `CLP` or `too_high`, not as a positive. ## Markov Blanket Buzzword Bad output: ```text 手感不好是马尔可夫毯问题，所以 B 组同时改输入、镜头、UI、敌人 AI、奖励和关卡节奏。 ``` Required correction: identify the concrete coupling break, such as latency, noise, unclear mapping, weak agency, or overload. Keep one primary lever per variant. ## Single-Player Metric Mismatch Bad output: ```text 这是买断制单机 Demo，所以 P1 直接用 D1/D7 和连续登录天数。 ``` Required correction: for `premium_single_player`, use total journey metrics such as total playtime, demo/chapter completion, core-loop reach, replay intent, and review/refund risk if available. D1/D7 are not default P1 unless the project has explicit liveops or return-cadence goals. ## Mobile Metric Mismatch Bad output: ```text 这是长线手游活动，只看总游戏时长上升即可，不需要看每日和持续天数。 ``` Required correction: for `mobile_liveops`, include D1/D3/D7/D30, daily active sessions, consecutive active days, event retention, return-session quality, and fatigue/complaint gates. ## No Anti-habituation For Long-term Fatigue Bad output: ```text 老玩家疲劳，所以把奖励数值翻倍并增加每日任务次数。 ``` Required correction: when the case involves seasons, looters, roguelikes, UGC, old players, repeat loops, or long-term fatigue, output `anti_habituation_plan` with a lever such as `alternative_use`, `attention_investment`, `conscious_reframing`, `context_shift`, or `combinatorial_depth`. ## No Evidence Gate Bad output: ```text 玩家说首局空，所以真实原因是事件太少，直接提高首局怪物和奖励频率。 ``` Required correction: first declare `evidence_gate`. With only text, use `L0_text_only` and `evidence_status: assumption_only`; output a hypothesis and minimum validation plan, not a factual cause. ## Static Screenshot Tempo Claim Bad output: ```text 从截图可以看出玩家会在第 90 秒退出，所以 B 组调整 90 秒后的节奏。 ``` Required correction: screenshots can support static UI and visual-noise judgments, but cannot prove real pacing, feel, or exit timing. Request recording/playtest/telemetry or downgrade the claim. ## Wrong Output Density Bad output: ```text 用户只问“首局太空，快速看一下”，于是输出完整 19 个模块和 4 页报告。 ``` Required correction: use `quick_ed_triage` for one-line quick requests. Output one boundary, one stimulation-window judgment, one primary lever, two minimal changes, three metrics, and one rollback condition. ## Hybrid Single-Win Fallacy Bad output: ```text 活动 D7 上升，所以 hybrid 项目整体成功，主线完成率下降可以忽略。 ``` Required correction: `hybrid` must split total-journey P1 and liveops P1. If either critical horizon is harmed, do not declare overall success; review negative gates and choose iterate, rollback, or a split experiment. ## Review Without Negative Gates Bad output: ```text P1 上升了，直接放大，不需要看投诉、失败率、经济或数据质量。 ``` Required correction: `review_and_decide` must check data quality and negative gates before P1/P2 interpretation. A P1 lift contaminated by technical, economy, fairness, fatigue, or dark-pattern risk is not a clean win.