Atlas Q&A 手册

WHY-first · 亮色主题 · 衍生指标已标注 · 全卡片中英双语

信息分类图例

本手册数据的四种来源

CASE FACT
案例事实 — 直接出现在case study中
DERIVED
衍生推算 — 用案例数据+公式，有明确假设
PLAN
方案建议 — recovery plan提出的措施
GATE
关卡待定 — Week 2/4/8才能确定
最重要的一件事

核心论点链（WHY-first）

1
Atlas完成了可见活动，但管理层仍缺乏稳定性、就绪状态和实际收益的可靠证据。40%降本目标未通过completed benefits tracking验证。
Management still cannot judge its value with confidence.
↓
2
5个已完成里程碑展示的是活动，不是scale决策所需的成果。
They show activity, not stability, readiness or realised benefits.
↓
3
三个控制失败：风险未被强制处理、WBS未实质性修订、没有端到端performance owner。
Informal change, untreated risk, unclear ownership → demonstrations without acceptance evidence.
↓
4
为原始范围追加资金只会延续同样的不确定性。取消是可辩护的后备，但会在受控测试前放弃最强用例。
More funding now would extend the same uncertainty.
↓
5
缩窄到simple customer support，稳定，衡量。为管理层提供透明证据做Week 8决策。
Narrowing concentrates limited capacity on one achievable target.
↓
6
两测试都通过→推荐扩展。任一不通过→默认关停。团队出证据，Sponsor做决策。
Both pass: recommend scale. Either fails: closure is the default. The Sponsor decides.
一句话版

Narrowing scope creates focus; that focus frees time and budget for quality activities; those activities reduce rollout risk and make the Sponsor's decision evidence-based.

缩窄范围→集中精力→释放时间和预算给质量活动→降低推广风险→让Sponsor的决策有据可依。

案例事实与分类数据

必记数据

CASE FACT
$4.5M
BAC 批准总预算
CASE FACT
$4.1M
AC 已花费
DERIVED
~$0.4M
名义预算额度，按BAC−AC计算，恢复工作尚未定价
CASE FACT
0.76
CPI 每花$1完成≈$0.76预算工作价值
CASE FACT
0.68
SPI 进度绩效指数
CASE FACT
40%
原始降本目标（未通过benefits tracking验证）
CASE FACT
M12/14
第12个月（共14个月）
PLAN
8 周
提交方案采用的恢复窗口 (2+2+4)
CASE FACT
5
已完成里程碑（展示活动非成果）
CASE FACT
6
未完成关键活动（验收驱动）
衍生指标 DERIVED ESTIMATE

课程EVM计算 — 用于Q&A展示理解

重要声明

以下计算用于说明原始基线在特定假设下意味着什么，并不是缩窄后恢复方案的报价。恢复工作的权威成本仍由Week 2基于narrowed WBS进行bottom-up ETC估算。

EV ≈ $3.116M
公式: EV = CPI × AC
代入: 0.76 × $4.1M = ≈ $3.116M
假设: 报告的CPI和AC在状态日期有效。
Q&A: 展示历史成本低效。
⚠ EV是预算工作价值(budgeted work value)，不是实际业务价值。
EAC ≈ $5.92M
公式: EAC = BAC ÷ CPI = $4.5M ÷ 0.76 = ≈ $5.92M
假设: 原始范围可比且历史CPI持续。
Q&A: 按当前效率继续原始范围可能超出批准预算。
⚠ 不是缩窄后恢复方案的成本。
ETC ≈ $1.82M
公式: ETC = EAC − AC = $5.92M − $4.1M = ≈ $1.82M
假设: 同EAC：原始范围+历史CPI不变。
⚠ Illustrative ETC for the original scope, not our recovery funding request.
VAC ≈ −$1.42M
公式: VAC = BAC − EAC = $4.5M − $5.92M = ≈ −$1.42M
假设: 同EAC。
⚠ 条件性预测(indicative projected overrun)，不是确定结果。
⭐ TCPI ≈ 3.46
剩余预算工作: BAC − EV = $1.384M
剩余批准预算: BAC − AC = $0.4M
公式: TCPI = $1.384M ÷ $0.4M = ≈ 3.46
假设: 原始范围必须在现有BAC内完成；计算的是所需未来效率，不是假设CPI=0.76持续。
Q&A: 最有力的衍生指标。说明原始基线financially implausible。
⚠ 说extreme improvement required，不要说mathematically impossible。仅在老师追问财务可行性时使用。
Burn Rate ≈ $0.342M/月 · 名义跑道 ≈ 5周
计算: $4.1M ÷ 12 = ≈ $0.342M/月；$0.4M ÷ $0.342M = ≈ 5周
假设: 支出近似线性；$0.4M未被占用；未来沿用历史支出模式。
⚠ 粗略跑道指标，不是恢复预算预测。缩窄范围改变了工作和burn profile。
SPI Duration ≈ 20.6月 — 进阶参考
计算: 14 ÷ 0.68 = ≈ 20.6月（超出约6.6月）
⚠ 这是粗略的警示性推算(rough heuristic)，不是正式的完工预测。SPI-based duration extrapolation在项目后期尤其不可靠，也不能用于定价或安排缩窄后的恢复计划。
每道题都用这个结构

答题方法 · 4步法

Position — 直接回答
第一句话给结论
Evidence — case事实
人名、数据或原话
Control — 方案中的控制或gate
指向recovery plan的机制
Boundary — 不做什么/不假设什么
划边界，展示纪律
关键原则

当问题挑战一个假设时，指向测试该假设的gate，不要声称假设已被证明。
目标20–35秒，通常40–75个英文词。复杂问题最多约90词，不要作为常态。不超过40秒。

Q&A · A类 — 方案逻辑与推荐理由

Recommendation & Rationale

必背Why narrow the scope instead of funding the original scope?为什么缩窄而不是追加资金？
ENGLISH ANSWER
Management does not yet have reliable evidence of stability, readiness or realised benefits. Adding funding to the original scope would extend the same uncertainty. Narrowing creates one bounded, measurable target and gives the Sponsor transparent evidence before deciding on further investment.
可选财务补充 · Optional financial support
If challenged on affordability, the original-scope TCPI is approximately 3.46 — completing the original scope within the remaining approved budget would require an extreme improvement in future cost efficiency.
⚠ 主答案先讲管理层缺乏决策依据。只有老师追问财务可行性时再用TCPI。
中文逻辑
核心问题不是钱不够——是花更多钱仍然回答不了"Atlas有没有价值"。先证明价值再谈投资。
必背Why not cancel Atlas immediately?为什么不直接取消？
ENGLISH ANSWER
Immediate cancellation is a defensible fallback, but simple support has not yet been tested under controlled conditions. The evidence does not show that every use case has failed — it shows that reliability and value remain unproven. If either test fails, closure is the default.
中文逻辑
不是因为花了$4.1M所以不能停（sunk cost fallacy）。是因为最强用例还没受控测试过，先测了再决定。
必背Why is rebaselining alone insufficient?为什么光rebaseline不够？
ENGLISH ANSWER
Rebaselining alone changes dates and costs but leaves the same scope and reliability problems. Our plan narrows scope first, then rebuilds the WBS, schedule, ETC and controls around different, bounded work.
中文逻辑
光改日期和预算没用——范围太大、可靠性问题还在。先缩窄再建新baseline，围绕的是不同的工作。
必背Why is simple customer support the recovery scope?为什么选simple support？
ENGLISH ANSWER
Simple support provides the clearest available test of repair effort, handling cost and operational value. The case says it performs reasonably well. Complex conversations are less reliable. Logistics still requires human verification. Sustained analytics adoption has not yet been measured.
中文逻辑
最清晰的测试目标：修复事件可计数、降本可衡量。其他模块要么不稳定、要么需人工验证、要么没人测量过使用率。
必背What is the trade-off?trade-off是什么？
ENGLISH ANSWER — 用因果链
Narrowing sacrifices breadth, but it frees time and budget for validation, regression testing and measurement. These are prevention and appraisal costs. We accept them now because they reduce rollout risk, customer harm and later rework.
中文逻辑
链条式：缩窄→释放资源→投入质量活动→减少推广风险和后期返工。不要用清单式回答。
高频Why retain logistics advisory use?为什么保留物流咨询？
ENGLISH ANSWER
No new development. Mandatory human verification. Outside the measured recovery scope. Cannot consume stabilisation capacity. The case notes it identifies useful efficiencies in straightforward cases.
中文逻辑
不做新开发、必须人工验证、不在受控试点范围内、不消耗稳定化产能。砍掉反而浪费已有的有限价值。
Q&A · B类 — 可行性与资源

Feasibility & Resources

必背Is eight weeks feasible?8周够吗？
ENGLISH ANSWER
Eight weeks is the remaining project window, not a guarantee. Week 2 explicitly tests feasibility. Named engineering, data, CX and vendor capacity must be committed and costed. If recovery work is not fundable or supportable, the plan stops.
中文逻辑
不说"够"。说"8周是窗口，Week 2测试可行性——不可行就停"。不确定性推到gate上。
高频Two senior developers left. Who performs the recovery work?两个高级开发走了，谁来做？
ENGLISH ANSWER
That is exactly what the Week 2 resource gate addresses. Proceed only if committed capacity is confirmed and costed. If capacity is not available, the plan does not proceed. We do not assume adequate staffing — we test it.
中文逻辑
不假设人够——Week 2的gate测试这个前提。测不过就停。
高频Who owns end-to-end recovery performance?恢复期间谁负责整体？
ENGLISH ANSWER
The Recovery Lead is accountable for integrated performance through Week 8. The Product Owner owns scope, acceptance and measurement design. The Data Team Lead owns the accepted, version-controlled dataset. Engineering and the vendor produce regression evidence. CX performs repair tagging and operational measurement. The Head of CX signs off readiness. Finance owns the ETC and net-benefit analysis. Risk and Compliance signs off AI controls. A permanent service owner is appointed before scale.
中文逻辑
记住Recovery Lead管整体、Product Owner管范围和验收、Finance管钱和效益分析、Risk/Compliance管AI风控。没有单方同时生产和验收。
高频Why not ask for additional funding today?为什么今天不直接要钱？
ENGLISH ANSWER
Management does not yet have enough evidence to justify funding the original scope. Week 2 prices the bounded recovery work. Further funding is considered only after the work is costed and the decision basis is clear.
中文逻辑
没有证据支持的资金请求不会被批准。先缩窄、定价，Week 2再谈钱。
Q&A · C类 — 衡量指标与质量成本

Measures & Cost of Quality

必背What is a repair event?什么算修复事件？
ENGLISH ANSWER
An unplanned human intervention after an unresolved, materially incorrect or unsafe AI response. Planned handoffs are excluded. The definition is fixed before the pilot starts.
中文逻辑
AI回答了但有问题（未解决/实质错误/不安全），人工不得不介入修复。计划内的人工转接不算。定义在试点前锁定。
高频What is a material regression?什么算重大回归？
ENGLISH ANSWER
Deterioration against the accepted regression baseline beyond the tolerance agreed at Week 2. The comparison must use controlled versions of the data, model and configuration. An unresolved material regression at the Week 8 decision point blocks scale.
中文逻辑
不是任何小波动都算失败。Week 2先定baseline和tolerance；超过容差且未解决的退化阻止scale。
高频How are the threshold and minimum volume set?阈值和最低样本量怎么定？
ENGLISH ANSWER
Set at Week 2 before results are known. Product Owner and Head of CX derive the threshold from the validated manual baseline and agreed risk tolerance. Minimum volume prevents a misleading result from too few interactions.
中文逻辑
所有参数在结果出来之前锁定，防止事后调整。最低样本量防止样本太少导致误导性结论。
高频What happens if stability passes but net benefit fails?稳定通过但效益没通过？
ENGLISH ANSWER
Closure remains the default. Technical viability cannot substitute for business viability. Both tests must pass. The Sponsor may consider a bounded exception, but that is not automatic.
中文逻辑
技术能跑≠商业划算。两个测试都必须通过。只通过一个=默认关停。
高频How can a four-week pilot establish net benefit?四周试点如何证明净效益？
ENGLISH ANSWER
The four-week pilot does not prove long-term value with certainty. It provides a controlled directional reading using actual handling savings and all recurring operating costs. The measurement method and minimum volume are fixed at Week 2. If the evidence is insufficient, the team cannot recommend scale.
中文逻辑
四周不能保证长期收益，但形成第一次受控、可审计的方向性证据。测量方法和最低样本量在结果前确定。
高频Where is cost of quality in the plan?质量成本体现在哪？
ENGLISH ANSWER
Prevention: data validation, version control, controlled configuration. Appraisal: regression testing, repair measurement, acceptance criteria. Failure costs avoided: rework, customer repair, operational harm, failed rollout. These costs are included in the Week 2 ETC.
中文逻辑
预防+评估=主动投入（现在花）。返工+客户修复+运营损害=被动损失（不投入以后花更多）。
高频What AI controls must be accepted?试点前必须接受哪些AI控制？
ENGLISH ANSWER
The detailed controls are confirmed at Week 2, but they must address accountability, auditability, unsafe responses and changes caused by data, model or configuration updates. Risk and Compliance must approve the controls before pilot entry and again before scale.
中文逻辑
不要编造完整AI治理框架。回答case里的四类风险：问责、可审计、不安全响应、数据/模型/配置变更。强调Risk/Compliance两次签字。
进阶Why only two primary success tests?为什么只有两个主要测试？
ENGLISH ANSWER
Stability and net benefit directly address the two unresolved decision questions. Other controls are mandatory prerequisites but do not independently justify scale.
进阶Why not retain the original 40% target?为什么不沿用原来的40%目标？
ENGLISH ANSWER
The 40% target belonged to the original scope and assumptions. The narrowed scope requires a new positive Sponsor-approved hurdle. This does not remove the requirement to demonstrate value.
进阶How do you prevent repair-rate manipulation?怎么防止修复率被操纵？
ENGLISH ANSWER
Definition fixed before pilot. Planned handoffs excluded. Threshold and volume set before results known. Model/data/config changes linked and recorded. Production, acceptance and authorisation roles are separate.
Q&A · D类 — PMBOK与治理

PMBOK & Governance

必背Which PMBOK knowledge areas does the plan address?涉及哪些PMBOK知识领域？
ENGLISH ANSWER
Six areas are central. Integration: Recovery Lead chairs change authority; decision points and the Week 4 evidence condition reconnect iterative work to executive governance. Scope: narrowed WBS, accepted enquiry catalogue, no informal expansion. Schedule: 2+2+4 sequence with Sponsor decisions at Week 2 and Week 8 and an internal entry condition at Week 4. Cost: bottom-up ETC from narrowed WBS, sunk-cost discipline, net-benefit test. Quality: regression testing, acceptance criteria, repair measurement. Risk: RAID treatment protocol with named owners, AI-control sign-off.
中文逻辑
记住6个领域各一个Atlas措施。Sponsor决策在Today/Week 2/Week 8；Week 4是内部准入条件不是Sponsor决策。
高频How does the change authority prevent informal refinements?变更控制怎么防止非正式"优化"？
ENGLISH ANSWER
Recovery Lead chairs the change authority. Product Owner protects scope. Any proposed change is assessed for impact on cost, quality, schedule and risk. No new features, flows or mappings enter during the pilot.
中文逻辑
之前变更被当作refinement绕过审批。现在所有变更必须经过正式影响评估。
高频How does the plan reconnect agile delivery with milestone governance?方案怎么把agile和milestone治理重新连接？
ENGLISH ANSWER
Technical work may remain iterative, but demos no longer count as completion. Each iteration sits inside baselines, acceptance evidence, change control and decision gates. The gates provide structured evidence and decision points that reconnect iterative delivery to milestone-based governance.
中文逻辑
技术工作可以继续迭代，但demo不再等于完成。每次迭代必须在baseline、验收证据和变更控制框架内进行。
进阶How do the three root causes interact?三个根因怎么互相影响？
ENGLISH ANSWER
Informal changes altered the work without updating the baseline. Untreated risks allowed the exposure to persist. Without an end-to-end owner, nobody was accountable for proving that the changed service remained acceptable. Each failure reinforced the others.
进阶Why trust vendor-produced regression evidence?为什么信任供应商出的回归测试证据？
ENGLISH ANSWER
The vendor does not approve its own evidence. Engineering and vendor produce it. Product Owner accepts coverage. Data Team controls the dataset. Recovery Lead authorises entry. No single party both produces and accepts.
进阶Why is Week 4 an internal gate, not a Sponsor decision?为什么Week 4是内部关卡不是Sponsor决策？
ENGLISH ANSWER
Week 4 is an operational entry condition verifying data, regression evidence, measurement logging and AI controls. No new funding or strategy decision is needed. Sponsor decisions remain at Today, Week 2 and Week 8.
Q&A · E类 — 关停与例外

Closure & Bounded Exception

必背What happens if the project closes?关停怎么处理？
ENGLISH ANSWER
Stop new development and expansion. Keep the approved evidence and artefacts. Transfer only work that has passed readiness and compliance checks. Retire everything else. Even closure produces a measured result.
中文逻辑
关停≠浪费。关停流程也产出有价值的东西：受控的测量结果、已验收的文档、通过合规的可转移工作。
高频Why allow one bounded exception?为什么允许一次有限例外？
ENGLISH ANSWER
Closure remains the default. The exception is not automatic. It exists for one specific residual gap: no new scope, fully costed, one fixed deadline, no further extension, Sponsor approval.
中文逻辑
关停是默认。例外不是自动的——5个条件必须同时满足。防止开放式延期。
高频What does "scale" mean?"扩展"具体指什么？
ENGLISH ANSWER
Broader use only within the accepted simple-support catalogue, approved data boundary and operating model. It does not reopen complex conversations, analytics or multi-region rollout.
中文逻辑
Scale≠回到原始范围。只在已验收的simple support边界内扩大使用。超出需要新的business case。
高频Does net benefit include the $4.1M already spent?净效益算上已花的$4.1M吗？
ENGLISH ANSWER
The $4.1M is sunk cost. It explains urgency but should not determine the continuation decision. Net benefit compares incremental recovery cost against future operational value.
中文逻辑
$4.1M是沉没成本。决策看的是未来：恢复增量成本 vs 未来运营价值。
Q&A · F类 — 矛盾证据

Reconciling Conflicting Evidence

策略：不选边站，指向evidence gates。

必背Green-Amber report vs. staff confidence — which should the Sponsor trust?报告说绿-琥珀但员工没信心，信谁？
ENGLISH ANSWER
The formal report records milestone progress. Operational interviews reveal unresolved stability, readiness and accountability concerns. The recovery plan does not choose one narrative. It introduces evidence gates that reconcile them with verified data.
中文逻辑
不替Sponsor选一边。我们引入evidence gates，用可验证的数据来代替两个互相矛盾的叙述。
必背The vendor says the platform is stable. Why still require stabilisation?供应商说平台稳定，为什么还要做稳定化？
ENGLISH ANSWER
The vendor may be describing the core platform. Atlas must prove end-to-end performance under ORG's changing data, configuration and use cases. Barry Styles said he would not trust the system to behave consistently under pressure. Regression evidence tests the integrated service, not the platform in isolation.
中文逻辑
供应商说的可能是核心平台。我们测试的是端到端表现——在ORG不断变化的数据和配置下系统是否稳定。
高频Atlas is within budget. Why call it financially unhealthy?还在预算内，为什么说财务不健康？
ENGLISH ANSWER
Being below the original ceiling does not prove affordability. Stabilisation, testing, readiness and transition remain incomplete. The remaining cost is unknown until the bottom-up ETC is completed. Arianna Venti noted that being technically inside the budget does not make the project financially healthy.
中文逻辑
没超预算≠钱够花。恢复工作还没定价，剩余成本是未知数。引用Arianna Venti的话作为evidence。
高频Simple support performs reasonably well. Why not scale immediately?Simple support表现还行，为什么不直接推广？
ENGLISH ANSWER
Reasonably well is not acceptance evidence. Atlas still needs controlled data, regression evidence, repair measurement, readiness sign-off and a positive net-benefit result.
中文逻辑
"还行"不等于验收证据。没有受控数据、回归测试、修复衡量、就绪签字和正效益结果就不能推广。
高频Why can CPI not price the recovery plan?为什么不能用CPI估算恢复成本？
ENGLISH ANSWER
CPI measures historical efficiency against the original baseline. After narrowing, the remaining work is no longer comparable. We need a bottom-up ETC from the narrowed WBS.
中文逻辑
CPI是基于原始范围的历史效率。缩窄后工作完全不同，需要从新WBS做bottom-up重新估算。
Q&A · G类 — 课程计算 DERIVED

Course Calculations

高频Why does the derived ETC show $1.82M when only ~$0.4M remains?衍生ETC $1.82M vs 剩余$0.4M？
ENGLISH ANSWER
The $0.4M is nominal budget headroom. The $1.82M is a CPI-based scenario for completing the original scope if historical efficiency continues. It is not our recovery funding request. Once scope changes, Finance rebuilds the ETC bottom-up from the narrowed WBS.
中文逻辑
$0.4M是名义额度，$1.82M是"如果继续原始范围+历史效率"的假设性估算。我们不要$1.82M——我们缩窄范围让Finance在Week 2重新算。
高频What does TCPI ≈ 3.46 tell the Sponsor?TCPI 3.46说明什么？
ENGLISH ANSWER
Atlas would need to improve from CPI 0.76 to about 3.46 to finish the original scope within the remaining budget. That is an extreme change. It shows why the original baseline is not credible and why scope must be narrowed before rebuilding the ETC.
中文逻辑
从0.76跳到3.46——效率要翻4.5倍。说明原始基线在财务上不可信，必须先缩窄范围。
进阶If historical runway is ~5 weeks, how can recovery run 8 weeks?历史跑道约5周，怎么做8周？
ENGLISH ANSWER
The five-week figure is a rough extrapolation from historical average spend. Narrowing changes the work and potentially the cost rate. We do not use it as a funding conclusion. Week 2 determines whether eight weeks is genuinely costed and affordable.
进阶Does EV ≈ $3.116M mean Atlas created $3.116M of business value?EV $3.116M等于创造了$3.116M的商业价值吗？
ENGLISH ANSWER
No. EV is the budgeted value of project work completed under the baseline. It is not realised savings, customer value or net operational benefit. That distinction is exactly why the Week 8 benefits test is still required.
进阶Why use derived metrics if the plan rejects CPI-based forecasting?方案不用CPI预测，为什么还算衍生指标？
ENGLISH ANSWER
We use them diagnostically, not as the recovery estimate. They show why continuing the original baseline is financially implausible. The authoritative estimate is the Week 2 bottom-up ETC.
速答题

一句话能答的题

"What is the single most important recovery action?"
Narrowing the scope — every later control depends on having a bounded, testable target.
缩窄范围——后面所有控制依赖一个有边界的可测试目标。
"What value does the plan create even if Atlas closes?"
It converts uncertainty into evidence. Management receives a controlled result instead of another demonstration without acceptance evidence.
把不确定性转化为证据。
"Who decides whether Atlas continues?"
The Sponsor. The recovery team produces evidence and a recommendation; it does not make the investment decision.
Sponsor。团队出证据，不做投资决策。
"What is the most important benefit of narrowing?"
It creates a bounded target that the team can stabilise and management can evaluate.
有边界的目标，可稳定、可评估。
"What is the single most important control?"
The Week 4 entry condition — it prevents a pilot from starting without accepted evidence.
Week 4准入条件。
"What is the biggest assumption?"
That sufficient capacity can be committed and funded. The plan tests this at Week 2.
能否在Week 2确保人力和资金。
"What would make you recommend closure immediately?"
Failure to secure capacity or funding at Week 2, inability to establish an accepted dataset, or an uncontrollable AI risk.
Week 2拿不到人或钱、数据集不达标、AI风险不可控。
"Why not request the derived $1.82M immediately?"
That assumes original scope and historical CPI. We narrow first. Week 2 prices the bounded recovery.
$1.82M假设原始范围不变。先缩窄，Week 2定价。
禁区

高风险表述

以下表述容易暴露逻辑错误、引发追问或削弱答案。

PMBOK 速查

6个关键PMBOK管理领域与Atlas应用

1. Integration Management 整合管理
问题：agile-style delivery与milestone-based治理和预测之间缺乏有效衔接
Fix: Recovery Lead chairs change authority. Decision points and the Week 4 evidence condition reconnect iterative work to executive governance.
2. Scope Management 范围管理
问题：WBS未实质性修订；变更被当作refinement绕过正式审批
Fix: narrowed WBS, accepted enquiry catalogue, no informal expansion.
3. Schedule Management 进度管理
问题：SPI=0.68；agile-style delivery与milestone-based治理缺乏有效衔接
Fix: 2+2+4 sequence with Sponsor decisions at Week 2 and Week 8 and an internal entry condition at Week 4.
4. Cost Management 成本管理
问题：CPI=0.76；恢复成本未知直到bottom-up ETC完成
Fix: bottom-up ETC from narrowed WBS. Nominal headroom not assumed sufficient. Net-benefit test. Sunk-cost discipline.
5. Quality Management 质量管理
问题：demo代替hardening；验收标准不明确或不完整；数据刷新后regression
Fix: data validation, version control, regression testing, acceptance criteria, repair measurement. Prevention and appraisal costs reduce later failure costs.
6. Risk Management 风险管理
问题：RAID items未被强制处理（一项挂了6周无enforced treatment owner）；AI风险后期才加入
Fix: RAID treatment protocol with named owners. Critical risks treated, accepted or escalated. Risk/Compliance signs off AI controls before pilot and before scale.
团队协作

Q&A现场规则

1️⃣
第一个人立刻给结论
Answer immediately
📌
跟一个case事实
Add one case fact
🔧
解释对应control或gate
Name the relevant control or gate
🚧
说清边界然后停
State boundary, stop
➕
第二个人只补充一个新信息
ONE new fact, implication or PMBOK link
🚫
绝不三个人答同一题
❌
不编数据
Do not invent costs, thresholds or volumes
🔍
不确定→指向gate
Name who confirms and at which gate
⚖️
区分case证明的/方案提出的/gate待定的
⏱️
20–35秒（40–75词；复杂问题最多≈90词）
附录

Case Study 人物速查

只有引号内是原话。引用人名拿evidence分。

Taylor Quick — COO & Executive Sponsor
"I feel like I'm getting two versions of this project."
Best use: conflicting executive information, Sponsor decision need
Billy English — PMO Analyst
"A risk that sits in the RAID log for six weeks with no treatment plan… that's just sitting there with a status next to it."
Best use: RAID enforcement failure, milestone-vs-agile governance gap
Arianna Venti — Finance Business Partner
"If the recovery plan is just, you know, spend more and keep going… that won't be enough."
Best use: within budget ≠ financial health
Ted Sheran — Head of Customer Experience
"They're not just solving the issue… they're having to basically repair the interaction."
Best use: repair burden, CX quality, origin of "repair event"
Barry Styles — Internal AI Engineer
"I wouldn't trust it to behave the same way twice in a row under pressure."
Best use: instability, why regression testing matters
Serena Gomez — Product Owner
"Nothing ever really feels finished. It's always close, or ready but needing one more pass."
Best use: unclear acceptance criteria and definition of done
Bruno Bars — Risk and Compliance Officer
"The outputs sound confident, so people assume they're more certain than they actually are."
Best use: AI auditability, explainability, overconfident outputs