Back to Details

predictive-planning

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Predictive Planning

预测性规划

Treat a plan as a forecast plus a policy:
If we take these actions, under these assumptions, we expect this outcome within these time and cost limits. If reality differs, we will respond in a defined way.
Use only the detail that can change a decision.
将计划视为预测加策略：
如果我们采取这些行动，基于这些假设，我们预期在既定时间和成本限制内达成该结果。若实际情况与假设不符，我们将按既定方式应对。
仅保留可能影响决策的细节。

Route

实施路径

For routine, reversible work, use this minimal forecast-policy and stop:
If [action], [owner] has [probability]% chance of [scoreable outcome] by [date]. If [trigger], then [response].
For work with meaningful stakes or uncertainty, use the full method. First establish this operating contract:
  • Store each timestamped forecast and revision in one durable record.
  • Freeze the success event, measurement rule, and evidence source before the decision.
  • Name the forecast author and the decision owner who can apply response rules.
  • Schedule the review before the next expensive or irreversible commitment.
  • Assign the forecast to a group of comparable forecasts made at a similar time horizon.
  • When approval incentives could distort the estimate, obtain an independent forecast before choosing the decision forecast.
If a control is missing, make establishing it the first action instead of inventing precision.
对于常规、可逆转的工作，使用以下极简预测-策略即可：
如果执行[行动]，[负责人]有[概率]%的机会在[日期]前达成[可量化结果]。若出现[触发条件]，则采取[应对措施]。
对于存在重大风险或不确定性的工作，使用完整方法。首先确立以下操作约定：
  • 将每个带时间戳的预测及修订版本存储在一个持久化记录中。
  • 在做出决策前，冻结成功事件、衡量规则和证据来源的定义。
  • 指定预测作者以及可执行应对规则的决策负责人。
  • 在进行下一项高成本或不可逆承诺前安排评审。
  • 将该预测归入一组时间范围相似的同类预测中。
  • 当审批激励可能导致估算失真时，在选定决策用预测前获取独立预测。
若缺少某项控制措施，应先建立该措施，而非盲目追求精度。

Method

方法步骤

1. Define the predicted result

1. 定义预测结果

Define one scoreable success event: the outcome reaches a stated threshold by a stated deadline without exceeding a stated cost or effort limit. Attach the probability of success to that complete event.
Also state:
  • The owner
  • P50 and P90 completion dates when timing is uncertain
  • P50 and P90 cost or effort when cost is uncertain
  • Why the proposed actions should produce the outcome
Forecast completion date and cost separately. P50 means a 50% chance of finishing by the date or within the cost estimate. P90 means a 90% chance, so a P90 date or cost cannot be earlier or lower than P50.
These are separate forecasts: two P90 estimates do not imply a 90% chance of meeting both. The probability of a success event requiring several conditions cannot exceed the probability of any required condition.
Describe the result rather than the activity. Replace “ship a dashboard” with the change the dashboard should cause.
定义一个可量化的成功事件：在规定的截止日期前达成既定结果阈值，且不超出规定的成本或工作量限制。为该完整事件赋予成功概率。
同时说明：
  • 负责人
  • 当时间存在不确定性时，给出P50和P90完成日期
  • 当成本存在不确定性时，给出P50和P90成本或工作量
  • 拟采取的行动为何能产生预期结果
分别预测完成日期和成本。P50表示有50%的概率在该日期前完成或控制在成本估算内。P90表示有90%的概率，因此P90对应的日期或成本不能早于或低于P50的数值。
这些是独立的预测：两个P90估算并不意味着同时满足两者的概率为90%。需要多个条件的成功事件，其概率不会超过任一必要条件的概率。
描述结果而非活动。将“交付仪表盘”替换为该仪表盘应带来的改变。

2. Establish the probability basis

2. 确立概率依据

Find comparable past work. Use its actual duration, cost, and success rate as the starting point. Record each adjustment and the evidence for it.
When comparable evidence is unavailable, label the forecast as judgment-based. Do not present unsupported precision as measured knowledge. Record any independent forecast separately before choosing the probability used for the decision.
寻找同类过往工作。以其实际时长、成本和成功率为起点。记录每一项调整及对应的证据。
当无同类证据时，将该标记为基于判断的预测。不得将无依据的精度伪装成可衡量的知识。在选定决策用概率前，单独记录任何独立预测。

3. Expose critical uncertainty

3. 暴露关键不确定性

Record only assumptions and dependencies that are both important and uncertain. State the evidence and confidence for each one.
仅记录重要且不确定的假设和依赖关系。说明每项假设的证据和置信度。

4. Test the most valuable uncertainty first

4. 优先测试最具价值的不确定性

Prioritize a test when the assumption is important and uncertain, the result could change the plan, and the test is cheap. Put the best such test before expensive or irreversible work.
Sequence work by useful information gained, not by convenience.
当某项假设重要且不确定、测试结果可能改变计划，且测试成本低廉时，优先进行测试。将此类最优测试安排在高成本或不可逆工作之前。
按获取有用信息的价值排序工作，而非按便利性。

5. Choose the policy and actions

5. 选择策略与行动

Before committing, compare the proposed policy’s expected benefit, cost, and downside with doing nothing and the best practical alternative. Choose the policy with the best supported tradeoff.
List its smallest complete sequence of actions. Connect each action to the assumption it tests, evidence it creates, or result it causes. Remove work that does none of these.
在做出承诺前，比较拟议策略的预期收益、成本和风险，与不采取行动及最佳可行替代方案的情况。选择有最可靠权衡依据的策略。
列出其最小完整行动序列。将每个行动与它所验证的假设、生成的证据或带来的结果关联。移除无法实现上述任一目标的工作。

6. Turn milestones into predicted evidence

6. 将里程碑转化为可预测的证据

Give each checkpoint an observable expected state, not only a task or date.
For example:
By Tuesday, five users can complete onboarding unaided, with at least four succeeding in under three minutes.
为每个检查点设定可观察的预期状态，而非仅设定任务或日期。
示例：
截至周二，5名用户可独立完成入门流程，且至少4人能在3分钟内完成。

7. Define response rules in advance

7. 提前定义应对规则

For each important signal, state the response:
If [observed condition], then [continue, revise, reduce scope, pause, or stop].
Choose thresholds before observing the result. Name the person authorized to apply each response and when they must decide.
针对每个重要信号，说明应对方式：
若出现[观测到的情况]，则[继续执行、修订计划、缩小范围、暂停或终止]。
在观测结果前选定阈值。指定有权执行每项应对措施的人员及决策时限。

8. Update the forecast

8. 更新预测

At each checkpoint:
  1. Compare predicted evidence with actual evidence.
  2. Identify the assumption that changed.
  3. Update the probability and the P50 and P90 completion and cost estimates.
  4. Continue, change course, or stop according to the response rules.
  5. Append the evidence and revision to the durable record without replacing earlier forecasts.
在每个检查点：
  1. 比较预测证据与实际证据。
  2. 识别已发生变化的假设。
  3. 更新成功概率以及P50、P90完成日期和成本估算。
  4. 根据应对规则继续执行、调整方向或终止。
  5. 将证据和修订内容追加到持久化记录中，不得替换早期预测。

9. Learn from prediction error

9. 从预测误差中学习

Afterward, compare predicted and actual outcome, duration, cost, and failed assumptions. Score the frozen forecast used for the decision; keep later revisions separate and compare forecasts made at similar time horizons.
For decimal probability 
p
, set 
y = 1
when the complete success event occurs and 
y = 0
otherwise. Calculate the Brier score 
BS = (p - y)²
; lower is better. Across repeated forecasts, check that events forecast near 70% occur about 70% of the time, and that actual completion dates and costs fall at or below P50 and P90 estimates about 50% and 90% of the time.
Use recurring errors to update local base rates and future estimates. Judge forecast quality separately from outcome quality. One result cannot establish calibration: a good probabilistic forecast can fail, and a poor plan can succeed through luck.
事后比较预测与实际结果、时长、成本，以及未达成的假设。对用于决策的冻结预测进行评分；将后续修订单独保存，并比较同一时间范围做出的预测。
对于小数概率
p
，当完整成功事件发生时设
y = 1
，否则设
y = 0
。计算Brier评分
BS = (p - y)²
；分数越低越好。在多次预测中，验证预测概率约为70%的事件实际发生概率是否约为70%，以及实际完成日期和成本是否约有50%和90%的概率落在P50和P90估算值以内。
利用反复出现的误差更新本地基准率和未来估算。将预测质量与结果质量分开评判。单一结果无法证明校准效果：好的概率预测也可能失败，而差的计划也可能因运气成功。

Output

输出格式

For routine work, output the minimal forecast-policy from Route. For consequential uncertainty, use the relevant sections of this template:
markdown
undefined
对于常规工作，输出实施路径中的极简预测-策略即可。对于存在重大不确定性的工作，使用以下模板的相关部分：
markdown
undefined

[Plan name]

[计划名称]

Operating controls

操作控制项

  • Durable record:
  • Forecast frozen at:
  • Measurement rule and evidence source:
  • Forecast author:
  • Decision owner and authority:
  • Calibration group and forecast horizon:
  • Independent forecast when incentives require one:
  • 持久化记录：
  • 预测冻结时间：
  • 衡量规则与证据来源：
  • 预测作者：
  • 决策负责人及权限：
  • 校准组与预测时间范围：
  • 激励机制要求时的独立预测：

Forecast

预测内容

  • Scoreable success event: [outcome threshold] by [deadline] within [cost or effort limit]
  • Probability of success:
  • Probability basis and adjustments:
  • Owner:
  • P50 / P90 completion:
  • P50 / P90 cost or effort:
  • Comparable past cases:
  • Causal rationale:
  • 可量化成功事件：[结果阈值]在[截止日期]前达成，且不超出[成本或工作量限制]
  • 成功概率：
  • 概率依据及调整：
  • 负责人：
  • P50 / P90完成日期：
  • P50 / P90成本或工作量：
  • 同类过往案例：
  • 因果逻辑：

Policy choice

策略选择

  • Proposed policy: [expected benefit, cost, and downside]
  • Doing nothing: [expected benefit, cost, and downside]
  • Best practical alternative: [expected benefit, cost, and downside]
  • Decision:
  • 拟议策略：[预期收益、成本和风险]
  • 不采取行动：[预期收益、成本和风险]
  • 最佳可行替代方案：[预期收益、成本和风险]
  • 决策结果：

Critical assumptions

关键假设

  • [Assumption]: [evidence, confidence, or dependency]
  • [假设内容]：[证据、置信度或依赖关系]

Cheapest next test

成本最低的下一项测试

  • Belief being tested:
  • Test:
  • Expected evidence:
  • 待验证的观点：
  • 测试内容：
  • 预期证据：

Actions

行动项

  1. [Action] — [owner] — [expected effect or evidence]
  1. [行动内容] — [负责人] — [预期效果或证据]

Milestones

里程碑

  • [Checkpoint]: [date or range] — [expected evidence]
  • [检查点]：[日期或范围] — [预期证据]

Response rules

应对规则

  • If [condition], then [authorized owner] takes [response] by [decision time].
  • 若出现[情况]，则[授权负责人]在[决策时限]前采取[应对措施]。

Review and scoring

评审与评分

  • Next review before commitment:
  • Outcome scoring date:
  • Actual outcome source:
  • Brier score when resolved:

For an existing plan, retain its useful content while converting activities, dates, and assumptions into explicit forecasts and response rules.
  • 下一次承诺前的评审时间：
  • 结果评分日期：
  • 实际结果来源：
  • 完成后的Brier评分：

对于现有计划，保留其有用内容，同时将活动、日期和假设转换为明确的预测和应对规则。