planning-under-uncertainty（不确定性下的规划）

Planning Under Uncertainty

Scope

Covers

• Turning ambiguity into an executable plan via hypotheses, experiments, and decision triggers
• Diagnosing “what’s actually happening” before acting (especially in crisis / wartime situations)
• Using data as a compass (directional checks) rather than a GPS (false precision)
• Building buffers and contingencies so the plan survives chaos
• Setting a cadence for learning, decision-making, and stakeholder communication

When to use

• “We need a plan, but the requirements are unclear and the outcome is uncertain.”
• “Create a hypothesis-driven plan (experiments + decision rules) for this initiative.”
• “We’re in a crisis (drop in retention/revenue/reliability) and need a wartime diagnosis + action plan.”
• “Help us build contingencies, buffers, and pivot triggers before we commit.”

When NOT to use

• You don’t agree on the underlying problem/opportunity (use problem-definition).
• You need to choose what to do among many options (use prioritizing-roadmap).
• You already have a clear plan and only need dates/milestones and stakeholder cadence (use managing-timelines).
• You need a decision-ready PRD/spec for build execution (use writing-prds / writing-specs-designs).

Inputs

Minimum required

• The initiative context and desired outcome (“what are we trying to change?”)
• Time horizon and urgency (wartime vs peacetime)
• Constraints/guardrails (quality, compliance, brand, budget, “must not worsen” metrics)
• Stakeholders and decision rights (who decides pivot/stop/scale?)
• Top unknowns/assumptions (what would change the plan?)
• Current signals (what data exists; what feels true but unproven?)

Missing-info strategy

• Ask up to 5 questions from references/INTAKE.md.
• If answers aren’t available, proceed with explicit assumptions and list Open questions that could change the plan.

Outputs (deliverables)

Produce an Uncertainty Planning Pack in Markdown (in-chat; or as files if the user requests), containing:

1. Decision frame (objective, “why now”, success + guardrails, time horizon, decision owner)
2. Uncertainty map (assumptions/unknowns, confidence, impact, validation plan)
3. Hypotheses + experiment portfolio (what we’ll learn, how, and what decision it enables)
4. Plan v0 with buffers + contingencies (phases/options, triggers, fallbacks, pivot criteria)
5. Cadence + comms (learning review ritual, update template, decision log)
6. Risks / Open questions / Next steps (always included)

Templates: references/TEMPLATES.md Expanded guidance: references/WORKFLOW.md

Workflow (7 steps)

1) Intake + mode setting (wartime vs peacetime)

• Inputs: User request; references/INTAKE.md.
• Actions: Clarify urgency, stakes, and what decision is needed. Decide whether you’re in diagnosis-first wartime mode or exploration peacetime mode.
• Outputs: Short decision frame draft + mode declaration.
• Checks: You can state: “We’re optimizing for <fast stabilization / learning / growth>. The decision we need by is <pivot/stop/scale/commit>.”

2) Diagnose reality (humility first)

• Inputs: Current signals, anecdotes, dashboards, incident reports, qualitative inputs.
• Actions: Separate symptoms from hypotheses. Write 3–7 plausible explanations, and identify what evidence would falsify each. Avoid prematurely picking a favorite story.
• Outputs: “What we know / don’t know” + initial hypothesis set.
• Checks: At least one hypothesis contradicts the team’s initial intuition (to reduce confirmation bias).

3) Build the uncertainty map (assumptions → validation plan)

• Inputs: Hypotheses; constraints; stakeholders; time horizon.
• Actions: Create an uncertainty map of assumptions/unknowns with confidence and impact; prioritize the top items that would change the plan.
• Outputs: Uncertainty map table + prioritized “top 5 unknowns”.
• Checks: Every top unknown has a clear validation method and an owner.

4) Define hypotheses + decision rules (learning over “wins”)

• Inputs: Top unknowns; success/guardrails; risk tolerance.
• Actions: Turn unknowns into testable hypotheses. For each hypothesis, define: expected learning, success signal(s), guardrails, and the decision the result enables (stop/pivot/scale).
• Outputs: Hypothesis statements + decision rules.
• Checks: Each hypothesis ties to a decision; “winning” is defined as learning, not just positive results.

5) Design a reproducible testing process (many shots at bat)

• Inputs: Hypothesis set; available tools; team capacity.
• Actions: Create an experiment portfolio that balances speed vs confidence (smoke tests, prototypes, A/Bs, customer calls, operational drills). Set a cadence to run and review tests continuously.
• Outputs: Experiment portfolio table + review cadence.
• Checks: At least 1 fast test can run within the next 1–2 weeks (or faster in wartime).

6) Turn learning into a plan with buffers, contingencies, and triggers

• Inputs: Experiment portfolio; constraints; dependencies; timeline needs.
• Actions: Draft Plan v0 with phases/options; add buffers; define contingencies and explicit triggers for pivot/rollback/escalation. Use data as a compass: focus on directional signals and early warnings, not false certainty.
• Outputs: Plan v0 + buffer/contingency section + trigger list.
• Checks: There is a clear “if X happens, we will do Y” for the top risks/unknowns.

7) Quality gate + finalize

• Inputs: Full draft pack.
• Actions: Run references/CHECKLISTS.md and score with references/RUBRIC.md. Ensure Risks / Open questions / Next steps exist with owners and time bounds.
• Outputs: Final Uncertainty Planning Pack.
• Checks: A stakeholder can approve the plan async and the team can execute without re-litigating the ambiguity.

Quality gate (required)

• Use references/CHECKLISTS.md and references/RUBRIC.md.
• Always include: Risks, Open questions, Next steps.

Examples

Example 1 (ambiguous initiative): “We think onboarding is hurting conversion, but we’re not sure why. Create an uncertainty plan with hypotheses, experiments, and pivot triggers.” Expected: an uncertainty map + experiment portfolio (qual + quant) + a Plan v0 that commits to learning milestones, not premature delivery dates.

Example 2 (wartime): “Retention dropped 15% this week after a release. We need a wartime plan: diagnose root causes, run rapid tests, and decide whether to rollback or patch.” Expected: diagnosis-first workflow with falsifiable hypotheses, tight guardrails, and explicit rollback/escalation triggers.

Boundary example: “Write a full PRD for Feature X.” Response: clarify uncertainty first (this skill), then use writing-prds once the hypotheses, constraints, and decision gates are clear.