Back to Details

heuristics-and-checklists

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Heuristics and Checklists

Heuristics and Checklists

Table of Contents

目录

Purpose

用途

Heuristics and Checklists provides practical frameworks for fast decision-making through mental shortcuts (heuristics) and systematic error prevention through structured procedures (checklists). This skill guides you through designing effective heuristics for routine decisions, creating checklists for complex procedures, and understanding when shortcuts work vs. when they lead to biases.
Heuristics and Checklists 通过心智捷径（启发式方法）提供快速决策的实用框架，并通过结构化流程（检查表）实现系统性错误预防。该方法将引导你为常规决策设计有效的启发式规则，为复杂流程创建检查表，同时帮助你理解何时捷径有效、何时会导致偏差。

When to Use

适用场景

Use this skill when:
  • Time-constrained decisions: Need to decide quickly without full analysis
  • Routine choices: Repetitive decisions where full analysis is overkill (satisficing)
  • Error prevention: Complex procedures where mistakes are costly (surgery, software deployment, flight operations)
  • Checklist design: Creating pre-flight, pre-launch, or pre-deployment checklists
  • Cognitive load reduction: Simplifying complex decisions into simple rules
  • Bias mitigation: Understanding when heuristics mislead (availability, anchoring, representativeness)
  • Knowledge transfer: Codifying expert intuition into transferable rules
  • Quality assurance: Ensuring critical steps aren't skipped
  • Onboarding: Teaching newcomers reliable decision patterns
  • High-stakes procedures: Surgery, aviation, nuclear operations, financial trading
Trigger phrases: "heuristics", "rules of thumb", "mental models", "checklists", "error prevention", "cognitive biases", "satisficing", "quick decision", "standard operating procedure"
在以下场景使用该方法：
  • 时间受限的决策：需要在无充分分析的情况下快速做出决定
  • 常规选择：重复性决策，充分分析属于过度投入（满意原则）
  • 错误预防：错误代价高昂的复杂流程（如手术、软件部署、飞行操作）
  • 检查表设计：创建飞行前、发布前或部署前检查表
  • 认知负荷降低：将复杂决策简化为简单规则
  • 偏差缓解：理解启发式方法何时会产生误导（可得性偏差、锚定效应、代表性偏差）
  • 知识传递：将专家直觉转化为可传递的规则
  • 质量保证：确保关键步骤不被遗漏
  • 新员工入职：向新人传授可靠的决策模式
  • 高风险流程：手术、航空、核操作、金融交易
触发关键词："heuristics"、"rules of thumb"、"mental models"、"checklists"、"error prevention"、"cognitive biases"、"satisficing"、"quick decision"、"standard operating procedure"

What Is It?

什么是Heuristics and Checklists？

Heuristics and Checklists combines two complementary approaches for practical decision-making and error prevention:
Core components:
  • Heuristics: Mental shortcuts or rules of thumb that simplify decisions (e.g., "recognition heuristic": choose the option you recognize)
  • Checklists: Structured lists ensuring critical steps completed in order (aviation pre-flight, surgical safety checklist)
  • Fast & Frugal Trees: Simple decision trees with few branches, good enough decisions with minimal information
  • Satisficing: "Good enough" decisions (Simon) vs. exhaustive optimization
  • Bias awareness: Recognizing when heuristics fail (availability bias, anchoring, representativeness)
  • Error prevention: Swiss cheese model, forcing functions, fail-safes
Quick example:
Scenario: Startup CEO deciding whether to hire a candidate after interview.
Without heuristics (exhaustive analysis):
  • Compare to all other candidates (takes weeks)
  • 360-degree reference checks (10+ calls)
  • Skills assessment, culture fit survey, multiple rounds
  • Analysis paralysis, miss good candidates to faster competitors
With heuristics (fast & frugal):
  1. Recognition heuristic: Have they worked at company I respect? (Yes → +1)
  2. Take-the-best: What's their track record on most important skill? (Strong → +1)
  3. Satisficing threshold: If 2/2 positive, hire. Don't keep searching for "perfect" candidate.
Outcome: Hired strong candidate in 3 days instead of 3 weeks. Not perfect, but good enough and fast.
Checklist example (Software Deployment):
Pre-Deployment Checklist:
☐ All tests passing (unit, integration, E2E)
☐ Database migrations tested on staging
☐ Rollback plan documented
☐ Monitoring dashboards configured
☐ On-call engineer identified
☐ Stakeholders notified of deployment window
☐ Feature flags configured for gradual rollout
☐ Backups completed
Benefit: Prevents missing critical steps. Reduces deployment failures by 60-80% (empirical data from aviation, surgery, software).
Core benefits:
  • Speed: Heuristics enable fast decisions under time pressure
  • Cognitive efficiency: Reduce mental load, free capacity for complex thinking
  • Error reduction: Checklists catch mistakes before they cause harm
  • Consistency: Standardized procedures reduce variance in outcomes
  • Knowledge codification: Capture expert intuition in transferable form
Heuristics and Checklists 结合了两种互补的方法，用于实用决策和错误预防：
核心组件:
  • Heuristics（启发式方法）：简化决策的心智捷径或经验法则（例如，"识别启发式"：选择你熟悉的选项）
  • Checklists（检查表）：结构化列表，确保按顺序完成所有关键步骤（如航空飞行前检查表、手术安全检查表）
  • Fast & Frugal Trees（快速节俭决策树）：分支较少的简单决策树，用最少信息做出足够好的决策
  • Satisficing（满意原则）：做出"足够好"的决策（西蒙提出），而非穷尽式优化
  • 偏差意识：识别启发式方法失效的情况（可得性偏差、锚定效应、代表性偏差）
  • 错误预防：瑞士奶酪模型、强制函数、故障安全机制
快速示例:
场景：初创公司CEO在面试后决定是否录用候选人。
无启发式方法（穷尽分析）:
  • 与所有其他候选人对比（耗时数周）
  • 360度背景调查（10+次通话）
  • 技能评估、文化适配调查、多轮面试
  • 分析瘫痪，优秀候选人被更快的竞争对手抢走
使用启发式方法（快速节俭）:
  1. 识别启发式：他们是否在我认可的公司工作过？（是→+1）
  2. 取最优启发式：他们在最重要技能上的过往表现如何？（优秀→+1）
  3. 满意原则阈值：如果2项都满足，就录用。不要继续寻找"完美"候选人。
结果：3天内录用了优秀候选人，而非3周。虽非完美，但足够好且高效。
检查表示例（软件部署）:
Pre-Deployment Checklist:
☐ All tests passing (unit, integration, E2E)
☐ Database migrations tested on staging
☐ Rollback plan documented
☐ Monitoring dashboards configured
☐ On-call engineer identified
☐ Stakeholders notified of deployment window
☐ Feature flags configured for gradual rollout
☐ Backups completed
益处：防止遗漏关键步骤。据航空、手术、软件领域的实证数据，可将部署失败率降低60-80%。
核心优势:
  • 速度：启发式方法让你在时间压力下快速决策
  • 认知效率：减少心智负荷，腾出精力用于复杂思考
  • 错误减少：检查表在错误造成危害前发现问题
  • 一致性：标准化流程减少结果差异
  • 知识固化：将专家直觉转化为可传递的形式

Workflow

工作流程

Copy this checklist and track your progress:
Heuristics & Checklists Progress:
- [ ] Step 1: Identify decision or procedure
- [ ] Step 2: Choose approach (heuristic vs. checklist)
- [ ] Step 3: Design heuristic or checklist
- [ ] Step 4: Test and validate
- [ ] Step 5: Apply and monitor
- [ ] Step 6: Refine based on outcomes
Step 1: Identify decision or procedure
What decision or procedure needs simplification? Is it repetitive? Time-sensitive? Error-prone? See resources/template.md.
Step 2: Choose approach (heuristic vs. checklist)
Heuristic for decisions (choose option). Checklist for procedures (sequence of steps). See resources/methodology.md.
Step 3: Design heuristic or checklist
Heuristic: Define simple rule (recognition, take-the-best, satisficing threshold). Checklist: List critical steps, add READ-DO or DO-CONFIRM format. See resources/template.md and resources/template.md.
Step 4: Test and validate
Pilot test with sample cases. Check: Does heuristic produce good enough decisions? Does checklist catch errors? See resources/methodology.md.
Step 5: Apply and monitor
Use in real scenarios. Track outcomes: decision quality, error rate, time saved. See resources/template.md.
Step 6: Refine based on outcomes
Adjust rules based on data. If heuristic fails in specific contexts, add exception. If checklist too long, prioritize critical items. See resources/methodology.md.
Validate using resources/evaluators/rubric_heuristics_and_checklists.json. Minimum standard: Average score ≥ 3.5.
复制此检查表并跟踪进度：
Heuristics & Checklists Progress:
- [ ] Step 1: Identify decision or procedure
- [ ] Step 2: Choose approach (heuristic vs. checklist)
- [ ] Step 3: Design heuristic or checklist
- [ ] Step 4: Test and validate
- [ ] Step 5: Apply and monitor
- [ ] Step 6: Refine based on outcomes
步骤1：确定决策或流程
哪些决策或流程需要简化？是否具有重复性？是否时间敏感？是否容易出错？查看 resources/template.md。
步骤2：选择方法（启发式vs检查表）
启发式方法适用于决策（选择选项）。检查表适用于流程（步骤序列）。查看 resources/methodology.md。
步骤3：设计启发式规则或检查表
启发式：定义简单规则（识别、取最优、满意阈值）。检查表：列出关键步骤，添加READ-DO或DO-CONFIRM格式。查看 resources/template.md 和 resources/template.md。
步骤4：测试与验证
用样本案例进行试点测试。检查：启发式方法是否能做出足够好的决策？检查表是否能发现错误？查看 resources/methodology.md。
步骤5：应用与监控
在实际场景中使用。跟踪结果：决策质量、错误率、节省的时间。查看 resources/template.md。
步骤6：根据结果优化
根据数据调整规则。如果启发式方法在特定场景失效，添加例外情况。如果检查表过长，优先保留关键项。查看 resources/methodology.md。
使用 resources/evaluators/rubric_heuristics_and_checklists.json 进行验证。最低标准：平均得分≥3.5。

Common Patterns

常见模式

Pattern 1: Recognition Heuristic
  • Rule: Choose the option you recognize over the one you don't
  • Best for: Choosing between brands, cities, experts when quality correlates with fame
  • Example: "Which city is larger, Detroit or Milwaukee?" (Choose Detroit if only one recognized)
  • When works: Stable environments where recognition predicts quality
  • When fails: Advertising creates false recognition, niche quality unknown
Pattern 2: Take-the-Best Heuristic
  • Rule: Identify single most important criterion, choose based on that alone
  • Best for: Multi-attribute decisions with one dominant factor
  • Example: Hiring - "What's their track record on [critical skill]?" Ignore other factors.
  • When works: One factor predictive, others add little value
  • When fails: Multiple factors equally important, interactions matter
Pattern 3: Satisficing (Good Enough Threshold)
  • Rule: Set minimum acceptable criteria, choose first option that meets them
  • Best for: Routine decisions, time pressure, diminishing returns from analysis
  • Example: "Candidate meets 80% of requirements → hire, don't keep searching for 100%"
  • When works: Searching costs high, good enough > perfect delayed
  • When fails: Consequences of suboptimal choice severe
Pattern 4: Aviation Checklist (DO-CONFIRM)
  • Format: Perform actions from memory, then confirm each with checklist
  • Best for: Routine procedures with critical steps (pre-flight, pre-surgery, deployment)
  • Example: Pilot flies from memory, then reviews checklist to confirm all done
  • When works: Experts doing familiar procedures, flow state preferred
  • When fails: Novices, unfamiliar procedures (use READ-DO instead)
Pattern 5: Surgical Checklist (READ-DO)
  • Format: Read each step, then perform, one at a time
  • Best for: Unfamiliar procedures, novices, high-stakes irreversible actions
  • Example: Surgical team reads checklist aloud, confirms each step before proceeding
  • When works: Unfamiliar context, learning mode, consequences of error high
  • When fails: Expert routine tasks (feels tedious, adds overhead)
Pattern 6: Fast & Frugal Decision Tree
  • Format: Simple decision tree with 1-3 questions, binary choices at each node
  • Best for: Triage, classification, go/no-go decisions
  • Example: "Is customer enterprise? Yes → Assign senior rep. No → Is deal >$10k? Yes → Assign mid-level. No → Self-serve."
  • When works: Clear decision structure, limited information needed
  • When fails: Nuanced decisions, exceptions common
模式1：识别启发式
  • 规则：选择你熟悉的选项，而非不熟悉的
  • 最佳适用场景：在品牌、城市、专家之间选择，且质量与知名度相关时
  • 示例："哪个城市更大，底特律还是密尔沃基？"（如果只认识底特律，就选它）
  • 有效场景：稳定环境中，知名度可预测质量
  • 失效场景：广告制造虚假知名度、小众高质量选项不为人知时
模式2：取最优启发式
  • 规则：确定单一最重要的标准，仅据此选择
  • 最佳适用场景：多属性决策，但有一个主导因素时
  • 示例：招聘时——"他们在[关键技能]上的过往表现如何？"忽略其他因素。
  • 有效场景：一个因素具有预测性，其他因素价值不大
  • 失效场景：多个因素同等重要、存在交互影响时
模式3：满意原则（足够好阈值）
  • 规则：设定最低可接受标准，选择第一个满足标准的选项
  • 最佳适用场景：常规决策、时间压力、分析收益递减时
  • 示例："候选人满足80%的要求→录用，不要继续寻找100%符合的人"
  • 有效场景：搜索成本高，足够好优于延迟的完美
  • 失效场景：次优选择的后果严重时
模式4：航空检查表（DO-CONFIRM）
  • 格式：凭记忆执行操作，然后用检查表逐一确认
  • 最佳适用场景：包含关键步骤的常规流程（飞行前、手术前、部署前）
  • 示例：飞行员凭记忆操作，然后查看检查表确认所有步骤完成
  • 有效场景：专家执行熟悉的流程，偏好流畅状态时
  • 失效场景：新手、不熟悉的流程（改用READ-DO格式）
模式5：手术检查表（READ-DO）
  • 格式：阅读每个步骤，然后执行，逐一完成
  • 最佳适用场景：不熟悉的流程、新手、高风险不可逆操作
  • 示例：手术团队大声朗读检查表，确认每个步骤后再继续
  • 有效场景：陌生环境、学习阶段、错误后果严重时
  • 失效场景：专家执行常规任务（会觉得繁琐，增加负担）
模式6：快速节俭决策树
  • 格式：只有1-3个问题的简单决策树，每个节点为二元选择
  • 最佳适用场景：分诊、分类、是否推进的决策
  • 示例："客户是企业客户吗？是→分配资深代表。否→交易额是否超过1万美元？是→分配中级代表。否→自助服务。"
  • 有效场景：决策结构清晰、所需信息有限时
  • 失效场景：决策复杂、例外情况多时

Guardrails

约束准则

Critical requirements:
  1. Know when heuristics work vs. fail: Heuristics excel in stable, familiar environments with time pressure. They fail in novel, deceptive contexts (adversarial, misleading information). Don't use recognition heuristic when advertising creates false signals.
  2. Satisficing ≠ low standards: "Good enough" threshold must be calibrated. Set based on cost of continued search vs. value of better option. Too low → poor decisions. Too high → analysis paralysis.
  3. Checklists for critical steps only: Don't list every trivial action. Focus on steps that (1) are skipped often, (2) have serious consequences if missed, (3) not immediately obvious. Short checklists used > long checklists ignored.
  4. READ-DO for novices, DO-CONFIRM for experts: Match format to user expertise. Forcing experts into READ-DO creates resistance and abandonment. Let experts flow, confirm after.
  5. Test heuristics empirically: Don't assume rule works. Test on historical cases. Compare heuristic decisions to optimal decisions. If accuracy <80%, refine or abandon.
  6. Bias awareness is not bias elimination: Knowing availability bias exists doesn't prevent it. Heuristics are unconscious. Need external checks (checklists, peer review, base rates) to counteract biases.
  7. Update heuristics when environment changes: Rules optimized for past may fail in new context. Market shifts, technology changes, competitor strategies evolve. Re-validate quarterly.
  8. Forcing functions beat reminders: "Don't forget X" fails. "Can't proceed until X done" works. Build constraints (e.g., deployment script requires all tests pass) rather than relying on memory.
Common pitfalls:
  • ❌ Heuristic as universal law: "Always choose recognized brand" fails when dealing with deceptive advertising or niche quality.
  • ❌ Checklist too long: 30-item checklist gets skipped. Keep to 5-10 critical items max.
  • ❌ Ignoring base rates: "This customer seems like they'll buy" (representativeness heuristic) vs. "Only 2% of leads convert" (base rate). Use base rates to calibrate intuition.
  • ❌ Anchoring on first option: "First candidate seems good, let's hire" without considering alternatives. Set satisficing threshold, then evaluate multiple options.
  • ❌ Checklist as blame shield: "I followed checklist, not my fault" ignores responsibility to think. Checklists augment judgment, don't replace it.
  • ❌ Not testing heuristics: Assume rule works without validation. Test on past cases, measure accuracy.
关键要求:
  1. 了解启发式方法的有效与失效场景：启发式方法在稳定、熟悉的环境和时间压力下表现出色。在新颖、具有欺骗性的环境（对抗性、误导性信息）中会失效。当广告制造虚假信号时，不要使用识别启发式。
  2. 满意原则≠低标准："足够好"的阈值必须经过校准。根据持续搜索的成本与更好选项的价值来设定。阈值过低→决策质量差。阈值过高→分析瘫痪。
  3. 检查表仅包含关键步骤：不要列出每个琐碎操作。聚焦于以下步骤：(1) 经常被跳过的，(2) 遗漏会导致严重后果的，(3) 不明显的。短检查表的使用率远高于被忽略的长检查表。
  4. 新手用READ-DO，专家用DO-CONFIRM：格式要匹配用户专业水平。强迫专家使用READ-DO会引发抵触和放弃。让专家流畅操作，事后再确认。
  5. 实证测试启发式方法：不要假设规则有效。用历史案例测试。将启发式决策与最优决策对比。如果准确率<80%，则优化或放弃。
  6. 偏差意识≠消除偏差：知道存在可得性偏差并不能防止它。启发式方法是无意识的。需要外部检查（检查表、同行评审、基准率）来抵消偏差。
  7. 环境变化时更新启发式方法：针对过去环境优化的规则可能在新环境中失效。市场变化、技术变革、竞争对手策略演变。每季度重新验证一次。
  8. 强制函数优于提醒："不要忘记X"会失效。"完成X才能继续"才有效。构建约束（例如，部署脚本要求所有测试通过），而非依赖记忆。
常见陷阱:
  • ❌ 将启发式方法视为通用法则："总是选择知名品牌"在面对欺骗性广告或小众高质量选项时会失效。
  • ❌ 检查表过长：30项的检查表会被跳过。最多保留5-10个关键项。
  • ❌ 忽略基准率："这个客户看起来会购买"（代表性启发式） vs "只有2%的潜在客户会转化"（基准率）。用基准率校准直觉。
  • ❌ 锚定第一个选项："第一个候选人看起来不错，录用吧"而不考虑其他选项。设定满意阈值，然后评估多个选项。
  • ❌ 将检查表作为免责盾牌："我遵循了检查表，不是我的错"忽略了思考的责任。检查表是辅助判断，而非替代判断。
  • ❌ 未测试启发式方法：假设规则有效而不验证。用历史案例测试，衡量准确率。

Quick Reference

快速参考

Common heuristics:
HeuristicRuleExampleBest For
RecognitionChoose what you recognizeDetroit > Milwaukee (size)Stable correlations between recognition and quality
Take-the-bestUse single most important criterionHire based on track record aloneOne dominant factor predicts outcome
SatisficingFirst option meeting thresholdCandidate meets 80% requirements → hireTime pressure, search costs high
AvailabilityJudge frequency by ease of recallPlane crashes seem common (vivid)Recent, vivid events (WARNING: bias)
RepresentativenessJudge by similarity to prototype"Looks like successful startup founder"Stereotypes exist (WARNING: bias)
AnchoringAdjust from initial valueFirst price shapes negotiationNumerical estimates (WARNING: bias)
Checklist formats:
FormatWhen to UseProcessExample
READ-DONovices, unfamiliar, high-stakesRead step → Do step → RepeatSurgery (WHO checklist)
DO-CONFIRMExperts, routine, familiarDo from memory → Confirm with checklistAviation pre-flight
Challenge-ResponseTwo-person verificationOne reads, other confirmsNuclear launch procedures
Checklist design principles:
  1. Keep it short: 5-10 items max (critical steps only)
  2. Use verb-first language: "Verify backups complete" not "Backups"
  3. One step per line: Don't combine "Test and deploy"
  4. Checkbox format: ☐ Clear visual confirmation
  5. Pause points: Identify natural breaks (before start, after critical phase, before finish)
  6. Killer items: Mark items that block proceeding (e.g., ⚠ Tests must pass)
When to use heuristics vs. checklists:
Decision TypeUse HeuristicUse Checklist
Choose between options✓ Recognition, take-the-best, satisficing✗ Not applicable
Sequential procedure✗ Not applicable✓ Pre-flight, deployment, surgery
Complex multi-step✗ Too simplified✓ Ensures nothing skipped
Routine decision✓ Fast rule (satisficing)✗ Overkill
Error-prone procedure✗ Doesn't prevent errors✓ Catches mistakes
Cognitive biases (when heuristics fail):
BiasHeuristicFailure ModeMitigation
AvailabilityRecent/vivid events judged as frequentOverestimate plane crashes (vivid), underestimate heart diseaseUse base rates, statistical data
RepresentativenessJudge by stereotype similarity"Looks like successful founder" ignores base rate of successCheck against actual base rates
AnchoringFirst number shapes estimateInitial salary offer anchors negotiationSet own anchor first, adjust deliberately
ConfirmationSeek supporting evidenceOnly notice confirming dataActively seek disconfirming evidence
Sunk costContinue due to past investment"Already spent $100k, can't stop now"Evaluate based on future value only
Inputs required:
  • Decision/procedure: What needs simplification or systematization?
  • Historical data: Past cases to test heuristic accuracy
  • Critical steps: Which steps, if skipped, cause failures?
  • Error patterns: Where do mistakes happen most often?
  • Time constraints: How quickly must decision be made?
Outputs produced:
  • heuristic-rule.md
    : Defined heuristic with conditions and exceptions
  • checklist.md
    : Structured checklist with critical steps
  • validation-results.md
    : Test results on historical cases
  • refinement-log.md
    : Iterations based on real-world performance
常见启发式方法:
Heuristic规则示例最佳适用场景
识别启发式选择你熟悉的选项底特律 > 密尔沃基（城市规模）知名度与质量稳定相关的环境
取最优启发式仅依据最重要的标准选择招聘时仅看关键技能的过往表现一个因素具有预测性，其他因素价值不大
满意原则选择第一个满足阈值的选项候选人满足80%要求→录用时间压力大、搜索成本高时
可得性启发式依据回忆的难易程度判断频率高估坠机事件（生动），低估心脏病近期、生动事件（注意：存在偏差）
代表性启发式依据与原型的相似性判断"看起来像成功的创始人"存在刻板印象的场景（注意：存在偏差）
锚定启发式依据初始数值调整判断初始薪资报价影响谈判数值估算场景（注意：存在偏差）
检查表格式:
格式适用场景流程示例
READ-DO新手、不熟悉的流程、高风险阅读步骤→执行步骤→重复手术（WHO检查表）
DO-CONFIRM专家、常规、熟悉的流程凭记忆执行→用检查表确认航空飞行前检查
Challenge-Response双人验证一人朗读，另一人确认核发射流程
检查表设计原则:
  1. 保持简短：最多5-10个项（仅关键步骤）
  2. 使用动词开头的表述："Verify backups complete" 而非 "Backups"
  3. 一步一行：不要合并"测试与部署"
  4. 复选框格式：☐ 清晰的视觉确认
  5. 暂停点：识别自然停顿（开始前、关键阶段后、结束前）
  6. 致命项：标记阻止推进的项（例如，⚠ 必须通过所有测试）
启发式方法vs检查表的适用场景:
决策类型使用启发式方法使用检查表
在选项间选择✓ 识别、取最优、满意原则✗ 不适用
顺序流程✗ 不适用✓ 飞行前、部署、手术
复杂多步骤✗ 过于简化✓ 确保无遗漏
常规决策✓ 快速规则（满意原则）✗ 过度投入
易出错流程✗ 无法预防错误✓ 发现错误
认知偏差（启发式方法失效时）:
偏差关联启发式失效模式缓解方法
可得性偏差依据近期/生动事件判断频率高估坠机事件，低估心脏病使用基准率、统计数据
代表性偏差依据刻板印象相似性判断"看起来像成功创始人"忽略成功的基准率对照实际基准率检查
锚定偏差依据初始数值调整估算初始薪资报价锚定谈判先设定自己的锚点，刻意调整
确认偏差寻找支持性证据只注意确认自己观点的数据主动寻找反驳性证据
沉没成本偏差因过去投入而继续"已经花了10万美元，不能停止"仅基于未来价值评估
所需输入:
  • 决策/流程：哪些需要简化或系统化？
  • 历史数据：用于测试启发式准确率的过往案例
  • 关键步骤：哪些步骤遗漏会导致失败？
  • 错误模式：错误最常发生在何处？
  • 时间约束：决策需要多快做出？
产出:
  • heuristic-rule.md
    : 定义带有条件和例外的启发式规则
  • checklist.md
    : 包含关键步骤的结构化检查表
  • validation-results.md
    : 历史案例的测试结果
  • refinement-log.md
    : 基于实际表现的迭代记录