Series ID: EDKS-BASIC-ZH-0106 · Lesson No.106 · Expert Maintenance: Model Literacy
Explaining Models, Assumptions and Limits | 解释模型、假设与边界
A model is useful because it simplifies reality. A model becomes dangerous when its simplifications are forgotten.
Expert English often needs to explain models: scientific models, business forecasts, diagrams, scoring systems, frameworks, simulations and mental models. The language must show what the model represents, what it leaves out, which assumptions it depends on and where its conclusions should stop.
专家级解释的关键不是把模型讲得“像真的一样”，而是同时讲清楚：模型代表什么、依赖什么假设、忽略了什么，以及在哪些条件下不能再使用。
Chinese Edition Hub · 中文版入口 · ← Lesson No.105 · Incident Reports, Postmortems and Learning Reviews
The model-literacy map | 模型素养地图
- representation
- purpose
- variables
- assumptions
- simplification
- inputs
- outputs
- sensitivity
- validation
- limits
1. Model | 模型
A model is a simplified representation used to understand, explain, predict or decide.
2. Purpose first | 先看模型用途
Ask:
- describe?
- explain?
- predict?
- simulate?
- rank?
- decide?
3. Same model can be good for one purpose and poor for another | 模型适用性取决于任务
A simple model may explain structure well but predict poorly.
4. Representation | 表示关系
What part of reality does the model represent?
5. Variables | 变量
Which quantities or categories can change?
6. Parameters | 参数
Parameters control how the model behaves.
7. Inputs | 输入
What information goes into the model?
8. Outputs | 输出
What does the model produce?
9. Assumption | 假设
An assumption is something treated as true or sufficiently true for the model to operate.
10. Assumptions must be visible | 假设要显性
Do not hide critical assumptions in technical footnotes if they materially affect interpretation.
11. Example assumption | 假设例子
The forecast assumes demand remains near the recent average.
12. Assumption vs fact | 假设 vs 事实
Do not present an assumption as observed reality.
13. Simplification | 简化
Every model leaves something out.
14. Good simplification | 有效简化
Leaves out detail that does not matter for the model’s purpose.
15. Dangerous simplification | 危险简化
Leaves out a factor that changes the conclusion.
16. Boundary conditions | 边界条件
Conditions under which the model is intended to work.
17. Out-of-domain use | 超出适用范围
A model trained or designed for one population may perform poorly elsewhere.
18. Explain the domain | 说明适用域
This model applies to short-term demand under normal operating conditions.
19. Do not say “the model says” too casually | 不要把模型人格化
Better:
Under these assumptions, the model estimates…
20. Model output is conditional | 输出是有条件的
It depends on:
- data
- assumptions
- structure
- parameter choices
21. Calibration | 校准
Adjust parameters so model behaviour aligns with observed data.
22. Validation | 验证
Test whether the model performs adequately on relevant evidence.
23. Validation ≠ proof | 验证不等于永远正确
Successful validation in one setting does not guarantee universal reliability.
24. Training vs test data | 训练数据 vs 测试数据
When relevant, distinguish data used to build the model from data used to test it.
25. Overfitting | 过拟合
A model can fit past data closely while performing poorly on new data.
26. Underfitting | 欠拟合
A model may be too simple to capture important structure.
27. Complexity trade-off | 复杂度取舍
More complex is not automatically better.
28. Interpretability | 可解释性
Can a user understand why the model produces a result?
29. Accuracy vs interpretability | 准确度 vs 可解释性
Sometimes there is a trade-off.
30. Sensitivity analysis | 敏感性分析
Ask how output changes when assumptions or inputs change.
31. Robust conclusion | 稳健结论
A conclusion is more robust if it survives reasonable changes in assumptions.
32. Fragile conclusion | 脆弱结论
If a small assumption change reverses the result, communicate that clearly.
33. Scenario model | 情景模型
Use multiple plausible scenarios rather than one false-precision forecast.
34. Best/base/worst credible | 最佳/基准/最坏可信情景
Explain what changes across scenarios.
35. Forecast model | 预测模型
A forecast is conditional on assumptions about the future.
36. Prediction interval/range | 预测范围
Where possible, show uncertainty around the estimate.
37. Classification model | 分类模型
May assign categories or probabilities.
38. Threshold choice | 阈值选择
Changing threshold can change false positives and false negatives.
39. Cost of errors | 错误成本
Different contexts care differently about:
- false positive
- false negative
40. Model score ≠ reality | 分数不等于现实本身
A score is a representation generated by the model.
41. Measurement model | 测量模型
A test score may estimate an underlying construct rather than directly observe it.
42. Proxy | 代理指标
A proxy is used when the thing we care about is difficult to measure directly.
43. Proxy risk | 代理指标风险
The proxy may drift away from the real goal.
44. Goodhart-style caution | 指标被目标化后的风险
When a measure becomes a strong target, behaviour may adapt to the measure rather than the underlying objective.
45. Diagram as model | 图示也是模型
A diagram simplifies relationships. Arrows can imply sequence or causality, so explain what they mean.
46. Framework as model | 框架也是模型
Frameworks organise thinking but may not make numerical predictions.
47. Mental model | 心智模型
A mental model is a simplified internal representation used to reason.
48. Models can conflict | 模型会冲突
Two models may emphasise different mechanisms or scales.
49. Compare models by task | 按任务比较模型
Ask which model is useful for which question.
50. Model language frames | 模型表达句型
- The model represents…
- It assumes…
- It does not include…
- It is most useful when…
- It becomes unreliable when…
51. Explain to a general audience | 向公众解释
Start with purpose, then one key assumption, then the main limitation.
52. Explain to a specialist | 向专家解释
Add data, parameters, validation and sensitivity.
53. Avoid model mystique | 避免“模型神秘化”
Technical complexity does not remove assumptions.
54. Avoid model dismissal | 也不要因为简化就否定模型
All useful models simplify. The question is whether the simplification fits the job.
55. Mandarin transfer: “模型”有很多类型 | model/function must be explicit
State whether you mean statistical model, conceptual framework, simulation or representation.
56. Mandarin transfer: “假设”可指 assumption/hypothesis | 两者不同
assumption = condition treated as given; hypothesis = claim to be tested.
57. Practice A | 练习 A
Take a simple forecast and list inputs, assumptions, output and limits.
58. Practice B | 练习 B
Explain the same model to a general audience and a specialist.
59. Practice C | 练习 C
Change one assumption and explain how the conclusion changes.
60. Error clinic | 常见问题
| Problem | Repair |
|---|---|
| Model treated as reality. | State representation. |
| Assumptions hidden. | Make them explicit. |
| Output treated as certain. | State range/conditions. |
| Used outside domain. | State boundaries. |
| Complexity mistaken for quality. | Match model to task. |
61. First weak link diagnosis | 第一个卡点诊断
- purpose unclear → task definition.
- assumptions invisible → model explanation.
- conclusion fragile → sensitivity.
- audience confused → adaptation.
- model overtrusted → boundary control.
62. Seven-day training cycle | 七天训练
| Day 1 | purpose/representation | 用途表示 |
| Day 2 | inputs/outputs | 输入输出 |
| Day 3 | assumptions | 假设 |
| Day 4 | validation | 验证 |
| Day 5 | sensitivity | 敏感性 |
| Day 6 | audience versions | 受众版本 |
| Day 7 | model explainer | 综合 |
63. Self-test | 自测
Explain a model’s purpose, variables, assumptions, inputs, outputs, validation, sensitivity and boundaries without treating the model as reality itself.
64. For parents and teachers | 给家长和老师
Use everyday models too: maps, grading rubrics, schedules and diagrams.
Ask learners what the model leaves out.
65. Final real-world challenge | 最终真实任务
- Choose one model.
- State its purpose.
- List assumptions.
- List inputs/outputs.
- Explain validation.
- Run one sensitivity thought experiment.
- State boundary conditions.
- Write a public explanation.
- Write a technical explanation.
- Explain what evidence would make you stop trusting the model.
Next: Lesson No.107 | 下一课
The next lesson develops facilitation language for meetings, panels and expert Q&A: opening the question, keeping speakers on scope, surfacing disagreement, protecting time and closing with clear decisions or unresolved issues.
Lesson No.107 · Facilitating Meetings, Panels and Expert Q&A · 主持会议、讨论与专家问答
Reference floor: expert-maintenance model literacy. Models are explained as conditional representations with explicit assumptions and limits.
