← How an Economy Grows and Why It Crashes
第 7 章 · 20 分钟
Reverse audit: checking the reasoning of chapters 1–6
The audit covers chapters 1 to 6, including right conclusions reached wrongly.
概念地图
Reason audit
A right conclusion reached by wrong reasoning is booked as a defect, separately.
深色为本章概念，浅色为其他章
陪练模式
不想只是读？让它带你走一遍。
导师把这一章拆成小步，每一步先问你，再讲；你用自己的话答，它先认下对的部分再纠偏。用你自己的 AI，对话只存在这台设备上。
Locus01
延伸阅读
Popper · mid twentieth century · falsifiability as a criterion
Held that a theory's standing lies in what it forbids, not in what it can explain. He argued against judging theories by explanatory reach; the cost is that applying the criterion strictly would exclude a great deal of useful social science, so this course uses it for bookkeeping and not for selection.
Foundational
White · 2000 · data snooping and the reality check
Showed that repeated selection on one sample systematically inflates in-sample performance, and by an amount that can be estimated. This turned 'right for the wrong reason' from a warning into a computable quantity.
Turn
Chapters 1 to 6 of this course
The subject of the audit. The third block (derivation) and fifth block (cracks) of each chapter are the main entry points.
Under audit
机制02
Right, unsupported
A conclusion can be correct while the reasoning behind it is not. It happened at least three times in the first six chapters: falling prices need not be harmful (right, but the book's reason — smaller numbers on the same output — is wrong; the real reason is the debt structure); saving supports growth (right, but the book's reason is an identity, and the empirical content sits elsewhere); malinvestment needs clearing (right, but the book defaults the collateral-damage rate to zero). Booking that case separately is what this course calls right, unsupported. Without that entry, frameworks with bad reasoning accumulate credit whenever they happen to guess correctly.
机制An accidentally correct conclusion, through concealing the faulty reason, earns a framework credit it has not earned.
可迁移性测试
Move it to diagnosis: a clinician follows a wrong differential and happens to prescribe the right drug; the patient recovers. Record only outcomes and the same path will prescribe the wrong drug next time. The isomorphism breaks in that medicine keeps a chart, while economic judgements usually leave no record of the path.
机制03
Cracks crossing over
The chapter 1 framework fails under idle capacity, the chapter 3 framework fails under heavy debt, the chapter 6 framework fails with long collateral chains. Each chapter declared its own failure conditions. The danger appears when chapters are quoted across each other: use chapter 1's conclusion on chapter 6's setting and the failure conditions do not travel with it. The ledger built from every chapter's fifth block exists to stop this crack crossover.
机制Cross-chapter quotation, through dropping the original failure conditions, applies a framework where it has declared itself invalid.
可迁移性测试
Move it to a drug label: a medicine is contraindicated in hepatic impairment, and each retelling of its efficacy drops the contraindication a little further, until nothing is left. The isomorphism breaks in that drug labels must legally accompany the product, while opinions carry nothing.
Derivation04
Counting the categories in chapters 1–6
This aggregates the tag of every derivation step in the first six chapters and derives one conclusion: the book's empirical risk sits on very few steps.
Chapters 1 through 6 have 3–4 derivation steps each, 21 in total
Counting rule: only the steps inside each chapter's third block, not cracks or debate.
Ten steps are tagged [identity]
These hold under any data and carry no empirical risk.
Five are tagged [assumption], six are tagged [empirical]
Assumptions are chosen; empirical claims can be overturned.
推导 · 0 / 3
裂缝05
Cracks in this framework
争议地形06
本书主张
This course's position: a right conclusion reached by wrong reasoning is booked as a defect, or bad reasoning keeps earning credit whenever it happens to guess correctly.
另一种看法
The instrumentalist line holds that a theory is worth what its predictions are worth, and that the realism of its assumptions does not matter — Friedman's 1953 essay on the methodology of positive economics is the version most often cited. By that standard, a wrong reason with good predictions is not a defect.
分歧扎在
The disagreement is rooted in what a theory is for. For prediction within the same distribution, instrumentalism holds; for carrying a structure into a new setting, correct reasoning is the entire basis of transferability. This course aims at the second and therefore takes the first position — a choice made by the goal, not a refutation of instrumentalism.
什么证据能裁决
What would settle it is out-of-distribution performance: split prediction rules into 'realistic assumptions' and 'unrealistic assumptions with comparable in-sample fit' and compare them in visibly different environments. No difference favours instrumentalism; the realistic group holding up better gives correct reasoning independent value.
The evidence points to 'it depends on the distance of the extrapolation': at short range the two groups are comparable, at long range the realistic group is steadier. What is missing is a measure of that distance — current practice judges after the fact whether an extrapolation was near or far.
Falsification07
Crack crossover as a preregistrable test
The mechanism under test: cross-chapter quotation, through dropping the original failure conditions, applies a framework where it has declared itself invalid.
Data source: published economic commentary, with the failure-condition ledger from chapters 1–6
The ledger is fixed in each chapter's fifth block and is not edited before the test.
Sample: all commentary in one period that quotes any conclusion of the book, with no filtering by the writer's position
Filtering by position turns the result into a description of the filter.
Window: one full credit cycle, from the start of expansion to the end of clearing
Crossover clusters at the turning points, so the window must contain a turn.
Threshold: the share of quotations that also state the failure conditions is below a pre-registered fifty per cent
Fifty per cent is set in advance and not adjusted afterwards.
Failure condition: the share exceeds fifty per cent, or inter-coder agreement falls below the acceptable level
The second voids the test rather than supporting or refuting.
推导 · 0 / 3
接口08
Which slot it hangs on
挂在哪个槽位The conclusion you already hold is probably 'a model that predicts well is a good model' — a judgement about how to rate a tool. This chapter goes at its range of application.
Does this chapter (a) replace 'predicting well is good enough', or (b) constrain it to cases where you do not intend to extrapolate?
慢变量Register two slow variables: whether you state failure conditions when you quote a conclusion, and how many of your working conclusions are second-hand with the original argument unread. Look at the second first.
小结09
本章小结
01Of 21 derivation steps in chapters 1–6, ten are identities, five are assumptions, and only six carry empirical risk.
02'Right, unsupported' is booked as a defect, or bad reasoning accrues credit by guessing correctly.
03Failure conditions are lost most easily in cross-chapter quotation, which is why the ledger is explicit.
04Taking the position that reasoning must be correct is a choice made by the goal, not a refutation of instrumentalism.
05The audit is partly circular — reasoning is usually judged through conclusions — so it is bookkeeping, not a verdict.
提取练习 · 合上书，先自己答一遍。
?Is 'steps at risk equal total minus identities minus assumptions' (a) an identity or (b) an empirical claim?
?Using chapter 1's conclusion in a deep recession is (a) right-but-unsupported or (b) crack crossover?
Forced choice10
B · Identify the tag
'Steps carrying empirical risk equal total steps minus identities minus assumptions' — what kind of step is that?
二选一
Forced choice11
D · Judge the interface
Which one hangs on a slot in your existing structure?
二选一
读到这里，把它记为已读。