01 THINKHow does he reason?

Most of the case studies were me

Charlie Munger recommended it, so I finally read it. Checking the principles one by one, I found I was not the one reading the case studies — I was inside them.

This sentence is already published in this repository.

The latest in the registry was TypeScript 7.0.2. Native port, far faster type checking. In it went.

There is no reason behind "in it went." Measured in that same piece, type checking in the build took 872ms. It was not the bottleneck. It did not go in because it was fast; it went in because it was newest, and the whole lint pipeline died.

At the time I wrote only that I had been hasty. I did not know the reaction had a name.

I finally read it

I read Influence. It is listed as recommended reading in Charlie Munger's Poor Charlie's Almanack, so I picked it up.

It is a book that organises, from behavioural experiments, when people skip judgement. I found it remarkable the whole way through — and the remarkable part was not the experiments. It was that I am one of their samples.

The six principles

PrincipleWhen it fires
ReciprocityReceiving something makes you feel you owe
ConsistencyYou avoid contradicting what you once said
Social proofIf everyone does it, it must be right
LikingYou inspect the words of people you like less closely
AuthorityAn expert marker makes you skip the judgement
ScarcitySomething running out looks more valuable

What kept catching me while reading was not the principles but the cases. I thought I was on the reading side of the case studies. I was inside them. Most of the time.

The TypeScript item above is authority. I did not judge; I reacted to a marker reading "latest." And that is an item I had already written up and published. There is no way to count how many went unwritten.

Knowing does not switch it off

The most uncomfortable passage in the book is this. Knowing the principle does not stop the reaction. What changes after reading is only that you can name it afterwards.

But this repository had already reached the same conclusion in a different domain.

A check that never fails is not a check says that every guard gets deliberately broken. The bugs the gates let through says this — without a gate you check by hand; with one you believe you checked.

Both rest on the same premise. Do not trust your own judgement in the moment. So machines were put in place instead.

Judgements about codesix checkersJudgements a person makesnoneThe same person makes both, and only one side has a device.And the side without one is the harder to undo.

A checker measures colour contrast. Another measures cascade order, another content rules, another per-route transfer size. Six of them stand there, and the build breaks when one disagrees.

For judgements a person makes, there are none.

I widened one existing rule

No new rule. One existing rule got wider.

From I cut the list short, judged from it, and pushed to the wrong remote — anything hard to undo gets confirmed from the whole list. No head, no grep.

That rule was hung on git remotes only. It now covers every decision that is hard to undo. If a decision arrives with a feeling of "this has to be decided now" attached, that is scarcity, and that is when the whole list has to be read.

Whether that is enough, I do not know. A guard has to be deliberately broken to be trusted, and I have not yet found how to deliberately break a guard aimed at a person. Until then this piece is a memo, not a guard.


The names and definitions of the principles come from the book. The experiments themselves were not re-verified. Every case matched against them above is already published in this repository; none was invented. I understand the count of principles differs between editions, and having been unable to confirm that, I write six.

readingpsychologydecision-makingbias

Record

First committed 2026.09.03, and untouched since.

  • 83ef676Feat: 쉬운 말 판을 라우트로 세운다 — 구현과 20편을 함께

The full build record →