I wrote that actions matter more than interpretations. I was half right.

Three years ago I published a short piece arguing that you should judge people by their actions rather than by your interpretation of them. Interpretation is a story you write about someone. Actions are a record.

I was fourteen. I still think the core of it is right, and I have had three years of running companies, hiring, being hired, mentoring thirty-four founders and losing some, to find out where it breaks.

It breaks in two places, and the second one is expensive.

#Where it holds

Every place involving money or delivery, it holds completely.

Someone tells you they will have it Friday. Friday is a fact. Whether they seemed sincere on Tuesday is not evidence about Friday, and the strength of your impression of their sincerity is not correlated with the outcome — I have checked this the hard way often enough to say it flatly.

The version of this I now use operationally: the last three instances beat any narrative. Not the last one, which is noise. Not the average over a year, which lets old good behaviour subsidise recent bad. The last three.

That rule has saved me from every hiring mistake I did not make and would have saved me from two I did.

#The first place it breaks

An action is not a self-interpreting object.

This is the flaw in the fourteen-year-old's version, and it is a real one. I wrote as though actions were data and interpretations were noise — as though you could observe behaviour cleanly and the story was something you added afterwards.

You cannot. "He did not reply for four days" is already an interpretation, because not replying is a category I constructed. The raw fact is that no message arrived. Whether that is a snub, a busy week, a lost email, or a person in a hospital is not visible in the data, and I chose one of those before I noticed I was choosing.

So "judge actions, not interpretations" quietly smuggles in the assumption that you have direct access to actions. You have access to a very thin slice of someone's behaviour, selected by circumstance, and then you interpret it and call the result an observation.

The correction is not to abandon the rule. It is: judge actions, and hold the interpretation of them loosely, and know which one you are doing. That is clumsier and it is what I actually believe.

#The second place it breaks, which is worse

The essay was about how to judge other people. It never occurred to me to point it at myself.

Applied inward it says something much less comfortable: you are also an actions record, and your interpretation of yourself is also a story.

I spent most of last year believing I was sleeping seven hours a night. I was averaging just over six. I did not lie. I had a completely sincere interpretation of myself as someone who sleeps enough, and it was wrong by five hours a week, and it stayed wrong until a wearable told me.

That is the same error I wrote the essay about. I just had it aimed outward, where it is cheap, instead of inward, where it costs.

I now think that is the general shape. It is easy to be rigorous about other people's actions, because their interpretation is obviously a story — you can see them telling it. Your own interpretation does not feel like a story. It feels like knowing yourself.

#What I would write now

If I rewrote it today, it would be shorter and it would have this in it:

On other people. The last three instances, not the narrative. And say out loud which part is the action and which part is your reading of it, because you will not otherwise notice you added the second one.

On yourself. You do not get to be the exception. Whatever standard of evidence you apply to a founder telling you the launch is nearly ready, apply to yourself telling you the training is going fine. Get a number. The number will disagree with you, and the number is right.

On the gap. The distance between what you believe about yourself and what your record says is the only genuinely useful thing in either version of this essay, and fourteen-year-old me did not mention it once, because he had not lived long enough to have a record that could disagree with him.

I have one now. It disagrees with me regularly.

That turns out to be the point.