Skip to main content
The Investor’s MindLesson 6 of 6 · Knowing yourself

Judgment & behaviour

Stories, and the evidence you go looking for

A coherent story raises confidence more than evidence raises accuracy, and once you hold a view you search asymmetrically. Both are cheap to correct if you know the shape of them.

6 min readIntermediateLast checked against the product on

Coherence is not evidence

The feeling that you understand a company comes from the story hanging together, not from the amount of evidence under it. A tidy narrative built on three facts feels more convincing than a messy one built on thirty.

That is backwards, and it is worth saying plainly because the direction of the error is predictable: the less you know, the easier it is to construct a coherent story, because there is less to contradict. Confidence peaks early.

The asymmetric search

Once a view exists, evidence-gathering stops being neutral. You do not usually ignore contrary facts — the effect is quieter than that. You search differently:

  • Contrary evidence gets scrutinised. Supporting evidence gets accepted.
  • A bad quarter becomes "one-off". A good one becomes "the thesis working".
  • You read the bull case in full and skim the bear case for its weakest point.

Each step is individually defensible, which is what makes the accumulation hard to notice from inside.

Two corrections that actually work

Write the thesis down, with numbers, before you buy. A written claim is the only version of your reasoning that cannot be quietly revised after the fact. Memory rewrites; a file does not.

Run a premortem. Kahneman's version: imagine it is two years from now and the position has been a disaster. Now explain why. Asking for the failure as a fact to be explained rather than a risk to be assessed reliably produces concrete reasons where "what are the risks?" produces platitudes.

The second one is worth doing even when you are confident. Especially then.

How this connects to Pythia

Several surfaces are built to be inconvenient in this specific way.

The guru frameworks disagree with each other on purpose. A company can pass Greenblatt and fail Graham; the bench is not a chorus. Where they disagree, the disagreement is the finding, and a lens that refuses names the reason.

The AI narratives are fenced two different ways, and neither lets the story outrun its evidence. Pythia Insight is grounded strictly in the score payloads it is given — it has no free-text sources to reach for. The news syntheses, which do read outside text, must cite from a fixed set of urls, and an uncited payload is discarded rather than rendered.

And the thesis builder asks for the premortem before the position moves. That field exists because asked afterwards, the question is rhetorical.

The limits of this idea

Seeking disconfirming evidence is not the same as being persuaded by all of it. Every thesis worth holding has a bear case, and a permanent openness to the last argument you heard is its own failure mode — you end up trading on whichever article you read most recently.

The discipline is narrower than "stay open-minded": decide in advance what evidence would change your mind, and how much. Then when it arrives, you are updating on a rule you set while calm, rather than negotiating with yourself while a position is down.

Check yourself

4 questions. Nothing is recorded unless you are signed in, and nothing here affects anything else.

1. An investor knows three facts about a young company and finds the story completely convincing. A colleague knows thirty facts about an older one and feels less sure. What explains the difference in confidence?
2. An investor files a bad quarter as a one-off and a good quarter as the thesis working, reads the bull case in full, and skims the bear case for its weakest point. What is happening?
3. Why does a premortem produce concrete failure reasons where asking what the risks are produces platitudes?
4. Determined to beat confirmation bias, an investor revises their thesis after every article they read, in whichever direction it argues. What does the lesson call this?

4 questions left. An unanswered question counts as a miss, so the check waits for all of them.

Sources