I check the math before I believe.

Five sentences that keep turning up in my writing. Each comes with three pieces of my own evidence, one sentence where I could be wrong, and my interest in it.

Runs through 4 of 8 topic fields: AI costs, Judgment, Society, Perception

A good story is not yet the truth.

“An explanation that fits victory and defeat equally well explains neither.”Germany Loses to Paraguay. Our Brain Wins a Story.

1AI costsAI Gets 13 Times Cheaper Every Year? For Whom, Exactly?“13 times cheaper” only holds for the bargain hunter at the margin. We measure against yesterday's requirements and sell tomorrow's savings.
2JudgmentExplaining Is Easy, Predicting Is Hard: I Back-Tested the Famous Economic LinksFamous leading indicators look significant in hindsight and are worthless out of sample. After the event there's always a reason, before it almost never.
3JudgmentGermany Loses to Paraguay. Our Brain Wins a Story.Coherence is no proof of truth. An explanation that explains victory and defeat equally well explains nothing. The goal is calibration.
4JudgmentJobs to Be Done: The Job Explains Nothing. The Switch Does.A job stated after the fact is a story. JTBD only becomes useful when the statement rules something out and you ask why people canceled.
5JudgmentStatistical Deception: Predicting Lottery Numbers with AIAn elaborate AI model predicts lottery numbers exactly as well as chance. What's being sold is only the feeling of control.
6Society“The World Has Changed” Is Not a Diagnosis. It's a Negotiating Move.A statement can be true and still not be an argument. “The world has changed” turns decisions into weather and only points one way.
7JudgmentTechnology Doesn't Repeat. Our Narratives Do.Success generates explanations faster than explanations generate success. We mistake a description of a winner for a theory of winning.
8JudgmentWhy We Tell Ourselves Stories: What Managers, AI and the Human Mind Have in CommonCEOs, language models and brains all produce coherent explanations that aren't the mechanism behind their decisions. Convincing doesn't mean proven.
9PerceptionThe Sun Is a Blank Page, and Plato Was Still RightPlato was right about ontology and wrong about morality: knowledge never leaves the cave, it only casts more stable shadows.

A good story is not yet the truth.

Why it matters now

AI makes good stories almost free. Every slide, every report, every headline sounds convincing now. So the explanation gets cheap and the evidence gets expensive.

Checked myself

512ways to compute Epoch's data. The 13x cheaper holds, but only for the bargain hunter.Read →0.7359average hits for an AI model trained on 70 years of lottery draws. Chance gets 0.7352.Read →0famous leading indicators survived my out-of-sample test.Read →

What it means for youAsk what every number measures. And whether every explanation would have explained the opposite just as well.

Where I could be wrongIf well-told forecasts keep coming true without anyone checking the math. Then storytelling is knowledge, and I calculate too much.

My interest in itI get paid to check the math. The more people distrust stories, the more I earn.

Dependency doesn't disappear. It moves.

Why it matters now

AI makes building cheap. That doesn't end dependency, it moves it: from the software vendor to the model provider, from the license to the token.

Checked myself

10 hfor eight office apps, rebuilt with AI. Typing was cheap, deciding wasn't.Read →600packages came with 16 building blocks in one self-built app. Each one is a dependency.Read →$218with one provider, $7,926 with another. Same research, comparable models.Read →

What it means for youDon't ask "Should we build it ourselves?" Ask "How fast can we get out again, and what does that cost?"

Where I could be wrongIf providers stay cheaper and more stable than any switching option costs. Then lock-in is the sensible choice.

My interest in itBalane builds software and helps companies switch. The more they think about dependency, the better for me.

Advice is worth exactly as much as its evidence.

Why it matters now

AI writes expert language in seconds. So what advice is worth no longer depends on tone, title or slides, but on whether someone shows what it stands on and answers for it.

Checked myself

178businesses relied on my news search when a retired model broke it. A consultant called switching models "trivial".Read →1 firm, 2 answersA partner says one thing, the same firm's white paper another. The client buys a mood of the day.Read →✓The certificate was valid. Compliance proves a procedure on a given date, not security.Read →

What it means for youPay for evidence and accountability, not certainty. Ask: how will we know your advice was wrong, and what happens then?

Where I could be wrongIf advisers without numbers of their own demonstrably lead to better decisions than those with them. Then experience is evidence you don't have to show.

My interest in itI advise companies myself. So this sentence applies to me first.

Being wrong has become cheap. Whoever measures, wins.

Why it matters now

An experiment now costs a weekend instead of a quarter. With twenty-four tries a year, you don't have to be right. You have to be able to count.

Checked myself

€14for a weekend experiment. 24 tries a year, twice paying to find the wrong customers.Read →1,100green tests, and the update was still broken. Test count measures effort, not safety.Read →12 / 40books read. The others are options on questions I don't have yet.Read →

What it means for youBudget for measuring, not for forecasts. And decide before the experiment what counts as success.

Where I could be wrongWhere a single mistake costs everything, in safety, health or liability, being wrong is never cheap. There the sentence doesn't hold.

My interest in itI build tools for measuring and calculating. People who measure need them.

Systems tip over when nobody pushes back anymore.

Why it matters now

The more decisions run through approvals, committees and optimization, the earlier dissent gets filtered out. Not out of malice. Because each time it looks reasonable.

Checked myself

(1 − p)ⁿTwelve reasonable gatekeepers with a veto, and every candidate with character drops out.Read →Enrondidn't die of fraud but of having trained itself out of saying no.Read →2020–2026BMW in numbers: success that gets managed instead of tested.Read →

What it means for youMeasure your organization by how expensive it is to disagree. Who pushed back last, and what happened to them?

Where I could be wrongIf consensus-heavy organizations get through crises better than argumentative ones, decade after decade. Then calm is a strength.

My interest in itI get paid to disagree from the outside. That makes the sentence convenient for me.

Books as tools

Not book tips. Each book here answers a question of today. Below it is the sentence that came out of it.

PlatoAllegory of the Cave“We look for the truth in the light, and we're surprised it blinds us. The image was never in the sun.”George OrwellNineteen Eighty-Four"Whoever controls the past" was a metaphor in 1949. Today it's a question of write access to a database.Aldous HuxleyBrave New World“And the most effective control is the one you’re grateful for.”Hermann HesseThe Glass Bead Game“Beauty that serves nothing is a beautiful cage.”Oscar WildeThe Picture of Dorian Gray“Your feed is the painting. You're the attic.”Friedrich NietzscheOn the Genealogy of Morality“The morality of the powerless as revenge by way of values.”Ludwig WittgensteinTractatus“You are not a fact in your own life. You are its edge.”George SantayanaThe Sense of Beauty“When pleasure gets objectified into "quality," beautiful design is an ethical question, not a cosmetic one.”Karl PopperThe Paradox of Tolerance“Quote only one of them and you are not quoting Popper. You are quoting yourself with Popper's name underneath.”Francis FukuyamaThe End of History“A system that stops serving interests does not go quiet. It changes currency.”Thomas BernhardWoodcutters“Contempt that leaves the person feeling it out of the picture is not criticism. It is a pose.”Nassim TalebAntilibrary“The answer got cheap. The question did not.”Morgan HouselThe Psychology of Money“And taking it for granted is the beginning of the end.”Bethany McLean, Peter ElkindThe Smartest Guys in the Room“The question is not whether your company has good values. The question is what happens to the person who uses them.”Krause and TrappeHybris“Purity” is, genetically, an illusion.Christian MikundaDer verbotene Ort“We optimised every measurable variable. Then we wondered why nothing was memorable.”

I check the math before I believe.

Five sentences that keep turning up in my writing. Each comes with three pieces of my own evidence, one sentence where I could be wrong, and my interest in it.

Runs through 4 of 8 topic fields: AI costs, Judgment, Society, Perception

A good story is not yet the truth.

“An explanation that fits victory and defeat equally well explains neither.”Germany Loses to Paraguay. Our Brain Wins a Story.

1AI costsAI Gets 13 Times Cheaper Every Year? For Whom, Exactly?“13 times cheaper” only holds for the bargain hunter at the margin. We measure against yesterday's requirements and sell tomorrow's savings.
2JudgmentExplaining Is Easy, Predicting Is Hard: I Back-Tested the Famous Economic LinksFamous leading indicators look significant in hindsight and are worthless out of sample. After the event there's always a reason, before it almost never.
3JudgmentGermany Loses to Paraguay. Our Brain Wins a Story.Coherence is no proof of truth. An explanation that explains victory and defeat equally well explains nothing. The goal is calibration.
4JudgmentJobs to Be Done: The Job Explains Nothing. The Switch Does.A job stated after the fact is a story. JTBD only becomes useful when the statement rules something out and you ask why people canceled.
5JudgmentStatistical Deception: Predicting Lottery Numbers with AIAn elaborate AI model predicts lottery numbers exactly as well as chance. What's being sold is only the feeling of control.
6Society“The World Has Changed” Is Not a Diagnosis. It's a Negotiating Move.A statement can be true and still not be an argument. “The world has changed” turns decisions into weather and only points one way.
7JudgmentTechnology Doesn't Repeat. Our Narratives Do.Success generates explanations faster than explanations generate success. We mistake a description of a winner for a theory of winning.
8JudgmentWhy We Tell Ourselves Stories: What Managers, AI and the Human Mind Have in CommonCEOs, language models and brains all produce coherent explanations that aren't the mechanism behind their decisions. Convincing doesn't mean proven.
9PerceptionThe Sun Is a Blank Page, and Plato Was Still RightPlato was right about ontology and wrong about morality: knowledge never leaves the cave, it only casts more stable shadows.

A good story is not yet the truth.

Why it matters now

AI makes good stories almost free. Every slide, every report, every headline sounds convincing now. So the explanation gets cheap and the evidence gets expensive.

Checked myself

What it means for youAsk what every number measures. And whether every explanation would have explained the opposite just as well.

Where I could be wrongIf well-told forecasts keep coming true without anyone checking the math. Then storytelling is knowledge, and I calculate too much.

My interest in itI get paid to check the math. The more people distrust stories, the more I earn.

Dependency doesn't disappear. It moves.

Why it matters now

AI makes building cheap. That doesn't end dependency, it moves it: from the software vendor to the model provider, from the license to the token.

Checked myself

What it means for youDon't ask "Should we build it ourselves?" Ask "How fast can we get out again, and what does that cost?"

Where I could be wrongIf providers stay cheaper and more stable than any switching option costs. Then lock-in is the sensible choice.

My interest in itBalane builds software and helps companies switch. The more they think about dependency, the better for me.

Advice is worth exactly as much as its evidence.

Why it matters now

AI writes expert language in seconds. So what advice is worth no longer depends on tone, title or slides, but on whether someone shows what it stands on and answers for it.

Checked myself

What it means for youPay for evidence and accountability, not certainty. Ask: how will we know your advice was wrong, and what happens then?

Where I could be wrongIf advisers without numbers of their own demonstrably lead to better decisions than those with them. Then experience is evidence you don't have to show.

My interest in itI advise companies myself. So this sentence applies to me first.

Being wrong has become cheap. Whoever measures, wins.

Why it matters now

An experiment now costs a weekend instead of a quarter. With twenty-four tries a year, you don't have to be right. You have to be able to count.

Checked myself

What it means for youBudget for measuring, not for forecasts. And decide before the experiment what counts as success.

Where I could be wrongWhere a single mistake costs everything, in safety, health or liability, being wrong is never cheap. There the sentence doesn't hold.

My interest in itI build tools for measuring and calculating. People who measure need them.

Systems tip over when nobody pushes back anymore.

Why it matters now

The more decisions run through approvals, committees and optimization, the earlier dissent gets filtered out. Not out of malice. Because each time it looks reasonable.

Checked myself

What it means for youMeasure your organization by how expensive it is to disagree. Who pushed back last, and what happened to them?

Where I could be wrongIf consensus-heavy organizations get through crises better than argumentative ones, decade after decade. Then calm is a strength.

My interest in itI get paid to disagree from the outside. That makes the sentence convenient for me.

Books as tools

Not book tips. Each book here answers a question of today. Below it is the sentence that came out of it.