OpenIDEA

Case 1: Deduction Trap The pricing paper rested on a rule every software executive knows. Software has no marginal cost. Each new customer costs almost nothing to serve, so margins can be generous and growth can be aggressive. The team treated it as geometry, and nobody argued with it. Then the company added an AI assistant to its product. Usage climbed fast, and so did the bill for the model that answers every query. The heaviest users, the ones the paper had treated as its most valuable, were now the most expensive to serve. A flat monthly fee covered a small group running thousands of requests a day. The rule had not been false. It had been true for the old product, and the paper had quietly assumed it still held. Once every answer carried a real cost, the reasoning still followed perfectly from its premise. The premise had expired. The team priced by usage, capped the heaviest tiers, and tested margins on real accounts before writing the next plan.

A rule tells you what must follow. A pattern tells you what probably will. A story tells you what might be true. Confusing them is how a confident room makes a confident mistake.

The Monolith Fallacy: believing logic is one clean tool, when it is a switchboard of reasoning modes. Each runs on its own fuel, is strong over a certain range, and fails in its own way. The first skill is seeing that there is more than one mode at all.

Someone says the logic is clear, and the room nods. But logic is not one clean tool. It is a switchboard of reasoning modes, each with its own fuel and its own way of failing.

Three people in the meeting, three different machines. One reasons from a rule, one from a pattern, one from a story that seemed to fit. All of them call it logic, and the room nods as if it saw the same thing. The essay above shows why they are not the same machine.

A model that forbids nothing explains nothing. Before you accept the next confident claim, ask what currency it is paid in.

Case 3: The Labelled Bet Two product teams at the same lender got the same idea: a credit line for gig workers, built on a new model of income stability. Both models were coherent, and both teams wrote clean memos. Team A labelled it a bet: a structured argument not yet confirmed. It launched to 2% of applicants, set the default rate that would trigger a pause in advance, and scheduled a review at ninety days. Its memo named the signal that would show it had failed. Team B labelled it a certainty and wrote no stop threshold, because a certainty does not need one. The board approved a full rollout within the quarter. At ninety days, defaults ran above the model’s prediction in both books. Team A paused and tightened eligibility for the income groups driving the losses. Team B blamed the data and spent two more quarters defending the model against its own numbers. Both teams were clever. What separated them was the label. A bet can be staged, hedged, and watched. A certainty can only be defended. Call it the labelled bet: act on coherence if you must, but never let it pose as a fact.

You are allowed to act on coherence when evidence cannot yet testify. You are just not allowed to let it pose as something it is not.

The objection is right: most decisions cannot wait for a clean signal. Gravitational waves were sound reasoning for a century before any instrument could register one. So act on coherence. But label the bet as a bet, so you can hedge it, stage it, and watch it.

A number tells us that something happened. It never tells us why. A room that mistakes the metric for the mechanism will pull a lever wired to nothing.

Case 2: coherence spent as proof The expansion paper was the strategy team’s finest work. Footfall drives conversion. Conversion drives basket size. Basket size drives margin, and margin repays each lease inside three years. The board read it twice, found no gap, and approved forty stores in one afternoon. Eleven months later, the first twelve stores opened. Footfall rose, as promised. Conversion did not. People walked in, looked around, and walked out. Nothing in the logic had forbidden that outcome. The paper described a store that existed only in the document. No one had been careless. The argument was airtight, and that was the problem. Every step followed from the last, so the feeling of rigour settled over the room. No one asked whether the chain had ever met a real shopper before forty leases were signed. Call it coherence spent as proof: an argument that hangs together has earned the right to be considered, not the right to be believed.

The most dangerous plan is not the sloppy one. It is the elegant one no one ever checked against the world. The luminiferous aether was consistent, defended by careful people, and completely wrong. Coherence bought it decades. Contact eventually killed it.

Coherence is not evidence. Measurement is not meaning. Two currencies, and neither can buy what the other sells.

The confusion runs both ways. Spending coherence as evidence: it must be true, look how consistent it is. Spending measurement as understanding: we tracked it, so we must know what drives it. Both feel exactly like rigour. Both are category errors.

A perfectly coherent map can describe a country that does not exist. Logic can clear vast territories of nonsense in an afternoon. It cannot confirm the world agrees.

Case 1: The perfect hiring rubric The scorecard was the most rigorous document the people team had ever produced. Eleven competencies, each with behavioural anchors. Every interviewer was trained on it, and the calibration panels agreed almost every time. Leadership read the calibration results as proof. Interviewer agreement stood at 94%. The process was coherent, and no one could find a gap in it. The board approved it as the standard for every executive hire. Six months later, a third of the senior hires were struggling, and two had already left. The rubric had rewarded polished answers to structured questions, not the qualities that work in the role. Nobody had checked. The panels had only ever tested whether they agreed with one another. The fix was not more training. The team pulled the scores of every executive hired in three years and compared them with performance reviews. The correlation was close to nothing. Agreement inside the room had been mistaken for contact with the job. The rubric needed testing against early performance, not against itself. Coherence is not evidence.

Logic and evidence are different currencies. Logic buys consistency. Evidence buys contact with reality. Neither can buy what the other sells. An argument that hangs together has earned the right to be considered, not the right to be believed.

Two smart people argue. Neither moves an inch. One says the logic is airtight, the other says show me the data. Both are right, and the meeting dies anyway. They are not paying in the same currency. The essay above shows which currency each is spending.