The Explanation Trap

The answer came fast, and it came close to unanimous.

“The new girl was just worse.”

Two more explanations sat beside it. The weather must have turned. There must have been fewer people at the lake.

All three sound like answers. None of them is one.

What Mira actually knows

Across both days she wrote down exactly five things: the selling price, the buy-in price, the hours, the wage she paid, and the total number sold.

That is the whole record. Which produces an uncomfortable list of everything nobody wrote down:

• Did the weather change?
• Was it the same day of the week?
• Were there as many people at the lake?
• Was anyone else selling ice cream along the shore?
• Which of the two sold how many?

Mira would answer two of those on the spot: it was sunny, and it was busy. That is how she remembers the day. But none of it is written down, counted, or held against the day before. Memory is not measurement. It is the story we tell ourselves about a day once we already know how it ended — and it adjusts itself to that ending with remarkable willingness.

That last question is the one that exposes the board. “The new girl was just worse” is not merely unproven. Mira’s data cannot test it at all, because nobody ever tracked who sold which ice cream. The number 180 is a total, not a split.

A sentence that cannot be wrong is not an explanation. It is a reassurance.

Which is exactly why it is so popular. It costs nothing, it sounds like analysis, and it spares you having to say “I don’t know.” The bill arrives later, paid by the next decision built on top of it. Had Mira believed the explanation, her next move would have been obvious: find a different classmate. She would have replaced the one part of the setup she knows nothing about.

The forgotten wage

There is something missing from every one of those lists, and it is bigger than all of them.

Mira paid her classmate €5 an hour. She paid herself nothing.

Redo day one, this time with her own labor at the same rate: €100 in revenue, €50 in cost of goods, €50 for her own ten hours. What is left is zero.

Now day two: €180 in revenue, €90 in cost of goods, €50 for the classmate, €50 for Mira. What is left is minus €10.

The celebrated first day was not €50 of profit. It was a day of work that barely covered itself — and Mira called the wage she never paid out a profit.

That also wrecks the comparison the whole board has been arguing about. Day one and day two are not two profits. Day one is a day when one worker was free; day two is a day when only half the workforce was. The comparison was broken before anyone started explaining it.

This is not a child’s mistake. It is the same mistake inside every calculation where your own hours cost nothing — the side project, the founding year, the department whose people are already on payroll and therefore appear in nobody’s business case.

The planes that came back

American bomber losses over Europe were heavy in 1943. Armor helps, but armor is heavy, so it cannot go everywhere. The military put a very reasonable question to the Statistical Research Group at Columbia University: where should it go?

They examined the aircraft that returned and mapped the hits. The pattern was clear: fuselage and wings were covered in damage, while other areas came back nearly untouched. The obvious recommendation was to armor the places taking the hits.

Abraham Wald, a mathematician in that group, disagreed. Every aircraft in the sample had one thing in common: it had come back. The areas without holes were not the areas nobody hit. They were the areas where a hit meant the plane never returned to be measured. The armor belonged where the data showed nothing.

The error has a name: survivorship bias. You reason from what is left over to the whole, and overlook that what is left over was selected by exactly the criterion you are investigating. Wald’s work on this appeared in the 1943 memoranda of the Statistical Research Group, and the title says it already: “A Method of Estimating Plane Vulnerability Based on Damage of Survivors.”

Mira’s data is the planes that came back. She counted what sold. She did not count who stood at the stand and walked off, who saw the line and kept going, who was at the far end of the lake and never learned she existed. Her 180 is a success report with a blind spot in the middle of it.

Now turn the mirror around. Which report on your desk shows you only the planes that came back? Won deals get reviewed; lost ones rarely do. Customers who stayed get surveyed. The ones who left do not answer anymore.

What changes tomorrow

The board did not give Mira an explanation. It gave her something more useful: a list of things nobody knows.

So the question for tomorrow is not “whose fault was it” but “what do I write down.” And because nobody at an ice cream stand can record twenty things at once, she has to choose.

She may measure one additional number from tomorrow on. Which one — and what do you want to be able to decide with it?

The second half of that question is the harder one. A number you cannot decide anything with is decoration, however interesting it looks. In Episode 4 the answer turns into a calculation everyone should run once in their life.