Sample frameDay 144·AI Fluency · Consistency

Good morning. You are consistent about the decisions you make once. It's the ones you make fifty times a year that drift.

AI Fluency.

The Second Opinion

5 min read·Apply by lunch

The question

Take the judgement you make most often and write down the rule you actually apply to it. If the whole thing fits in four lines, what were all those years of applying it from memory, differently, depending on the day, actually buying you?

The idea

Paul Meehl, professor of psychology at the University of Minnesota, published Clinical versus Statistical Prediction in 1954 and then spent decades watching his own profession decline to believe it. He had compared two ways of predicting an outcome: a trained expert weighing a case in their head, and a simple formula applied mechanically to the same information. The formula matched or beat the expert almost every time. When Grove and Meehl returned to the question in 1996 across a hundred and thirty-six studies, the pattern held — sixty-four favoured the formula, sixty-four were a draw, and eight favoured the expert. The reason isn't that experts know less than a checklist. It's that a formula weighs the same factors on a Friday afternoon as on a Monday morning, isn't moved by the vivid case it saw last week, and doesn't get tired or hungry or impatient. You are the best available source of the rule. You're an unreliable executor of it.

The framework

01
The rule you are already using, in four lines.
Take the decision you make most often and list the factors that genuinely moved your last ten calls — not the ones you'd recite in an interview. Give each a rough weight and a threshold. It usually comes out at four or five lines and produces a sentence you feel slightly embarrassed by: advance the candidate when two of these three signals are present. That embarrassment is the finding, not a problem with the exercise. A rule that plain has been running your decisions for years, unwritten, applied a little differently every time depending on who asked and how your week was going.
02
Hand the rule to the model — not the decision.
Give it your written criteria and feed every case in an identical format, asking for three things: the verdict, the factor that decided it, and anything the rule doesn't cover. What you're buying here isn't intelligence; it's sameness, the hundredth application identical to the first. The failure mode is asking the wrong question. Ask what it thinks of the candidate and you get an agreeable essay that mostly returns what you signalled. Ask it to apply your rule and name the clause that settled the case, and you get something you can check, argue with and correct.
03
Override it whenever you must, but never silently.
You will disagree with the rule, and you should be allowed to. Meehl's own exception was the broken leg: a formula predicting that a man will go to the cinema tonight knows nothing about the plaster cast he acquired this morning, and no rule should stand against a fact like that. He also warned that people reach for the exception constantly, on evidence far softer than a broken leg. So make an override cost one line, written on the day in the manner of Day 62: the case, the rule's verdict, your verdict, and the fact you held that the rule couldn't see.
04
Which beat which — your exceptions, or the rule?
Once a quarter, read the override file against what actually happened. Two findings are possible and both are worth the hour. If your exceptions beat the rule, the rule is missing a factor you can evidently see, and it gets written in — you have improved the instrument rather than merely disagreed with it. If the rule beat your exceptions, you've learned the less comfortable thing, and the correct response is to override less often, not to keep a private tally of the times you were right. Almost everybody believes their own exceptions are the good ones, and the file is the only thing that can settle it.

Members only · 4 principles + template + AI mentor

364 more frameworks are waiting.

The Second Opinion is Day 1 of 365. One framework every morning for a year — across strategy, sales, negotiation, leadership, and more.

$1 /day

Billed $365/yr · cancel any time

365 frameworks

12 topics · templates · AI mentor

Reserve your place — $365/yr →

One payment a year · the whole library · no auto-renewal.

More AI Fluency frameworks

All 20 ai fluency frameworks →