Calibrating Yourself

An unfinished system

Three quantities to measure about yourself

Quantity one — your own baseline

You have recorded eight parameters for other people. Record them for yourself. This needs another person observing, or a recording, because by definition you cannot see yourself.

This quantity matters for a very practical reason: you are the largest source of noise in every measurement you make. The contagion trap in chapter 28 is not an abstract possibility — it runs in every meeting, and its strength depends on everything you have just listed above.

Quantity two — your real hit rate

This is the most uncomfortable quantity, because almost nobody measures it, and almost everybody believes theirs is higher than it is.

Quantity three — systematic bias

Random error cannot be corrected, but a directional bias can be — and almost everyone has one.

The prediction notebook

This is the only tool in the chapter, and it is the cheapest tool in the whole set. A small notebook, four columns, and one rule: write it down before you know the outcome.

DATE    SITUATION              MY PREDICTION               CONFIDENCE  OUTCOME
──────  ─────────────────────  ──────────────────────────  ──────────  ─────────────
12/03   Candidate A interview  “The CV gap was a layoff”   7/10        ✗ caring for
                                                                       a relative

14/03   Negotiation with B     “They will concede on       5/10        ✓
                                price before the weekend”

15/03   Team meeting           “C will vote against”       9/10        ✓

18/03   Client meeting with D  “They had decided not to    8/10        ✗ bought, 2
                                buy before we met”                     weeks late

The four conditions for being allowed to score

This is the hardest statement in the whole set, and it is here because it is the natural conclusion of everything before it.

If any condition is missing, you may still use the whole of this set to steer the conversation — ask better, de-escalate at the right moment, build the relationship better. You simply may not end with a judgement about the other person. That is a much narrower limit than it looks, and most of the value of the set lies in the part that is not limited.

The calibration schedule

1. First month — record only, judge nothing

   Write ten to fifteen predictions. Do no arithmetic at all. The aim is to build the habit of writing first.

2. End of the first month — fill the four boxes

   How many hits, how many false alarms, how many misses did you catch? Subtract the base rate.

3. Second month — draw the calibration curve

   Group the predictions by confidence. In the “eight or nine out of ten certain” group, what is the real hit rate? This is usually the most uncomfortable moment.

4. Third month — find the systematic bias

   Which way do your errors lean? Then apply the matching correction above.

5. After that — maintain

   Three predictions a week is enough. A calibration system kept up lightly is better than a perfect one abandoned after three weeks.

Chapter summary

• The system measures the subject on eight baseline parameters, plus nine zones, four domains, one verbal layer — and measures the observer on nothing at all. The observer is the only measuring instrument in the whole procedure, and the only one that has never been calibrated.
• Record those eight parameters for yourself. This needs another person observing, or a recording, because by definition you cannot see yourself. Almost everyone has a default pacifying behaviour, and almost nobody knows they do. You are the largest source of noise in every measurement you make.
• All four boxes have to be filled: hit · false alarm · miss · right but dull. Natural memory keeps only the hit box; the other three leave no trace — which is why everyone thinks they read people well. The only way to get all four is to record them before you know the outcome.
• The base rate. If nine people in ten around you tell the truth, simply always guessing “this person is telling the truth” makes you right 90% of the time without looking at anything. The number that means something is your hit rate MINUS the base rate; if the difference is zero, your skill has contributed nothing.
• Random error cannot be corrected, but a directional bias can. Leaning towards suspicion: raise the minimum number of signals in the cluster factor from three to four, and write two alternative hypotheses instead of one. Leaning towards trust: use verifiable questions mechanically, even when you see no need.
• The prediction notebook, four columns, one rule: write it down before you know the outcome. A prediction recorded afterwards is worthless, and memory will quietly correct it into a right answer.
• The confidence column, out of ten, is the most important column — it measures overconfidence, not just error. And the prediction has to be verifiable: “this person is not honest” cannot be checked, “this person will not call back within two weeks” can.
• When you say you are nine out of ten certain, the real hit rate is usually much lower — and the gap is widest at the highest confidence level.
• The four conditions for being allowed to score: their baseline established in this context, today · your own posture and state checked · at least one alternative hypothesis written down before observing · knowing how you will verify independently, specifically, not “I will keep an eye out”.
• With any condition missing, you may still use the whole set to steer the conversation — ask better, de-escalate at the right moment, build the relationship better. You simply may not end with a judgement about the other person.
• What the best readers have that others do not: not more signals, not better intuition, but an honest record of where they have been wrong.