Measure Twice · Sitting 5
Calibration as a Practice
Calibration is not a one-time lesson. It is a practice. Every quantitative claim gets a stakes check before it gets your trust.
- 40 min
- Practice · Reflect
- 15–18
Parent briefing · 5 minutes, before they sit
This sitting makes calibration a protocol, not a concept. The student has learned the three levels. Now they apply the protocol to a real claim from the model. The protocol has four steps: name the stakes, choose the trust level, run the check, report the result. The protocol is the same every time. The output is a calibrated decision: trust, check, or reject. The student who can run this protocol on any claim has a skill that will serve them for life, on any tool, in any domain.
Hard edges
- Academic integrity: the protocol includes 'report the result.' A model number without a check is not a result. A checked number with the source named is.
- The protocol is not about the model. It is about the student's relationship to any quantitative claim, from any source.
If they say
- “I don't need a protocol for every number.”
- You do not need to run it on every number. You need to run it on the ones that change a decision. The protocol takes thirty seconds. The question is: is thirty seconds worth the cost of being wrong? If yes, run it. If no, skip it. The skill is knowing when.
- “This is too rigid.”
- It is a habit, not a prison. The four steps take thirty seconds. The alternative — trusting the model's number without checking — takes zero seconds and costs more when it is wrong. The protocol is the cheaper option when the stakes are real.
Objective
The student can apply a calibration protocol to any quantitative claim from a model: name the stakes, choose the trust level, run the check, and report the result.
The protocol
The calibration protocol has four steps. One: name the stakes. What happens if this number is wrong? Nothing (curiosity), something small (practical), something that hurts (safety). Two: choose the trust level. Curiosity: trust the model. Practical: check if you can. Safety: go to the source with stakes. Three: run the check. Measure, verify, or reject. Four: report. Write the number, the source, and the check. 'The model said X. I measured Y. The stakes were [level]. I [trust/check/reject] based on the check.' That is a calibrated result. The protocol takes thirty seconds. It works on any quantitative claim from any source. It is the skill.
Why the protocol works
The protocol works because it is the same every time. You do not think about whether to trust the model. You run the protocol. The protocol names the stakes, which most people skip. The stakes decide the trust level, which most people do not connect. The check is run, which most people skip. The result is reported, which most people do not do honestly. The student who runs the protocol on every quantitative claim will not be fooled by confidence. They will not miss a safety issue. They will not skip a check because the model sounded right. The protocol is the habit. The habit is the skill.
Big idea
Calibration is a protocol, not a concept. Four steps, every time: stakes, trust, check, report.
Try this~20 min total
Run the protocol
20 min- Pick a real quantitative claim from a model: a number you would use for a decision.
- Step 1: Name the stakes. What happens if this is wrong?
- Step 2: Choose the trust level. Curiosity, practical, or safety.
- Step 3: Run the check. Measure, verify with a second source, or reject and go to the source with stakes.
- Step 4: Report. Write: 'The model said X. I checked and got Y. The stakes were [level]. I [trust/check/reject].'
- Talk About It: did the protocol change your decision? Would you have trusted the model without it?
Lesson guide
Ask after you try
After the protocol is run.
- Show the model the report. Ask: 'What did I miss?' If it says 'your approach is thorough,' that is flattery. If it names a specific step you skipped, that is useful. Compare its critique to your protocol. Did it catch something you missed?
- Did they run all four steps?
- Did the stakes check change the trust level?
- Is the report honest about the source and the check?
8 turns left this sitting. User-started only. Never on page load.
Light this sitting
Pair with Hermes
Currently reading WisdomForge lesson: Calibration as a Practice.
Pair this sitting
Copies the sitting card and the USER.md one-liner. The child profile reads only this card. It does not browse the catalog.
For the child profile
Paste this into the child’s USER.md. It names the sitting so the guide knows the context. The [v:1:e8b72040] tag lets you detect if the sitting’s content has changed since you paired it.
Optional: currently working on WisdomForge sitting: Measure Twice — calibration-trust. [v:1:e8b72040]
For your adult profile
Send this from your trusted adult Hermes profile. It starts the guide for this band and sitting.
You are a WisdomForge emerging guide sitting beside the lesson "Calibration as a Practice". The lesson is the text. You are the guide. Hint-first. Do not recite. Do not write the work. Warm, not a friend. If the topic is hard or tender, point to a trusted adult.
Tools on
- conversation
- optional parent-approved files
Ritual reminder
Real argument. Practice. Reflect. Chat. Optional narrow search or school files. Not an adult team agent.
Fresh profile only. Never clone an adult profile. No child names, photos, or school. Hint-first. User-started. The guide does not make AI safe. You may refuse it.
Dinner table
What claim did we calibrate this week, and did the protocol change what we did?
Sits beside
- Science. Hypothesis before search: the hypothesis includes the prediction and the test. The calibration protocol is the same structure.
- Thinking. Claim and check: the calibration protocol is the check applied to quantitative claims.
Integrity. The protocol includes 'report.' A model number without a report of the check is not a calibrated result. Report the source, the check, and the decision.