# 0.4.5 Deciding, and explaining why (/how-ai-works/critical-thinking/deciding-and-explaining-why) --- type: Document title: "0.4.5 Deciding, and explaining why" description: "How much evidence a decision needs, and the one-page decision note." status: stable order: 15.5 ksor: owner: team:panaversity audience: [ public ] approval: by: process:panaversity at: 2026-11-11T08:18:17Z chapter: "0.4" part: "0" expert_status: null # as the Part 0 overview: Part 0 is optional on every route concepts: [ "0.4.5" ] last_verified: 2026-11-10 generated: at: 2026-11-11T08:18:17Z by: process:panaversity trust_tier: unverified build_id: sha256:2c8363a5436bca5eb3cc81adfe9c5cdb55380e522b3ebd7618270caef15a8351 dirty: true ksor_version: 0.0.60 --- **In everyday life.** The weather forecast says there is a 40% chance of rain. You still have to decide whether to take an umbrella. You cannot wait until you know. **What you will learn.** How much evidence a decision needs, what a good judgment includes, and how to write a one-page decision note. **Why it matters.** Critical thinking shows what you know and what you do not know. Judgment decides what to do anyway. And one person must be responsible for that decision. ## Judgment when you are not certain ### How much evidence is enough? Ana's check showed that the memo was wrong. But she still could not know what the discounts would be worth next year, and the payment run was on Friday. To decide how much evidence you need, ask two questions from [Chapter 7](../../managing-ai-workers/the-authority-envelope/choose-the-rung.md): - **How easy is the decision to undo?** This is called **reversibility**. Taking a discount on one payment run is easy to stop next week. A three-year contract is not. - **How much could you lose?** This is called **exposure**. A $26 discount needs less proof than a $148,500 payment. ![A grid of four boxes. Easy to undo and small loss: decide quickly. Easy to undo and large loss: check the key numbers. Hard to undo and small loss: check the assumptions. Hard to undo and large loss: check fully and involve the owner.](img/evidence-scale.png) *Figure 0.4.6. Match the evidence to how hard a mistake is to undo and how much you could lose.* ### What does a good judgment include? 1. What to decide now, at the size the evidence supports. 2. What you are not sure about, and how much it matters. 3. The check that would reduce the doubt, and when it will happen. 4. Who is responsible for the decision. For Brightline: take the discounts that are clearly worth more than the 9% cost of borrowing. Do not take the one that is worth less. Review after three months of real results. Dave decides. "I need more data first" is also a decision, and usually a weaker one. And when an AI sounds more confident than you, that does not make it a better decision owner. A confident tone is a way of writing. It is not evidence ([Chapter 0.1](../what-ai-actually-is/why-it-behaves.md)). Three questions help you slow down before you choose: - What happens next if we do this? - What happens after that? This is called a **second-order effect**. Paying early saves a discount. Then it uses up cash. Then Brightline has to borrow. - What would I tell a colleague to do in my position? It is often easier to see the answer for someone else. ### How do you test a decision before you make it? Do a **premortem**. Imagine that six months have passed and the decision went badly. Ask: what is the most likely reason? For Brightline the answer comes quickly: "We paid early and had to borrow from the bank." A premortem finds a risk while you can still plan for it. You can ask AI to do a premortem with you. Then check its answers like any other output. ## The decision note ### What goes on the page? Dave did not want to hear every step Ana took. He wanted to know what to do, why, and how sure she was. A decision note is one page with five parts: | Part | What goes in it | | --- | --- | | Recommendation | One or two sentences. What to do | | Evidence | The key facts, each with its source. Numbers calculated again, not copied | | Assumptions | What must be true for the recommendation to be right | | What would change it | The facts that would change the answer, and when you will check them | | Decision owner | Who decides, and what you need from them | ![A page with five blocks from top to bottom: recommendation, evidence, assumptions, what would change it, and decision owner. A note says to add one sentence on what AI wrote and what you checked.](img/decision-note.png) *Figure 0.4.7. A decision note fits on one page. The recommendation comes first.* ### How do you write it for a senior reader? Start with the recommendation. Keep the evidence short and easy to check. Say clearly where you are not sure. A senior reader trusts a note that states its own weaknesses more than a note that sounds certain. Doubt that you hide is usually found later, at a worse moment. Put the details in an appendix, which is extra pages at the end. A decision note has a date. When the facts under it change, check the decision again. ### Do you have to mention AI? Yes. [Chapter 0.3](../ai-fluency/the-four-ds.md) calls this transparency diligence: say what AI wrote and what you checked. One sentence is enough: "An AI assistant wrote the first memo. Its numbers were run again by code on the Q3 export, and I checked them against the file." Follow your organization's AI policy first. Some workplaces do not allow AI for certain work, and some require a fuller record. If the reader would benefit from seeing the conversation itself, share a link to it, but only if it contains no private data. ### When is the work finished? When you decide it is. If you ask an AI to improve a draft, it always finds something to change. It will never tell you to stop, and each round improves the draft less. The work is your responsibility, so you decide when it is ready. ## Key words - **Reversibility:** how easy a decision is to undo. - **Exposure:** how much you could lose. - **Second-order effect:** what happens after the first result. - **Premortem:** imagining that a decision failed, to find the reason before you decide. - **Decision note:** one page with the recommendation, evidence, assumptions, what would change it, and the decision owner.