Wrapping up
This week you built a BOTEC. To do that honestly you needed to forecast — to produce numbers for value and probability that you’d defend. Forecasting rests on one prerequisite and three techniques:
Calibration (the prerequisite): give numbers that track reality at the rates you claim.
Fermi decomposition for magnitudes: break a hard-to-estimate quantity into a product of easier ones; flush your ignorance into the open so someone can actually disagree with it.
Base rates for probabilities: anchor on what usually happens in cases like this, then adjust. Decompose multi-step probabilities when the outcome depends on several things going right.
Considering the other side as a check: pre-mortem, steel-man the sceptic.
Three adjustments then turn raw expected value into counterfactual impact: credit (what fraction is actually yours), speed-up (what would have happened anyway), persistence (how long the change sticks).
What a BOTEC is — and isn’t — for
The BOTEC won’t give you certainty. It gives you transparency (others can disagree with specific inputs rather than the whole conclusion), learning (if you predicted 30% and it didn’t happen, you can ask why), and comparability (grants, career choices, and time allocations on the same footing).
It’s also worth knowing when not to trust a BOTEC. The tool has real limits.
When the uncertainty is about values, not facts. “How much does a chicken’s suffering matter compared to a human’s?” isn’t a forecast — it’s a moral question. A BOTEC with different moral weights gives different answers, and the tool itself can’t tell you which weights are correct. That’s moral uncertainty, and it’s what Week 4 takes up directly.
When the situation is genuinely unprecedented. There’s no reference class for transformative AI. A base rate built from scratch can look quantitative while being no better than a clearly marked guess — and worse, because the number carries more authority than the hand-waving it replaced.
When the question is which problem to work on at all. A BOTEC compares options inside a frame; it can’t tell you whether the frame is the right one. “Why animal welfare rather than global health?” is a worldview-level question, not a spreadsheet-level one. (Again, Week 4.)
When your answer swings by an order of magnitude on one input you could defensibly change. That’s the model telling you that you don’t know yet. The right response is often “run a smaller experiment first,” “ask someone who knows more,” or “make this decision differently” — not “trust the number.”
This is also why the summative asks which of your numbers is most load-bearing — that’s where your doubt belongs, and that’s what you and your partner will push on each other in the session.
Remember the move from Lesson 1, too: a single BOTEC evaluates a single bet. The hits-based logic — accepting that most bets return nothing because the few that hit are large enough to pay for the losses — is a claim about the portfolio, not a reason to skip the BOTEC on any individual bet. If anything, it’s the reason to do the BOTEC carefully: you want to know which bets are worth making when most are going to fail.
The skills are general
Everything this week was framed around grantmaking because the tools are well-developed there. The skills are general. A BOTEC is how you reason about any decision where stakes are high, evidence is limited, and you have to act anyway — including the one 80,000 Hours argues is probably the highest-leverage decision you’ll make in your life, which is what to do with your career.
What to bring to the session
Your completed BOTEC (Option A or B) open on your screen.
Your forecast from the Lesson 3 exercise.
A note of your calibration result from Clearer Thinking.
In the session you’ll swap BOTECs with a partner and critique each other’s reasoning. The skill of catching a weak number in someone else’s model is how you learn to catch it in your own.
Preview: Week 4
Next week picks up the hardest question yet: which problems should we work on in the first place? You’ll learn a framework for comparing cause areas, and you’ll confront a different kind of uncertainty — not empirical (what will happen?) but moral (what matters?). Different ethical frameworks give different answers, and your moral uncertainty turns out to have real practical consequences for how you prioritise.
Summary
This week you learned to:
Build a rough BOTEC for a decision you face
Calibrate your estimates so the numbers mean what you claim they mean
Estimate magnitudes using Fermi decomposition
Estimate probabilities using base rates and conditional decomposition
Stress-test your estimates by considering the other side
Adjust raw expected value for credit, speed-up, and persistence
Identify which of your numbers is load-bearing — and treat that as where your doubt belongs
Reasoning under uncertainty isn’t waiting until you’re sure. It’s making your uncertainty explicit — and deciding anyway.