This week's learning objectives
Learning objectives
By the end of this week, you will be able to:
Build a rough back-of-the-envelope calculation (BOTEC) for a decision you face
Translate a gut feeling into a calibrated probability you’d defend
Use three forecasting techniques — Fermi decomposition, base rates, and considering the other side — to sharpen your estimates
Adjust raw expected value for credit, speed-up, and persistence to get honest counterfactual impact
Identify the numbers doing the real work in a BOTEC, and critique your own and others’ reasoning
Time commitment
Reading: ~55 minutes (course text + two Coefficient Giving articles)
Calibration tool: ~20 minutes
Forecasting exercise: ~10 minutes
Quizes: ~10 minutes
Summative BOTEC: ~30 minutes
Total pre-session work: ~2 hours 5 minutes
A practical note
This week has one exercise (in Lesson 4) and one summative BOTEC (in Lesson 6) that ask you to write things out. You can do that wherever works for you — a notebook, a doc, the back of an envelope. Nothing needs to be submitted. The writing is for you, and you’ll bring your completed BOTEC to the facilitated session so you and a partner can critique each other’s work.
Where we’ve been
Week 1 established the project: that some ways of doing good are orders of magnitude better than others, that our intuitions don’t scale, and that making your reasoning transparent is the foundation skill everything else rests on.
Week 2 gave you the evidence-based toolkit: theories of change, the evidence hierarchy, QALYs and DALYs and WELLBYs, and the mechanics of a cost-effectiveness model. You read a GiveWell CEA critically and built a rough one yourself. You also met the limits of that toolkit — it works best for direct interventions with short feedback loops, and struggles with anything where the causal chain is long or the outcomes are hard to measure.
This week picks up at those limits. It’s also the methodological response to something Week 1 flagged: EA’s own recognition of measurability bias — the historic tendency to count what could be counted, and to neglect systemic change and speculative-but-potentially-transformative work as a result. A BOTEC is one specific tool for that response. It extends careful quantitative reasoning into hits-based territory — policy, movement building, long causal chains — without pretending the answers are tidy. It doesn’t do everything. Next week takes on a different kind of uncertainty altogether (moral, not empirical) and a different question (which problems are worth working on at all). Treat this week as one piece of a larger toolkit, not the whole answer.
The shape of the week
You’re learning one tool — the BOTEC — and the skills to fill it in honestly. A BOTEC combines how much good something might do with how likely it is to work. Doing that well means producing numbers for value and probability that you’d actually defend, which is what forecasting is.
Forecasting needs one prerequisite: calibration (Lesson 3). If you can’t give honest numbers in the first place, no technique on top of that will save you. Three techniques then do most of the work — Fermi decomposition for magnitudes, base rates for probabilities, and considering the other side as a check on both (Lesson 4).
Finally, three adjustments turn raw expected value into counterfactual impact — credit, speed-up, and persistence (Lesson 5). Lesson 6 is the assembly job: you’ll build a BOTEC for a decision you actually face, and bring it to the session for critique.
Why this matters
Last week’s toolkit was powerful but narrow. RCTs, DALYs, transparent spreadsheets. It works for direct interventions with measurable outcomes.
Many important things don’t fit that mould. Policy change works through complex causal chains with murky counterfactuals. Movement building might take decades to bear fruit. Some moral patients — animals, future generations — can’t fill out surveys about their wellbeing. Sometimes the deepest uncertainty isn’t empirical (what works?) but moral (what matters?).
The aim this week isn’t to abandon rigour; it’s to extend it. A BOTEC is a way of reasoning carefully when evidence is limited but stakes are high. The numbers will be wrong. Making them explicit lets you see how they’re wrong, learn from results, and disagree with someone without talking past them.
We’ll use grant evaluation as the running example because it’s concrete and the tools are well-developed. But the skills transfer. The same framework applies to career decisions (which path has higher expected impact?), time allocation (volunteer or earn to give?), personal choices (how much weight to give animal welfare?), and community strategy (where should we focus?). Reasoning under uncertainty is the core skill; grants are the training ground.