Discussion Guide — Chapter 37

Six prompts. Each is written to be run cold, and each has a predictable wrong turn that is the actual teaching moment.

1. "Which would you rather have: no evidence, or evidence that says no?"

Run it as a forced choice with a show of hands before any discussion.

Listen for: the split in the room, and specifically for people choosing "no evidence" because it feels like preserved possibility. That is the intuition §37.3 is correcting, and hearing it stated out loud by a peer is worth more than reading it. Listen also for the sophisticated version — "it depends on what I'm doing with it" — which is the right answer and should be drawn out: for predicting whether a row will move, absence is the volatile state; for deciding how much weight to put on current knowledge, absence is the impoverished one. Redirect if the discussion becomes about specific compounds; keep it abstract until the principle is secure.

2. "Is semaglutide good?"

Ask it flatly. Let people answer before you complicate it.

Listen for: confident answers in either direction, and note who is implicitly answering about which row. Someone will answer about weight loss (✅), someone about Alzheimer's (🔬), someone about a compounded product (❌). The moment to seize is when two students disagree and both are right — name it immediately: they are answering different questions, and the argument is unresolvable because the question is malformed. Listen for the student who asks "in whom, for what?" before answering. Say publicly that this is the whole chapter.

3. "The ratings come out 50 ✅ and 50 ❌. What does that tell you about peptides?"

Listen for: the trap answer — "peptides are about half nonsense" — which treats the distribution as a property of molecules. Push with: "would the ratio be the same if a different set of compounds had been funded for trials?" That question usually cracks it. Listen for students reaching independently for the funding explanation; when someone does, hand them §37.8's by-block table and let them find the pattern themselves rather than confirming it. Watch for the overcorrection into conspiracy — that trials were run and suppressed. The chapter's claim is narrower and stronger: for most of these compounds no application was ever filed, because no sponsor has a route to recovering the cost. Absence of a filing is not suppression.

4. "Read the Chapter 44 obesity claim aloud. Which single word makes it unrateable?"

Put the sentence on the board in full. Give ninety seconds of silence before taking answers.

Listen for: "obesity," "society," "should" — all near misses worth exploring. The answer is "primarily," and the reason is worth extracting rather than supplying: the two approaches are not mutually exclusive, so the word is not describing a fact about the world but an allocation of attention. Follow up with: "rewrite it so it is rateable." The rewrites will be revealing — most students produce something narrow and empirical and then notice, unprompted, that they have answered a different question than the one that was interesting. That noticing is the objective.

5. "Both generous drift and severe drift are errors. Which is more respectable?"

Expect the room to say severe drift is more respectable. Let them, and then work on it.

Listen for: the argument that skepticism is right more often in a market full of nonsense — which is true, and is exactly what makes severe drift hard to detect. Draw out the distinction between being right and knowing. Listen for anyone willing to name their own drift out loud; that is the best thing that can happen in this discussion and should be met with interest rather than reassurance. Do not let the group settle on "so we should just be balanced." The point is not moderation, it is that both failures substitute a prior about the source for an examination of the evidence.

6. "Is a disagreement with the table an error?"

Best run last, after the drift audit is assigned.

Listen for: the assumption that the book is the answer key. It is not, and saying so plainly changes how the audit gets done. The distinction to land: a disagreement you can defend — naming the evidence that moved you and what would move you back — is correct work; a disagreement you cannot defend is where your bias lives. Listen for students who want a rule for when disagreement is allowed. There isn't one, and the absence is the point: the standard is whether you can state your reasoning in the four-line form, not whether an authority has licensed you.

Close on this: ask the group what they would need to see, five years from now, to conclude that one of tonight's ❌ rows should have been ⚠️ all along. The ones who can answer have the skill the table is a byproduct of.