Discussion Guide — Chapter 42

Six prompts. Each is written to have no comfortable answer.

1. Someone you trust, who has no financial interest of any kind, tells you a peptide worked for them. How much should that move you — and does the absence of a financial interest make it move you more than it should?

What to listen for: students who correctly identify this as a single uncontrolled observation, and then — the harder step — notice that "no financial interest" removes one distortion while adding nothing. Listen for anyone who says the absence of incentive makes the anecdote better evidence; that is the symmetric error and it is worth surfacing early. The best answers arrive at: the report is worth something as a prompt to investigate and nothing as evidence of effect.

2. A creator discloses an affiliate arrangement, states the evidence accurately, includes the qualifiers, and links the primary literature. Is anything still wrong? If so, what — and who could fix it?

What to listen for: whether the class can locate the residual problem outside the piece of content. The individual video is unimprovable; the distortion is in the videos that were never made. Listen for students trying to fix it with better disclosure — that is the reflex §42.3 is designed to break. Push toward: no property of any single artifact can address a population-level absence.

3. Design a ranking system that favors well-evidenced health content. Specify the input.

What to listen for: the moment the specification fails. Every proposal reduces to a proxy — credentials, citation presence, source domain, institutional endorsement — and each proxy is gameable and excludes legitimate speakers. Strong students end up proposing a human-adjudicated layer and then notice it cannot scale. This is the right conclusion. Do not rescue them from it.

4. A clinician with a large audience says something careful and correct. It underperforms. A month later they say something confident and slightly overstated, and it reaches a hundred times more people. What, exactly, did they learn — and what should they do?

What to listen for: whether students recognize the learning as structural rather than moral. The creator did not decide to be less careful; they received a signal. Then push on the second half: any prescription for the individual ("be careful anyway") loses the audience to someone who isn't. Listen for anyone who reaches the population-level answer — this cannot be solved one creator at a time — and make them say what the alternative is.

5. The absence of an approval made advertising for compounded preparations less constrained, not more. Is this defensible policy? If not, what exactly would you change, and what breaks?

What to listen for: first, whether they can restate the mechanism correctly (rules attach to a label; no label, nothing to attach to). Then whether their fix survives contact with the reason compounding exists at all — patients who need preparations manufactured products do not supply. Listen for proposals that would eliminate legitimate compounding as collateral damage, and ask who that harms.

6. You have just learned to read content for its business model. Name the piece of health information you currently believe that this skill is most likely to make you wrong about.

What to listen for: the quality of the discomfort. Weak answers name a belief they already held loosely. Strong answers name something they are attached to and identify the specific mechanism by which the new tool would let them dismiss a challenge to it. This prompt is the calibration test (§42.9) in conversational form, and it is the one worth ending on — because a class that can only run the tool on claims it dislikes has learned a rationalization engine, not a method.