This is the whole course in one sitting. Hold the through-line in your head: things meet disorder in exactly three ways — fragile things are harmed by it, robust things ignore it, and antifragile things gain from it. The dividing line is not strength but curvature: a concave (accelerating-harm) response is fragile, a convex (accelerating-gain) response is antifragile, and by Jensen’s inequality volatility helps the convex and hurts the concave. From that one idea flows everything — hormesis (a small dose of stress triggers a stronger rebuild), the dark twin where removing stressors breeds hidden fragility (the turkey, the suppressed fire), the practical fix of via negativa (robustify by subtraction — the barbell, skin in the game), the skill of spotting fragility in advance (it’s a structural property, far more predictable than the shock), and the honest limits (the dose matters, survivorship bias, things you just protect). Take a breath. There is no going back once you commit.
How this exam works
This is a real exam, not a practice quiz. Questions appear one at a time. Once you submit an answer it is locked for good — there is no going back, no retry, and no restart. Your score stays hidden until the very end. A few questions ask you to select all that apply (read those carefully — partial credit is not a thing here). You need 70% to pass. Ready when you are.
The TRIAD is the backbone of the whole model. Which row correctly matches each response to disorder?
Select an answer to continue.
Big picture
Antifragility & via negativa, in one picture
- Antifragility & via negativa
- The triad & curvature
- Fragile = harmed by disorder (concave response); Robust = unchanged (flat); Antifragile = improved (convex). The line is CURVATURE, not strength. By Jensen's inequality, volatility helps the convex and hurts the concave. A concave thing is hurt more by one big shock than by many small ones summing to the same
- Hormesis & overcompensation
- A small, survivable dose of a stressor (harmful in large amounts) triggers a beneficial adaptive OVER-response: the system rebuilds MORE than was lost, treating the shock as information about a bigger one to come. Muscle, bone (Wolff's law), immunity/vaccines, desirable difficulties. The dose-response is an inverted-U — benefit only in the low-dose zone
- The touch of chaos
- Removing an antifragile system's stressors makes it fragile. The turkey (calm track record hides peak risk); the Lucretius problem (designing for the worst OBSERVED, not worst possible); stability breeds instability (Minsky, fire suppression -> megafires); redundancy/slack is hidden strength that over-optimisation strips away; iatrogenics of naive intervention
- Via negativa
- Robustify by SUBTRACTION — remove the fragilising agent (debt, single points of failure, the cigarette) rather than add a speculative fix, because removing a known harm is more reliable than adding an uncertain good (iatrogenics). The BARBELL: extreme safety + a little bounded risk, no fragile middle (capped downside, open upside). SKIN IN THE GAME: bear your own downside — it removes the hidden transfer of fragility to others
- Spotting fragility in advance
- Fragility is FAR more predictable than the shock — it is structural, visible now (no forecast needed). The convexity test: perturb +/- and see if variability makes the average WORSE (concave = fragile). Tells: debt, single points of failure, no slack / over-optimisation, dependence on a forecast, hates variance. A calm track record is not robustness (the turkey). Ask who bears the tail
- Where the model lies
- The DOSE matters — convexity holds only over a range; enough stress kills even the antifragile. "What doesn't kill me" is often SURVIVORSHIP BIAS (the destroyed don't testify; systems toughen by culling fragile parts while individuals are just harmed). Some fragile things you just PROTECT (the irreplaceable/irreversible). "Antifragile" is not a synonym for resilience, and not a licence for recklessness or fragility transfer. Respect RUIN
- The triad & curvature
Key takeaways
You now hold the whole model. Things meet disorder in exactly three ways — fragile (harmed), robust (unchanged), antifragile (improved) — and the dividing line is curvature, not strength: a concave response (harm accelerates) is fragile, a convex one (gain accelerates) is antifragile, and by Jensen’s inequality volatility helps the convex and hurts the concave. The biological engine is hormesis — a small, survivable dose of a stressor triggers overcompensation, rebuilding more than was lost (muscle, bone, immunity, desirable difficulties) — with the catch that the dose-response is an inverted-U, so only the low-dose zone helps. Its dark twin: removing an antifragile system’s stressors makes it fragile — the turkey mistakes calm for safety, the Lucretius problem designs for the worst observed shock, stability breeds instability (suppressed fires and recessions), and redundancy/slack is the hidden strength that over-optimisation strips away. The practical fix is via negativa: robustify by subtraction — remove the fragilising agent (debt, single points of failure, the toxin) because a known removal beats a speculative addition (iatrogenics) — sharpened by the barbell (extreme safety plus a little bounded risk, no fragile middle: capped downside, open upside) and skin in the game (bear your own downside; it removes the transfer of fragility to others). Best of all, fragility is far more predictable than the shock — it’s structural and visible now: run the convexity test and hunt the tells (debt, concentration, no slack, a calm track record that proves nothing). And the model lies at its edges: the dose matters (enough stress kills even the antifragile), “what doesn’t kill me” is often survivorship bias, some fragile things you simply protect, “antifragile” is not a synonym for resilience or a licence for recklessness, and you must always respect ruin. Never ask only “how do I make this stronger?” Ask: is this a thing that would gain from a little disorder — and if it’s fragile, what can I take away to save it?