l5-antifragile_lucretius-problem

The Lucretius problem is the error of treating the worst observed event as the worst possible event. Named after the Roman philosopher Lucretius, it describes how risk models anchored to historical maximums are blindsided by unprecedented extremes.

The hundred-year flood concept embeds the Lucretius problem by assuming past flood distributions reliably predict future ones. Since the true distribution is unknown and fat-tailed, nature routinely produces events worse than recorded history suggests.

Stop trying to predict the worst case and instead build systems robust to the unknown. The Lucretius problem shows prediction of extremes is impossible, so design for resilience rather than optimization around known historical boundaries.

The Fukushima nuclear plant was designed for the worst earthquake in recorded history, and nature delivered a worse one. Similar failures occur whenever engineers, actuaries, or planners treat historical maximums as absolute limits.

Historical data only reveals the worst that has happened so far, not the worst that can happen. In fat-tailed domains, the distribution itself is unknown, so empirical memory systematically underestimates nature's capacity for extreme outcomes.