Over-Optimization Fragility (Taleb's Turkey)

The pattern where a system optimized for a specific set of conditions becomes catastrophically fragile when conditions change. Taleb's turkey metaphor: a turkey is fed every day, and each day of feeding increases its statistical 'confidence' that it will be fed tomorrow. This confidence is highest the day before Thanksgiving. The mechanism: (1) a system is optimized for observed conditions, (2) the optimization process removes redundancy, buffers, and optionality (because they look like waste in the observed conditions), (3) the system becomes maximally efficient for the expected environment, (4) conditions change in a way not captured by the optimization model, (5) the very features removed for efficiency (buffers, redundancy, optionality) were the features needed to survive the new conditions. The 2021 Texas grid failure exemplified this: optimized for normal Texas weather, the system lacked the cold-weather hardening that seemed wasteful in a warm climate—until the unprecedented freeze hit. LTCM's failure in 1998 followed the same pattern: models optimized on historical data couldn't handle conditions outside the historical range.

When to use it

When optimizing any system for efficiency. When evaluating whether a system is robust or merely efficient. When conditions are stable (paradoxically, this is the highest-risk period for over-optimization). When designing systems that must survive conditions beyond current experience.

How it can help

Provides specific 'anti-fragility' design principles. (1) After any optimization, ask: 'What range of conditions does this optimization assume? What happens outside that range?' (2) Maintain 'strategic redundancy'—capacity that looks wasteful under normal conditions but provides survival under stress. (3) Apply Taleb's barbell strategy: be hyper-conservative in most areas (protect against ruin) and opportunistically aggressive in others (benefit from volatility). (4) Stress-test against scenarios your optimization model doesn't contemplate. (5) The rule: if the system hasn't been tested under the conditions you're protecting against, the protection is theoretical.