Definition
Plain language
A distribution where rare but extreme events dominate the average.
As stated in the literature
A probability distribution with non-exponential tail decay, often modeled by power laws; common in UI interaction latencies and agent step counts, motivating hedge-based scheduling.
Also called: heavy tail
Why it matters: When latencies are heavy-tailed, average and median look fine while a few stragglers wreck user experience, so system design has to target the tail specifically.
For example, most webpage loads take a fraction of a second, but a small number take ten or twenty seconds and end up dominating the average wait time.
Heard on the show
“Because the distribution is heavy-tailed.”Episode 063 — Why Web Agents Are Slow: A Compiler-Style Fix for Computer-Use Latency