Embodied AI Glossary中文

Number of Nines (Reliability)

几个 9(可靠性)Advanced

Counting the 9s in a percentage to describe reliability or success rate, e.g. 99.9% is “three nines.”

“Number of nines” is a shorthand for describing reliability by counting consecutive 9s in a percentage: 99.9% is “three nines,” 99.99% is “four nines.” It became popular first for describing IT system uptime — three nines corresponds to roughly 8.76 hours of downtime a year, four nines to about 53 minutes. In embodied AI it's often used for task success rate: hitting 90% in a lab demo isn't especially hard, but a factory line often demands 99.9% or higher. Each additional nine cuts the failure count by roughly an order of magnitude, and the failures that remain tend to be rare, long-tail cases, which is why each additional nine gets harder to reach than the last — one reason people often say going “from demo to deployment” is so difficult.

ExampleIf a sorting line picks 10,000 times a day, a 99% success rate means about 100 failures a day requiring human handling, while 99.99% means roughly 1.

Also called
Three Nines / Four Nines
Related
Success Rate · Long-tail Problem · Mean Time Between Failures · Robustness · Mean Time Between Interventions
Sources
High availability - Wikipedia(「nines」表示法) (Chinese)

See it in the full glossary →