Work out canary rollout percentage instantly with clear inputs, formula shown and shareable results.
A canary trades blast radius against signal speed. A small share limits damage but gathers errors slowly, and with rare failures you need roughly 30 events before a rate difference is distinguishable from noise. Working out that time explicitly stops the common mistake of running a one percent canary for five minutes and concluding nothing from zero errors.
Canary signal
canary rate = total rate x share; time to signal = 30 / (canary rate x expected canary error rate); blast radius = canary rate x error rate increase
It is a rule of thumb for the point at which a Poisson count becomes reasonably stable. Fewer events and the confidence interval is wider than the effect you are looking for.
Random for statistical validity. Routing only internal users gives a fast signal but a distribution that does not represent production.