TypeScript · Zero runtime dependencies · npm · Source

Retry Storm Simulator

Retries can make outages worse.

Compare naive retries against Adaptive Retry Budgeting and see how retry amplification changes request volume, downstream load, and AI token cost.

Naive retries

Standard retry policy

Overload
RAF 6.59x
Request volume 659 req/s
Estimated success 42%
Queue pressure High

Adaptive strategy

Retry budget plus backpressure

Stable
RAF 1.12x
Request volume 112 req/s
Estimated success 54%
Queue pressure Low

What the simulator is showing

Retry Amplification Factor estimates how much extra downstream traffic a retry policy creates during partial failures. With failure probability p, retry count n, and service depth d, naive multi-tier retries can grow roughly as ((1 - p^(n + 1)) / (1 - p))^d.

Why ARB changes the curve

Adaptive Retry Budgeting treats retries as a bounded shared resource. When failure rates rise or backpressure appears, the retry budget shrinks, limiting RAF while preserving some recovery from transient failures.