What it is
It turns a provider limit such as 500 requests per minute plus 30,000 tokens per minute into a sustainable requests-per-second figure, then compares it with how fast your code actually calls the API. Several limits can apply at once and the tightest one wins. A Monte Carlo simulation replays random request sequences through a fixed-window, sliding-window or token-bucket limiter and reports how often at least one request would get HTTP 429.
When to use it
Use it before integrating a third-party API, when sizing a worker pool or batch job, or when you keep seeing 429 responses and need to know whether the average rate or the bursts are the problem.
Gotchas & limits
An average rate below the limit can still trip it: bursts and Poisson-style arrivals spike above the average. Window alignment also matters, because a fixed window allows a double burst across a boundary. Providers apply limits per model and tier, so check the official docs; the presets are reference values only.