Function decide 

pub fn decide(
    policy: &RetryPolicy,
    attempt: u32,
    server_retry_after: Option<Duration>,
    elapsed: Duration,
    jitter_unit: f64,
) -> RetryDecision

Decide what to do after attempt attempts have failed.

attempt counts completed attempts, so it is 1 after the first failure. elapsed is the time consumed by the sequence so far. jitter_unit is a caller-supplied value in [0.0, 1.0); it is clamped defensively so a bad caller cannot produce a negative or unbounded delay.

Delay derivation

Without a server hint, this is full jitter — random(0, min(cap, base·2ⁿ)). The failure mode being defended against is several clients colliding on one model’s startup, and full jitter is the variant that decorrelates them most aggressively. Fixed backoff would have every waiter wake together and collide again.

With a Retry-After, the server’s value is treated as a floor rather than replaced by jitter: retrying earlier than the server asked just burns an attempt against a resource known to be unready. A small jitter of up to initial_backoff is added on top, so concurrent clients handed the same Retry-After still spread out. The floor is clamped to max_backoff first, so a buggy or hostile upstream cannot park a request indefinitely.

A delay that would overrun total_deadline yields GiveUpReason::DeadlineExceeded rather than a truncated sleep: waking early, before the moment the server nominated, is worse than stopping.