Struct UntilElapsedPolicy 

pub struct UntilElapsedPolicy {
    pub min_ms: u64,
    pub delta: u64,
}

Extend while ctx.elapsed_ms < min_ms, projecting the remaining work from the observed rate and committing it in one batch.

Per call: estimate repeats = floor((consumed * remaining_ms / elapsed_ms) / base) — the number of base-sized passes needed to fill the time remaining at the current rate. Apply a 5% under-bias (repeats * 95 / 100) so successive extensions converge from below the target rather than overshooting once. Commit repeats * base cycles.

The under-bias gives a geometric convergence: each call covers ~95% of the time remaining, so 3–4 calls typically land within 0.01% of the time budget. The first call (no rate signal yet — elapsed_ms == 0 or consumed == 0) falls back to one base chunk via the delta field.

Fields

min_ms: u64

The elapsed milliseconds to reach.

delta: u64

The extension step for the first call, before a rate is known.

Trait Implementations

impl ExtensionPolicy for UntilElapsedPolicy

fn next_extension(&self, ctx: &ExtensionContext) -> Option<u64>

Decide how to extend (if at all) given the current cursor context.