Struct BanditConfig 

pub struct BanditConfig {
    pub min_observations: usize,
    pub exploration: f64,
    pub algorithm: BanditAlgorithm,
    pub discount: f64,
}

Config for the UCB1 start-rung bandit (firstpass_proxy::bandit::StartRungBandit).

Absent (None) → start every request at rung 0 (byte-identical to today).

Fields

min_observations: usize

Minimum total gate-verdict observations in a context bucket before the bandit’s prediction kicks in. Below this, every request starts at rung 0 (cold-start safety).

exploration: f64

UCB1 exploration constant c (Auer et al. 2002). Higher values explore more; 1.0 is the theoretical default. Must be finite and >= 0; validated by Config::parse.

algorithm: BanditAlgorithm

Selection algorithm. "ucb1" (default) is deterministic and auditable; "thompson" samples Beta posteriors — stochastic by nature, so every decision logs a non-degenerate propensity (clean IPS/SNIPS/DR off-policy estimates without the epsilon overlay) and pairs with discount for non-stationary traffic.

discount: f64

Per-observation multiplicative decay of a context’s counts, in (0, 1]. 1.0 (default) = no forgetting. 0.99 adapts to model churn / workload drift at the cost of a slightly noisier estimate. Validated by Config::parse.

Trait Implementations

impl Clone for BanditConfig

fn clone(&self) -> BanditConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for BanditConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for BanditConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.