Struct StartRungBandit 

pub struct StartRungBandit { /* private fields */ }

Online UCB1 bandit for start-rung selection.

Keyed by ContextBucket; per context, tracks per-rung pass/fail counts of gate verdicts. Thread-safety comes from the caller wrapping this in Arc<std::sync::Mutex<_>> (mirroring AdaptiveConformal — both are per-process, in-memory, low-contention).

Implementations

impl StartRungBandit

pub fn new(min_observations: usize, exploration: f64) -> Self

Create a new UCB1 bandit with the given cold-start and exploration settings.

pub fn with_algorithm( min_observations: usize, exploration: f64, algorithm: Algorithm, discount: f64, seed: u64, ) -> Self

Create a bandit with an explicit algorithm, discount, and PRNG seed (seed matters only for Algorithm::Thompson; pass anything non-zero).

pub fn pass_estimate(&self, ctx: &ContextBucket, rung: u32) -> Option<f64>

Posterior-mean gate-pass estimate for (ctx, rung) — None when the context is cold (below min_observations) or the arm is unobserved. Deterministic; used by the speculative-deferral band, never for serving decisions.

pub fn observe(&mut self, ctx: &ContextBucket, rung: u32, verdict: Verdict)

Record a gate verdict for (context, rung).

Abstain is ignored — only clear Pass/Fail outcomes are informative for cost modelling.

pub fn choose_start( &self, ctx: &ContextBucket, ladder: &[String], prices: &PriceTable, ) -> u32

Choose the start rung that minimises expected cost for this context and ladder.

Returns 0 in all cold-start cases:

• bandit is cold (total observations < min_observations for this context)
• ladder is empty

Never returns a value ≥ ladder.len().

pub fn choose_start_with_propensity( &mut self, ctx: &ContextBucket, ladder: &[String], prices: &PriceTable, ) -> (u32, Option<f64>)

Choose the start rung and its selection propensity.

• Algorithm::Ucb1: deterministic — returns (choice, None); the epsilon-greedy overlay (if configured) supplies the logged propensity, exactly as before.
• Algorithm::Thompson: one posterior draw per rung drives the same expected-cost argmin as UCB1 (the decision), and the propensity is Monte-Carlo estimated by re-running the draw PROPENSITY_SAMPLES times — the standard estimator for TS selection probabilities (exact computation is analytically intractable). Cold-start contexts return (0, None) (deterministic, no propensity to log).

pub fn feed_trace_attempts(&mut self, ctx: &ContextBucket, attempts: &[Attempt])

Feed all attempts from a stored trace into this bandit — used for warm-start at startup.

Trait Implementations

impl Debug for StartRungBandit

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

Formats the value using the given formatter.