Struct Policy 

pub struct Policy<K> { /* private fields */ }

A persistent table of learned weights, one per key, nudged by feedback.

Decision logic is stateless per call; Policy is the small mutable state that lets an agent improve over time. Read a key’s learned weight and fold it into an action (as its base or a consideration); after an outcome, call reward to move that weight toward what actually worked.

The update is a bounded integer moving average, so it is deterministic and can never run away — it is not machine learning, just `weight += rate * (outcome

• weight)in fixed point, clamped to0.0..=1.0`.

Example

use reliakit_decide::{Policy, Score};

// Start every key at 0.5, learning at rate 0.5.
let mut policy = Policy::new(Score::from_ratio(1, 2), Score::from_ratio(1, 2));
assert_eq!(policy.weight(&"route_a"), Score::from_ratio(1, 2)); // unseen -> default

// "route_a" worked well (outcome 1.0): its weight rises toward 1.0.
policy.reward("route_a", Score::MAX);
assert_eq!(policy.weight(&"route_a").raw(), 7_500); // 0.5 + 0.5*(1.0-0.5)

Implementations

impl<K> Policy<K>

pub fn new(rate: Score, default: Score) -> Policy<K>

Creates an empty policy with a learning rate and a default weight for keys that have not been seen yet.

pub fn len(&self) -> usize

The number of keys that have learned weights.

pub fn is_empty(&self) -> bool

Returns true if no key has a learned weight yet.

pub fn entries(&self) -> &[(K, Score)]

The learned (key, weight) pairs, for snapshotting to storage. The host serializes these however it likes — the engine pulls in no serializer. Restore them later with set.

impl<K: PartialEq> Policy<K>

pub fn weight(&self, key: &K) -> Score

The learned weight for key, or the configured default if it is unseen.

pub fn reward(&mut self, key: K, outcome: Score)

Nudges key’s weight toward outcome by the learning rate. A previously unseen key starts from the default before moving.

pub fn set(&mut self, key: K, weight: Score)

Sets key’s weight directly (insert or replace) — used to restore learned weights from storage. Unlike reward this does not apply the learning rate; it stores the value as given.

Trait Implementations

impl<K: Clone> Clone for Policy<K>

fn clone(&self) -> Policy<K>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<K: Debug> Debug for Policy<K>

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

Formats the value using the given formatter.