Struct PolicySnapshot 

pub struct PolicySnapshot {
    pub policy_epoch: u64,
    pub catalog_epoch: u64,
    pub hard_risk_limit_bps: u16,
    pub minimum_confidence_bps: u16,
    pub risk_penalty_multiplier_bps: u16,
    pub latency_penalty_microunits_per_ms: u64,
    /* private fields */
}

An immutable snapshot of the decision policy and model catalog.

Create with PolicySnapshot::try_new (validated) or PolicySnapshot::new_unchecked, then call prescribe for each request. The snapshot is Clone and can be shared across threads via Arc.

Fields

policy_epoch: u64

catalog_epoch: u64

hard_risk_limit_bps: u16

minimum_confidence_bps: u16

risk_penalty_multiplier_bps: u16

latency_penalty_microunits_per_ms: u64

Cost penalty per millisecond of p95 latency, in microunits.

No static upper bound — overflow is dynamically guarded by all_latencies_fit which falls back to i128 arithmetic when max_p95_latency_ms * latency_penalty_microunits_per_ms would overflow u64.

Implementations

impl PolicySnapshot

pub fn new_unchecked( policy_epoch: u64, catalog_epoch: u64, hard_risk_limit_bps: u16, minimum_confidence_bps: u16, risk_penalty_multiplier_bps: u16, latency_penalty_microunits_per_ms: u64, models: Vec<KernelModel>, ) -> Self

Creates a policy snapshot without validation (escape hatch).

Production guidance

Use try_new for any policy served to real traffic. new_unchecked is for unit tests, fuzz fixtures, and experiments where you call validate yourself or intentionally construct invalid catalogs.

Serving from an unchecked snapshot without a successful validate() is an audit failure.

pub fn new( policy_epoch: u64, catalog_epoch: u64, hard_risk_limit_bps: u16, minimum_confidence_bps: u16, risk_penalty_multiplier_bps: u16, latency_penalty_microunits_per_ms: u64, models: Vec<KernelModel>, ) -> Self

👎Deprecated since 0.3.9:

use PolicySnapshot::try_new for validated snapshots or new_unchecked for tests

Creates a new policy snapshot from a model catalog (alias for new_unchecked).

pub fn models(&self) -> &[KernelModel]

Returns the model catalog.

pub fn validate(&self) -> Result<(), PolicyError>

Validate catalog and basis-point invariants before serving traffic.

pub fn try_new( policy_epoch: u64, catalog_epoch: u64, hard_risk_limit_bps: u16, minimum_confidence_bps: u16, risk_penalty_multiplier_bps: u16, latency_penalty_microunits_per_ms: u64, models: Vec<KernelModel>, ) -> Result<Self, PolicyError>

Build a snapshot and validate the catalog.

pub fn prescribe_batch(&self, inputs: &[KernelInput]) -> Vec<KernelDecision>

Evaluate many inputs. Allocates the output vector only.

pub fn prescribe_batch_checked( &self, inputs: &[KernelInput], ) -> Result<Vec<KernelDecision>, InputError>

Validate and evaluate many inputs, failing before producing decisions when any request is structurally invalid.

pub fn prescribe_with_trace( &self, input: KernelInput, ) -> (KernelDecision, DecisionTrace)

Evaluate input and return decision plus rejection histogram.

pub fn prescribe_with_trace_checked( &self, input: KernelInput, ) -> Result<(KernelDecision, DecisionTrace), InputError>

Validate input, then evaluate it with an explainability trace.

pub fn prescribe(&self, input: KernelInput) -> KernelDecision

Evaluate input against the policy and return the optimal decision.

The kernel checks 11 constraint gates per candidate, computes utility as quality_adjusted_value - risk_penalty - cost - latency_penalty, and selects the candidate with the highest positive utility.

If no candidate has positive utility, the request is rejected (fail-closed). The decision also records the counterfactual (second-best) candidate.

This function does not allocate.

pub fn prescribe_checked( &self, input: KernelInput, ) -> Result<KernelDecision, InputError>

Validate input, then evaluate it.

This is the recommended API at untrusted Rust boundaries. The unchecked prescribe path remains available for callers that have already validated or constructed the input through a safe builder.

pub fn utility_for_model( &self, input: KernelInput, model_id: u32, ) -> Option<i64>

Evaluate utility for a specific catalog model if it passes all constraint gates.

Unlike prescribe, this does not rank candidates — it only answers whether model_id is eligible and what its utility would be.

Trait Implementations

impl Clone for PolicySnapshot

fn clone(&self) -> PolicySnapshot

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for PolicySnapshot

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

Formats the value using the given formatter.