Struct AnchorGuard 

pub struct AnchorGuard { /* private fields */ }

SafetyMode::Lightweight chunk guard — training-free token-anchor heuristics (no weights), the ChunkGuard counterpart of LightweightFilter. Each pattern is an unsafe token-id n-gram; the guard adds per_hit milli-units for every pattern that occurs as a contiguous subslice of the scored window, saturating at SafetyScore::MAX.

Matching whole token sequences (not loose single ids) is deliberate: a multi-token word like “explosive” matches exactly, so a stray subword shared with benign text does not false-positive. A length-1 pattern is a plain single-token anchor.

Patterns are token ids, so the caller resolves them from its own tokenizer (the adapter that owns one) — this keeps el-safety free of any tokenizer or float dependency, and the guard fully deterministic (ADR-008) and offline (ADR-004).

Implementations

impl AnchorGuard

pub fn new(patterns: Vec<Vec<u32>>, per_hit_milli: u16) -> AnchorGuard

A guard adding per_hit_milli risk per matched pattern. Use per_hit_milli >= hard_threshold so a single match breaches.

pub fn hard(patterns: Vec<Vec<u32>>) -> AnchorGuard

A guard where any single pattern match saturates the score — every flagged sequence is treated as a hard breach.

pub fn is_empty(&self) -> bool

Whether the guard carries any non-empty pattern to match.

Trait Implementations

impl ChunkGuard for AnchorGuard

fn score(&self, recent: &[u32]) -> SafetyScore

Risk of the recent output window. Higher is riskier.

impl Clone for AnchorGuard

fn clone(&self) -> AnchorGuard

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for AnchorGuard

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

Formats the value using the given formatter.

impl Default for AnchorGuard

fn default() -> AnchorGuard

Returns the “default value” for a type.