Struct WriteInferenceEngine 

pub struct WriteInferenceEngine { /* private fields */ }

Implementations

impl WriteInferenceEngine

pub fn new() -> Self

pub fn with_config(config: WriteInferenceConfig) -> Self

pub fn looks_like_value_update(&self, new: &str, old: &str) -> bool

Whether new and old look like a value update (same sentence frame, changed value tail) under this engine’s configured thresholds. A deduplication pass that merges near-duplicates uses this as a guard so a correction (“coffee is a cortado” then “coffee is a flat white”) is never collapsed into the fact it corrects; the two must stay distinct for supersession to resolve them.

pub fn infer_on_write( &self, new_memory: &MemoryNode, existing_memories: &[MemoryNode], existing_edges: &[(MemoryId, MemoryId, EdgeType)], ) -> Vec<InferredAction>

pub async fn infer_on_write_with_llm<J: LlmJudge>( &self, new_memory: &MemoryNode, existing_memories: &[MemoryNode], existing_edges: &[(MemoryId, MemoryId, EdgeType)], llm: &CognitiveLlmService<J>, ) -> Vec<InferredAction>

LLM enhanced write inference. Uses the CognitiveLlmService when available to make smarter invalidation and contradiction decisions. Falls back to cosine similarity heuristics for moderate similarity (Related edges) and correction handling.

Trait Implementations

impl Default for WriteInferenceEngine

fn default() -> Self

Returns the “default value” for a type.