Trait Grounding 

pub trait Grounding {
    // Required methods
    fn ground(&self, claim: &str) -> SisterResult<GroundingResult>;
    fn evidence(
        &self,
        query: &str,
        max_results: usize,
    ) -> SisterResult<Vec<EvidenceDetail>>;
    fn suggest(
        &self,
        query: &str,
        limit: usize,
    ) -> SisterResult<Vec<GroundingSuggestion>>;
}

Grounding capability — sisters that verify claims implement this.

Implemented by: Memory, Vision, Identity, Codebase NOT implemented by: Time (no grounding concept)

The three methods mirror the actual {sister}_ground, {sister}_evidence, and {sister}_suggest MCP tools.

Required Methods

fn ground(&self, claim: &str) -> SisterResult<GroundingResult>

Verify a claim against stored evidence.

Searches for evidence that supports or refutes the claim. Returns verified/partial/ungrounded status with confidence.

Rule: NEVER throw on missing evidence

Return GroundingStatus::Ungrounded with confidence: 0.0 instead.

fn evidence( &self, query: &str, max_results: usize, ) -> SisterResult<Vec<EvidenceDetail>>

Get detailed evidence for a query.

Returns matching evidence items with full content and metadata. max_results limits the number of items returned.

fn suggest( &self, query: &str, limit: usize, ) -> SisterResult<Vec<GroundingSuggestion>>

Find similar items when an exact match fails.

Returns suggestions that are close to the query, helping the LLM recover from ungrounded claims.

Implementors