Evidence: harness probe 2026-07-27 (report at /opt/remember-stack/.benchmark-runs/v4-glm2/harness-probe-grok-45.md). Grok-4.5 as API consumer listed this among its top ergonomics gaps: after resolve_entity gives you Caroline's entity_id, there is no recipe or parameter to retrieve that entity's claims — you fall back to text search (claims_verbatim / claims_hybrid_rrf) and hope the name is in the claim text, or take the entity_timeline detour (currently misleading per #153).
An entity_id filter parameter on the claims search recipes (or a dedicated claims_for_entity recipe joining via MENTIONED_IN) would close the loop resolve_entity → scoped claims that every entity-centric question wants.
Note: adding a parameter changes recipe descriptors ⇒ tool-catalog hash ⇒ requires a protocol version bump per the design rule; version the recipes accordingly.
Evidence: harness probe 2026-07-27 (report at
/opt/remember-stack/.benchmark-runs/v4-glm2/harness-probe-grok-45.md). Grok-4.5 as API consumer listed this among its top ergonomics gaps: afterresolve_entitygives you Caroline's entity_id, there is no recipe or parameter to retrieve that entity's claims — you fall back to text search (claims_verbatim/claims_hybrid_rrf) and hope the name is in the claim text, or take theentity_timelinedetour (currently misleading per #153).An
entity_idfilter parameter on the claims search recipes (or a dedicatedclaims_for_entityrecipe joining via MENTIONED_IN) would close the loop resolve_entity → scoped claims that every entity-centric question wants.Note: adding a parameter changes recipe descriptors ⇒ tool-catalog hash ⇒ requires a protocol version bump per the design rule; version the recipes accordingly.