Skip to content

Augment golden eval set with synthetic query variations #19

Description

@sid-ak

Category: Operational & UX
Priority: Low
Phase: Phase 3 (Embeddings + dense retrieval) — eval harness

Problem
The golden-set eval only reflects one person's question style over one corpus — systematic blind spots in query types won't surface.

Solution
Augment the golden set with synthetic paraphrases, keyword-only, and multi-hop query variations generated once from the corpus.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions