Skip to main content

Archetype: Ranking recommendation engine

Use when​

Users must prioritize entities (jobs, securities, incidents) with interchangeable scoring knobs.

Common users​

  • Talent teams
  • Merchandisers
  • Security triage desks

Minimal MVP​

  • Feature store or tabular ingestion
  • Scoring pipeline with versioning
  • Explain-this-rank UX
  • Human override + feedback capture

Required files​

  • product.yaml
  • product-brief.md
  • data-contract.md
  • evaluation.md
  • demo.md

Evaluation​

  • NDCG / precision@k when labels exist
  • Fairness slices when regulated
  • Stability when features refresh

Reference products​

  • OddsFox-style ranking surfaces (see showcase)

Common mistakes​

  • Black-box scores without counterfactuals
  • Ignoring cold-start coverage
  • Optimizing offline metrics that ignore business guardrails

Agent prompt​

You are building a ranking microproduct. First define the user task, label availability, and failure costs for false positives vs false negatives. Only then implement scoring, explanations, and feedback capture.