Feature Request: Explainable MMR Score in k-NN

Summary:
For the Multi-Maximal Relevance (MMR) feature in version 3.3, add detailed explainability by including a debug payload for each hit.

Proposed Enhancement

• Introduce a debug field (e.g., mmr_debug) per result hit.
• The debug payload should include:
  • original_score: The initial similarity score for the hit.
  • computed mmr_score: The recalculated score after applying MMR.
  • max_similarity_to_selected: The maximum similarity to any already selected hit.

This improvement will make debugging, understanding, and tuning MMR much easier for developers and users by exposing inner calculations.

Version Target

• OpenSearch k-NN 3.3

Example

{
  "mmr_debug": {
    "original_score": 0.85,
    "mmr_score": 0.61,
    "max_similarity_to_selected": 0.20
  }
}

Benefits

• Improved transparency of MMR ranking
• Easier debugging and tuning for search quality
• Facilitates downstream analysis and validation