> ## Documentation Index
> Fetch the complete documentation index at: https://docs.negentro.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# UnifiedScorer & Reranking

> Combine dense semantic similarity, sparse BM25 text match, recency decay, and cross-encoders for precision retrieval.

## The 3-Stage Reranking Pipeline

## Tuning Retrieval Parameters

```python rerank_config.py theme={null}
results = m.search(
    query="User preferences for deployment target",
    user_id="usr_9918",
    alpha=0.7,             # 0.7 Vector weight / 0.3 BM25 text weight
    min_score=0.25,         # Rejects low-confidence matches
    recency_decay_days=30  # Half-life for temporal fading
)
```


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.