Field MinRelevanceScore

Namespace

LMKit.Retrieval

Assembly

LM-Kit.NET.dll

The default minimum relevance score used to filter retrieved partitions. Partitions scoring below this threshold are excluded from results.

public const float MinRelevanceScore = 0.5

Returns

float

The default minimum relevance score used to filter retrieved partitions. Partitions scoring below this threshold are excluded from results.

Examples

// Use the default threshold when searching for matching partitions.
var matches = ragEngine.FindMatchingPartitions(
    "semantic search",
    topK: 5,
    minScore: RetrievalDefaults.MinRelevanceScore);

Console.WriteLine($"Found {matches.Count} partitions above {RetrievalDefaults.MinRelevanceScore} relevance.");

Remarks

This is an absolute cosine floor and is the single source of truth for every retrieval surface that gates on relevance. Embedding spaces are anisotropic, so unrelated pairs already score well above zero, and a query and the passage answering it are near-paraphrases rather than duplicates. The value is the highest floor that still keeps essentially every correct match across the supported embedder families: raising it trades away correct matches that continue to rank first, far faster than it removes wrong ones.