Graphorin API reference v0.15.1

---

Graphorin API reference / @graphorin/memory / / WeightedRRFReranker

Class: WeightedRRFReranker

Defined in: packages/memory/src/search/rrf.ts:73

Stable

Weighted-RRF reranker. Fuses parallel ranked lists through fuseWeighted, applying a per-list weights[i] so a caller who has calibrated list reliability against labels (via @graphorin/evals) can trust one retriever over another - e.g. up-weight dense vector hits over lexical FTS hits. weights is positional: aligned to the input lists order (the built-in hybrid search passes FTS first, then vector). At equal weights it is identical to RRFReranker; RRF stays the framework default.

Like the RRF reranker it is deterministic and tie-broken by stable record id, and it preserves the upstream signals on each hit.

Implements

• ReRanker

Constructors

Constructor

new WeightedRRFReranker(args): WeightedRRFReranker;

Defined in: packages/memory/src/search/rrf.ts:78

Parameters

Parameter	Type
args	{ k?: number; weights: readonly number[]; }
args.k?	number
args.weights	readonly number[]

Returns

WeightedRRFReranker

Properties

Property	Modifier	Type	Description	Defined in
id	readonly	"weighted-rrf"	Stable lowercase identifier surfaced on every span.	packages/memory/src/search/rrf.ts:74
k	readonly	number	-	packages/memory/src/search/rrf.ts:75
weights	readonly	readonly number[]	-	packages/memory/src/search/rrf.ts:76

Methods

rerank()

rerank<TRecord>(
   query, 
   lists, 
options?): Promise<readonly MemoryHit<TRecord>[]>;

Defined in: packages/memory/src/search/rrf.ts:96

Rerank one or more parallel ranked lists and return the fused top-K (default topK = 10). Each input list must already be sorted by score descending.

Type Parameters

Type Parameter
TRecord extends MemoryRecord

Parameters

Parameter	Type
query	string
lists	readonly readonly MemoryHit<TRecord>[][]
options	ReRankOptions

Returns

Promise<readonly MemoryHit&lt;TRecord&gt;[]>

Implementation of

ReRanker.rerank