Module query 

Query types and search execution

Re-exports

pub use candidate_scoring::CandidateFeature;

pub use candidate_scoring::CandidateQuery;

pub use candidate_scoring::CandidateScores;

pub use candidate_scoring::CandidateScoringPlan;

pub use candidate_scoring::PassageFeatures;

pub use candidate_scoring::RankingModel;

pub use candidate_scoring::ScoreScope;

pub use candidate_scoring::ScoredCandidate;

pub use docset::*;

Modules

candidate_scoring

Named scores over a bounded candidate union: preserve organic retrieval values, optionally backfill missing cells, and apply the compiled formula.

docset

DocSet trait and concrete implementations for document iteration.

Structs

AllQuery

Matches every document, including documents with missing fields.

BinaryDenseVectorQuery

Binary dense vector query for Hamming distance similarity search

Bm25Params

Per-field BM25 parameters (indexed<k1: ..., b: ...> in the schema).

BooleanQuery

Boolean query with MUST, SHOULD, and MUST_NOT clauses

BoostQuery

Boost query - multiplies the score of the inner query

CountCollector

Collector that counts all matching documents

DenseVectorQuery

Dense vector query for similarity search

DocAddress

Unique document address: segment_id + local doc_id within segment. Stores segment_id as u128 internally (16 bytes) but serializes as hex string for backward compatibility with JSON/gRPC clients.

DocBitset

Compact bitset indexed by doc_id. O(1) lookup, ~2.25 MB for 18M docs.

EmptyScorer

Empty scorer for terms that don’t exist

FilteredQuery

GlobalStats

Global statistics aggregated across all segments (legacy)

GlobalStatsBuilder

Builder for aggregating statistics from multiple segments

GlobalStatsCache

Cached global statistics with automatic invalidation

HeapEntry

Entry for top-k min-heap

LazyGlobalStats

Lazy global statistics bound to a fixed set of segments

MatchedField

Matched field info with ordinals (for multi-valued fields)

MaxScoreExecutor

Unified Block-Max MaxScore executor for top-k retrieval

PhraseQuery

Phrase query - matches documents containing terms in consecutive positions

PrefixQuery

Prefix query — matches documents containing any term starting with prefix.

ProximityConfig

Proximity rescoring of a text query (MatchQuery.proximity_weight).

RangeQuery

Fast-field range query.

RegexQuery

Constant-score union of indexed terms matching a whole regular expression.

RerankerConfig

Configuration for L2 dense/binary vector reranking

RrfContribution

One vote from a branch’s complete nomination list, before backfill.

RrfRankedList

A nomination branch. None scope preserves legacy chunk fusion semantics.

RrfScore

ScoreCollector

Efficient top-k collector using min-heap (internal, scoring-layer)

ScoredDoc

Search result from MaxScore execution

ScoredPosition

A scored position/ordinal within a field For text fields: position is the token position For vector fields: position is the ordinal (which vector in multi-value)

ScorerOptions

Options that affect scorer construction rather than scoring semantics.

SearchHit

Search hit with unique document address and score

SearchResponse

Search response with hits (IDs only, no documents)

SearchResult

Search result with doc_id and score (internal use)

SharedThreshold

Cross-segment top-k score floor, shared across the parallel/concurrent per-segment searches of a single query.

SparseFieldStats

Statistics for a sparse vector field

SparseTermQuery

Query for a single sparse vector dimension.

SparseTermQueryInfo

Info for MaxScore-optimizable sparse term queries

SparseVectorQuery

Sparse vector query for similarity search

TermQuery

Term query - matches documents containing a specific term

TermQueryInfo

Info for MaxScore-optimizable term queries

TextFieldStats

Statistics for a full-text field

TopKCollector

Collector for top-k results

WildcardQuery

Constant-score union of indexed terms matching a whole-term wildcard.

Enums

FusionMethod

Method for fusing multiple ranked result lists.

LengthSource

Where a text cursor reads the length of a scoring unit: chunk lengths of a chunked field, or the persisted per-document field lengths (norms) of a plain field. Without either, tf stands in for the length.

MultiValueCombiner

Strategy for combining scores when a document has multiple values for the same field

QueryDecomposition

Decomposition of a query for MaxScore optimization.

RangeBound

Inclusive range bounds in the user’s type domain.

Constants

BM25_B

BM25 b parameter - controls length normalization 0 = no length normalization, 1 = full normalization

BM25_K1

BM25 k1 parameter - controls term frequency saturation Higher values give more weight to term frequency

DEFAULT_DENSE_RERANK_FACTOR

Default exact-rerank candidate multiplier for dense search.

DEFAULT_RRF_K

Default RRF rank constant (from Cormack et al., the standard choice).

MAX_CANDIDATE_OVERSUBSCRIPTION

Maximum candidate depth relative to the result window.

MAX_DENSE_NPROBE

Maximum number of IVF clusters a single dense query may probe.

MAX_DENSE_RERANK_FACTOR

Maximum exact-rerank candidate multiplier accepted by dense search.

MAX_FUSION_CANDIDATE_SLOTS

Maximum aggregate list slots retained before fusion.

MAX_FUSION_CHUNK_SLOTS

Maximum per-ordinal chunk contributions materialized during fusion.

MAX_FUSION_SUB_QUERIES

Maximum independently executed lists accepted by the Searcher fusion API.

MAX_QUERY_TERMS

Maximum number of query tokens (terms / dimensions) for text and sparse queries. Queries exceeding this limit are trimmed to the top-weighted terms.

Traits

Collector

Trait for search result collectors

Query

A search query (async)

Scorer

Scored document stream: a DocSet that also provides scores.

Functions

bm25_idf

Compute IDF (Inverse Document Frequency) using BM25 variant

bm25_score

Compute BM25 score for a term occurrence

bm25_upper_bound

Compute BM25 upper bound score for MaxScore pruning

bm25_upper_bound_with_len

BM25 upper bound with a known minimum length of the scoring units the bound covers (a block or a whole list): the shortest unit has the weakest length normalisation, so it bounds every longer one.

bm25f_score

Compute BM25F score with field boost

bm25f_upper_bound

Compute BM25F upper bound score for MaxScore pruning with field boost

collect_segment

Execute a query with one or more collectors (async)

collect_segment_with_limit

Execute a query with one or more collectors and a specific limit (async)

collect_segment_with_limit_seeded

Async collect_segment_with_limit with a cross-segment threshold seed.

collect_segment_with_limit_seeded_sync

Synchronous collect_segment_with_limit_sync with a cross-segment threshold seed (see collect_segment_with_limit_seeded).

collect_segment_with_limit_sync

Synchronous collect with limit — uses scorer_sync.

fuse_ranked_lists

Fuse multiple ranked result lists into a single top-limit list.

fuse_ranked_lists_chunked

Fuse multiple ranked result lists at chunk granularity.

max_candidate_limit

Largest default candidate pool for a requested result window.

rerank

Rerank L1 candidates by exact dense vector distance.

rrf_scores_for_hits

Compute RRF only for the selected hits, using ranks from complete lists. Input/output hit order is preserved. No retrieval, hydration or L1 scoring occurs here. Document context is broadcast to nominated passage scores.

search_segment_seeded

Per-segment search seeded with a cross-segment top-k floor (async).

search_segment_seeded_sync

Per-segment search seeded with a cross-segment top-k floor (sync).

search_segment_shared

Per-segment search with a live cross-segment top-k floor (async).

search_segment_shared_sync

Per-segment search with a live cross-segment top-k floor (sync).

search_segment_with_count

Execute a search query on a single segment and return (results, total_seen) (async)

search_segment_with_count_sync

Synchronous segment search — returns (results, total_seen).

search_segment_with_positions_and_count

Execute a search query on a single segment with positions and return (results, total_seen)

search_segment_with_positions_and_count_sync

Synchronous segment search with positions — returns (results, total_seen).

try_fuse_ranked_lists_chunked

Validated, bounded entry point for chunk-level fusion used by Searcher and the server. The legacy pure helper remains available for trusted embedded callers, while request-facing paths must account for ordinal expansion before allocating fusion maps.

try_fuse_ranked_lists_chunked_borrowed

Fuse borrowed nomination lists so diagnostics can reuse them without cloning documents or positions. Bounds and scoring match the owning entry point.

Type Aliases

CountFuture

Future type for count estimation

DocPredicate

Per-document predicate closure type (platform-aware Send+Sync bounds)

MatchedPositions

Matched positions for a field (field_id, list of scored positions) Each position includes its individual score contribution

ScoreBatch

Exact scores corresponding to one compact posting batch.

ScoreBatchMask

ScorerFuture

Future type for scorer creation