API reference

The public surface, generated from the assemblies and the XML documentation they ship. The What it is column is the type’s own summary, not a sentence written for this page: it is what the author wrote above the declaration, trimmed to its first paragraph. Where it is not enough, the Guide column is the page that argues the case.

This page indexes the surface; it does not teach it. The guide is where a pipeline is built, and the API reference answers the narrower question of what exists to build it with.

The contracts

Every interface the library asks you to satisfy, and the ones it asks you to satisfy something with. These are the seams: an engine, a scorer, a merger, a reranker, a model provider. Implementing one is how the library is extended, which is why they are listed first.

Core package

Contract	Extends	What it is	Guide
IContextualTextClassifier	ITextClassifier	A classifier that accepts request-scoped evidence in addition to the text.	guide
IFeatureSet	—	The features of one item, as a question sees them: a set of names, no domain.	guide
IFeatureSource<TItem>	—	Extracts the features of an item — the seam that keeps the question trees domain-agnostic.	guide
IAsyncCrossEncoderScorer	—	Asynchronous counterpart of ICrossEncoderScorer: scores a (query, document) pair without blocking the caller. Use this when the model is remote or otherwise…	guide
IAsyncReranker	—	Asynchronous counterpart of IReranker: re-orders a candidate list without blocking the caller. See IAsyncCrossEncoderScorer for the rationale and the limits of…	guide
IAsyncTextClassifier	ITextClassifier	Optional capability of an ITextClassifier: the same training and prediction operations, with provider calls that can complete asynchronously.	guide
IBatchEmbeddingProvider	IEmbeddingProvider	An IEmbeddingProvider that can embed several texts in one call. Engines that index a corpus check for this interface and use it instead of calling…	guide
ICandidateIndex	IReadOnlyTextIndex	Capability for an index that can enumerate only the documents containing at least one of the given query terms, instead of scanning the whole corpus.	guide
ICandidateTextSource	—	Produces categories that no trained classifier has learned, from the text alone.	guide
IChunkContextEnricher	—	Augments the searchable text of a SearchDocument before it is indexed — the « contextual retrieval » seam. A chunk that only says “the benefit rose 12%” is…	guide
IColumnAwareBatchEmbeddingProvider	IBatchEmbeddingProvider, IEmbeddingProvider	A batch provider that also knows which column the texts belong to — the pair of capabilities a PostgreSQL engine with EmbeddingColumns needs to batch one column…	guide
ICompressingSearchEngine	ITextSearchEngine	Optional capability of an ITextSearchEngine that can return, alongside the ranked page, each document compressed to its query-relevant content — the surface a…	guide
IContextCompressor	—	Reduces a document’s text to the content a query actually cares about — the token-reduction seam between a ranked page and a generation model. LexiSharp never…	guide
IContextualSearchEngine<TPayload>	ITextSearchEngine	An ITextSearchEngine that accepts a caller-supplied payload alongside a query, for the cases a query alone does not carry: who is asking, what time it is, which…	guide
ICrossEncoderScorer	—	Scores a (query, document) pair as a single relevance value — the job of a cross-encoder (concatenate query and document into one input) or, more generally, of…	guide
IDetailedSearchEngine	—	Optional capability of an ITextSearchEngine that can return, alongside the merged ranking, the per-source score breakdown that fed it (see Contributions).	guide
IEmbeddingProvider	—	Contract for producing text embeddings. LexiSharp never computes embeddings itself: implementing this interface is up to the consumer — a local ONNX model, an…	guide
IEmbeddingSignatureProvider	—	An IEmbeddingProvider that can identify the model and preprocessing that produce its vectors. The signature is what lets a persisted index refuse to compare…	guide
IExplainableSearchEngine	—	Optional capability of an ITextSearchEngine that can explain why a document received the score it did for a query, when its scorer is an IScoreExplainer.	guide
IFacetedSearchEngine	ITextSearchEngine	Optional capability of an ITextSearchEngine that can return, alongside the ranked results, facet counts over Fields.	guide
IFieldScopedScorer	ITextScorer	Capability for a scorer that can score a query term from one field’s statistics instead of the whole document’s: the arithmetic behind title:term.	guide
IFieldStatisticsIndex	IReadOnlyTextIndex	Capability for an index that tracks per-field statistics: how many times a term occurs in one field of one document, how long each field is, the corpus averages…	guide
IGraphSearchEngine	ITextSearchEngine	Optional capability of an ITextSearchEngine that can return, alongside the ranked page, the knowledge-graph facts connected to the query — the search-side half…	guide
IIncrementalAsyncTextClassifier	IAsyncTextClassifier, ITextClassifier	Optional capability of an IAsyncTextClassifier: add or retract one labelled document without a retraining pass, asynchronously.	guide
IIncrementalTextClassifier	ITextClassifier	Optional capability of an ITextClassifier: add or retract single labelled documents without a retraining pass over the whole corpus. Detected with pattern…	guide
IKnowledgeGraphBridge	—	Contracts for connecting the engines to a knowledge graph — the GraphRAG seam. LexiSharp never runs a model or a graph itself: relation extraction is the model…	guide
IListableSearchEngine	—	Optional capability of a ITextSearchEngine that can enumerate the identifiers of the documents it currently holds, without materializing the full documents.	guide
IQueryCostEstimator	—	Picks which of several engines a RoutedSearchEngine should run a query on.	guide
IQueryCostProbe	—	Optional capability of an ITextSearchEngine that can estimate, cheaply and without running a search, how many documents a query is likely to touch — the signal…	guide
IQueryRouter	—	Picks which route (engine + optional filters) a RoutingSearchEngine should run a query on — a semantic/intent decision, as opposed to IQueryCostEstimator which…	guide
IQuerySyntaxSupport	—	Declares which QueryFeatures an ITextSearchEngine interprets, so callers can discover the supported query syntax instead of discovering it by getting wrong…	guide
IQueryTransformer	—	Rewrites a raw query into one or more searchable variants before retrieval — the « pre-retrieval transformation » seam. Rewriting, decomposing a question into…	guide
IReadOnlyTextIndex	—	The read-only view of an ITextIndex: the corpus statistics and per-document lookups a scorer, a reranker or an extractor needs, with none of the mutation…	guide
IReinforceableTextClassifier	ITextClassifier	Optional capability of an ITextClassifier: adjust a trained model from user feedback, one text-to-category association at a time, without a retraining pass over…	guide
IReranker	—	Contract for re-ordering a list of candidate results after retrieval — the seam for second-stage ranking strategies (diversity preservation, cross-encoders,…	guide
IRetrievalMetrics	—	Receives the measurements a RetrievalTelemetry takes. Implement it to ship numbers to a metrics backend (OpenTelemetry, Prometheus, StatsD, …).	guide
IScoreExplainer	ITextScorer	Optional capability of an ITextScorer that can justify its scores: it produces a ScoreExplanation decomposing the contribution of every matched query term.	guide
ISparseEmbeddingProvider	—	Contract for producing sparse text embeddings — e.g. SPLADE/SPLADE-v2, uniCOIL, or any learned sparse model. Sparse embeddings map vocabulary terms to weights,…	guide
IStructureAwareSearchEngine	ITextSearchEngine	Optional capability of an ITextSearchEngine that knows the hierarchical structure its documents live in — the « structure-aware retriever » seam. Flat retrieval…	guide
ITermCorrector	—	Corrects a single query term — the query-side seam for spell correction, typo tolerance and keyboard-aware matching. The vocabulary is a construction concern of…	guide
ITermOverlapScorer	ITextScorer	Capability for a scorer whose score is exactly 0 for every document sharing no query term with the query, which lets the search engine skip such documents…	guide
ITextClassifier	—	Supervised text classifier trained on labelled documents.	guide
ITextIndex	IReadOnlyTextIndex	A write-capable index: the read surface of IReadOnlyTextIndex plus the mutations a caller that owns the corpus performs (Index/Add/Remove/Clear).	guide
ITextPriorSource	—	Produces additive log-prior evidence for categories a trained model already holds.	guide
ITextScorer	—	Computes how relevant a single document is with respect to a tokenized query, given a shared IReadOnlyTextIndex for corpus statistics.	guide
ITextSearchEngine	—	The public search surface of the library: index documents and rank them against a query.	guide
ITokenEmbeddingProvider	—	Produces a sequence of token-level embeddings for a text — the building block of late interaction (ColBERT-style) ranking, where a query is matched to a…	guide
IVocabularyIndex	IReadOnlyTextIndex	Capability for an index that can enumerate its vocabulary — the seam prefix/fuzzy query expansion reads at search time.	guide
IWeightedPredictor	—	Optional capability of a ITextClassifier that accepts WeightedTokens instead of raw text, bypassing internal tokenization. Detected with pattern matching…	guide
ITermExpander	—	Derives the additional terms to index for a document from its own tokenization.	guide
IMergeExplainer	—	A merger that can state the arithmetic behind the list it returned, without keeping any state between searches.	guide
IResultMerger	—	Strategy that turns several ranked result lists — one per source engine — into a single coherent, de-duplicated, ordered list.	guide
IKeywordExtractor	—	Strategy for pulling the most representative terms out of a single text — tag generation, query expansion seeds, summary highlights.	guide
ISpanTokenizer	ITokenizer	A tokenizer that can also report where each normalized term sits in the source text — the basis for search-result highlighting.	guide
IStemmer	—	Pluggable stemmer used by the Tokenizer to fold inflected forms onto a common stem. The library ships one implementation, PorterStemmer (English, no…	guide
ITokenizer	—	Converts raw text into a list of normalized terms.	guide

LexiSharp.Postgres package

Contract	Extends	What it is	Guide
IColumnAwareEmbeddingProvider	IEmbeddingProvider	Optional capability of an IEmbeddingProvider used by a multi-column PostgresVectorSearchEngine: the provider is told which embedding column the text is being…	guide

LexiSharp.Onnx package

Contract	Extends	What it is	Guide
IOnnxTokenizer	—	The tokenizer seam an ONNX encoder needs: text in, ids out. LexiSharp does not ship a model vocabulary, so the implementation carries the checkpoint’s own rules…	guide

The types

Everything else that is public, by namespace. record, record struct and class are told apart by the members the compiler adds to each — the difference is in the page because SearchDocument is constructed and InMemoryTextIndex is not, and because a record struct is neither: it is a value type with value equality and a with expression, and calling it a struct would be as wrong as calling a record a class.

Core package

LexiSharp

Type	Kind	Contracts	What it is
LexiSharpFacetedResult<TDocument>	record	—	Ranked hits plus one facet bucket per requested field, as returned by LexiSharpIndex.
LexiSharpHit<TDocument>	record	—	One search hit over a user document, as returned by LexiSharpIndex.
LexiSharpIndex<TDocument>	class	IDisposable	The drop-in entry point of the library: an opinionated, typed search index that maps your own documents to the engine internals and back — from “a list of…
LexiSharpIndexOptions<TDocument>	class	—	Configuration of a LexiSharpIndex: how caller documents map to indexable records, and which engine pieces to use.
LexiSharpQueryOptions	record	—	Per-search options for LexiSharpIndex and friends.

LexiSharp.Benchmarking

Type	Kind	Contracts	What it is
BenchmarkComparer	class	—	Builds a BenchmarkComparison from two runs.
BenchmarkComparison	record	—	A baseline-versus-candidate comparison, per query and in aggregate. This is the artifact that answers “what did this change actually do”, which a pair of means…
BenchmarkConfig	record	—	A named recipe for building the search engine a benchmark run evaluates. The build closure receives the shared index and the query set so a configuration can…
BenchmarkConfigResult	record	—	The outcome of benchmarking one BenchmarkConfig against a corpus and a query set: the retrieval metrics averaged over the judged queries, plus the latency…
BenchmarkDeltaVerdict	enum	—	How a query’s nDCG moved between two configurations.
BenchmarkMetrics	record	—	Mean retrieval metrics of one benchmark configuration over the judged queries, at the configured topK. Every value lives in [0, 1]; all numerators and…
BenchmarkOptions	record	—	Tuning knobs of a Run.
BenchmarkQuery	record	—	One labeled query of a benchmark run: a raw query together with the relevance judgments a good ranking should reproduce.
BenchmarkQueryDelta	record	—	One query’s movement between two benchmark runs, with the evidence for it.
BenchmarkQueryResult	record	—	The outcome of benchmarking one query: its own metrics, the ids the engine returned, and where the first judged document landed.
CorpusBenchmark	class	—	Runs a retrieval benchmark: builds one shared in-memory index over a corpus, evaluates every requested BenchmarkConfig against the same query set, and reports…
PanelEvaluationResult	record	—	The outcome of evaluating a live engine against a RetrievalEvaluator’s panel: the retrieval metrics averaged over the judged queries, plus the wall-clock time…
RetrievalEvaluator	class	—	Evaluates a liveITextSearchEngine against a fixed panel of labeled queries, on demand — the runtime complement to CorpusBenchmark, which builds a fresh index…

LexiSharp.Classification

Type	Kind	Contracts	What it is
BlendContext	record	—	Evidence supplied for one prediction, instead of being configured as a source on the blend.
BlendTextClassifier	class	IContextualTextClassifier, IIncrementalTextClassifier, ITextClassifier	Blends any number of text classifiers, then optionally injects candidate categories and log-prior bonuses.
CalibratedClassifier	class	ITextClassifier	Wraps an ITextClassifier with a fitted temperature, so Predict returns the calibrated distribution and PredictBest the same top category.
CalibrationSample	record	—	One validation observation for calibration: the distribution a model produced and the category that was actually right.
CandidateLane	record	—	One source of categories no model learned, and the weight of its evidence.
CentroidClassifier	class	IAsyncTextClassifier, IIncrementalAsyncTextClassifier, IIncrementalTextClassifier, ITextClassifier	Text wrapper over the generic nearest-centroid model: embeds documents with an IEmbeddingProvider, computes one centroid per Category, and classifies a text by…
CentroidClassifier<TItem>	class	—	Nearest-centroid classifier over caller-supplied vectors: one mean per category, ranked by cosine similarity, with a softmax over the cosines as the posterior.
CentroidClassifierOptions	record	—	Tunable knobs for the nearest-centroid classifiers: the softmax temperature, and the embedding role the text wrapper uses on each side.
ClassificationInterval	record struct	—	A percentile confidence interval.
ClassificationModelSerializer	class	—	Saves and loads trained text classifiers in a versioned, dependency-free binary format.
CompositeFeatureSource<TItem>	class	IFeatureSource	Combines several feature sources by union: an item carries every feature any source reports.
CorpusCluster	record	—	One terminal region of a corpus division: its document ids, the features that best represent it, and the depth at which it was found.
CorpusDivider	class	—	The text convenience over CorpusDivider: divides a SearchDocument corpus by the terms of Text.
CorpusDivider<TItem>	class	—	Divides an unlabelled corpus over any feature space into two recursively: at each node it asks the feature whose presence splits the items into the two most…
CorpusDividerOptions	record	—	Tunable knobs for CorpusDivider; the defaults keep the tree shallow enough to read and bound the vocabulary the split search compares.
CorpusDivision	class	—	The tree and terminal clusters produced by CorpusDivider.
DecisionBranch	record	DecisionTreeNode	An internal node: a question, the two subtrees it selects between, and the score the split earned when the tree was built.
DecisionExplanation	record	—	The path an item took through a tree, and the leaf it reached.
DecisionLeaf	record	DecisionTreeNode	A leaf: the category the tree predicts and the smoothed class distribution behind it.
DecisionSplitMetrics	record	—	The measurements behind one chosen split.
DecisionTreeClassifier	class	ITextClassifier	The text convenience over DecisionTreeClassifier: a SearchDocument classifier that reads Text and implements ITextClassifier.
DecisionTreeClassifier<TItem>	class	—	A supervised classifier over any feature space: at each node it asks the feature whose presence best separates the training labels, so the most discriminative…
DecisionTreeNode	record	—	A node of a trained decision tree: either a question or a leaf.
DecisionTreeOptions	record	—	Tunable knobs for DecisionTreeClassifier; the defaults stop a tree before it memorizes the training corpus.
DelegateFeatureSource<TItem>	class	IFeatureSource	A feature source built from a delegate, for the one-off projection that does not deserve a class.
DivisionBranch	record	DivisionNode	An internal node of a corpus division: the question that split this region of the corpus and the two regions it produced.
DivisionLeaf	record	DivisionNode	A terminal region of the corpus.
DivisionNode	record	—	A node of a corpus division: either a question or a cluster.
DivisionSplitMetrics	record	—	The measurements behind one chosen corpus split.
FeatureKind	enum	—	What a feature’s value means: a binary presence, a term count, or a numeric measurement.
FeaturePresenceQuestion	record	FeatureQuestion	True when the item carries a given feature — the only question the built-in builders ask.
FeatureQuestion	record	—	A yes/no question asked of an item’s features — the unit a question tree is built from.
FeatureSet	class	IFeatureSet	The default IFeatureSet: a set of names with ordinal comparison, optionally with a numeric value attached to some of them.
FeatureState	enum	—	What a feature set knows about one feature: it is present, it is known to be absent, or the source could not determine it. A binary question answers no for both…
FusedTextClassifier	class	IIncrementalTextClassifier, ITextClassifier	Blends two text classifiers by probability: (1 - w) · P(first) + w · P(second).
IdfMode	enum	—	How term evidence is scaled by corpus rarity during classification. Term frequency is always counted from the training corpus; this knob decides whether (and…
LinearClassifier	class	ITextClassifier	The text convenience over LinearClassifier: a SearchDocument model over Text.
LinearClassifier<TItem>	class	—	A multinomial logistic regression over the same IFeatureSource seam as the question trees, trained with AdaGrad and a deterministic shuffle.
LinearClassifierOptions	record	—	Tunable knobs for LinearClassifier: the AdaGrad pass, the vocabulary floor, and the abstention threshold.
LinearFeatureWeight	record	—	One feature and the weight the linear model gives it for a category.
MaskedFeatureSource<TItem>	class	IFeatureSource	Suppresses an inner feature source for the items a mask predicate selects, and reports the mask’s own features instead.
NaiveBayesClassifier	class	IIncrementalTextClassifier, IReinforceableTextClassifier, ITextClassifier, IWeightedPredictor	Multinomial Naive Bayes trained on labelled documents.
NaiveBayesOptions	record	—	Tunable knobs for NaiveBayesClassifier; the defaults reproduce the classic Laplace-smoothed, unweighted multinomial Naive Bayes.
NearestNeighborClassifier	class	IAsyncTextClassifier, IIncrementalAsyncTextClassifier, IIncrementalTextClassifier, ITextClassifier	K-nearest-neighbour text classifier: BM25 over terms, or cosine over embeddings, with a configurable vote.
NearestNeighborOptions	record	—	Tunable knobs for NearestNeighborClassifier: how many neighbours vote, how their scores become weights, the BM25 parameters of the lexical path, and the…
NeighborVote	enum	—	How a neighbour’s similarity becomes a vote.
NumericThresholdQuestion	record	FeatureQuestion	True when a feature’s numeric value is at or above the threshold. A missing value — the feature is absent, carries no number, or carries NaN — answers no, which…
OnlineBucketMetrics	record	—	Accuracy and macro-F1 for one slice of the stream.
OnlineClassificationEvaluator	class	—	Replays a labelled stream through any ITextClassifier: predict first, then learn.
OnlineCoveragePoint	record	—	One point of the precision/coverage curve.
OnlineEvaluationExample	record	—	One labelled document in the stream, with the block (day, period, …) it belongs to and an optional request context for a contextual classifier.
OnlineEvaluationOptions	record	—	Options for Evaluate.
OnlineReport	record	—	The outcome of one predict-then-learn run.
PriorLane	record	—	One source of log-prior bonuses on categories a model already holds.
QuestionAnswer	record	—	One question and the answer a particular item gave it.
RandomForestClassifier	class	ITextClassifier	The text convenience over RandomForestClassifier: a SearchDocument forest over Text.
RandomForestClassifier<TItem>	class	—	A bagged forest of DecisionTreeClassifier trees: each tree sees a bootstrap sample and a random subset of the features, and the forest averages their…
RandomForestOptions	record	—	Tunable knobs for RandomForestClassifier: how many trees, how deep, and how much of the feature space each tree is allowed to see.
SplitCriterion	enum	—	How a candidate split is scored when a decision tree chooses its next question.
SynchronizedTextClassifier	class	IAsyncTextClassifier, IContextualTextClassifier, IDisposable, IIncrementalAsyncTextClassifier, IIncrementalTextClassifier,…	Wraps an ITextClassifier so that concurrent prediction and concurrent mutation stop being the caller’s problem: any number of Predict and PredictBest calls may…
TemperatureCalibrator	class	—	A one-parameter post-hoc calibrator: it divides the log-probabilities by a temperature fitted on a validation set to minimize log-loss.
TextClassifierLane	record	—	One trained classifier and the weight its distribution carries in the blend.
TextFeatureSource	class	IFeatureSource	The text feature space: the normalized terms of Text.
UnknownIndicatorSource<TItem>	class	IFeatureSource	Turns every unknown feature into an explicit binary indicator named prefix + feature, so a model that only asks binary questions can learn the difference…

LexiSharp.Compression

Type	Kind	Contracts	What it is
CompressingTextSearchEngine	class	ICompressingSearchEngine, ITextSearchEngine	A decorator that adds the ICompressingSearchEngine surface to any engine: searching and writing are forwarded unchanged, and SearchCompressed returns each hit’s…
QueryWindowCompressor	class	IContextCompressor	The model-free IContextCompressor: keeps the windows of words around each occurrence of a query term and drops everything else — the deterministic, offline…

LexiSharp.Core

Type	Kind	Contracts	What it is
AtomicClassifierReference	class	—	The atomic publish point for the « snapshot swap » concurrency pattern on a classifier: one thread trains, clones or otherwise builds a fresh model off-lock,…
AtomicEngineReference	class	—	The atomic publish point for the « snapshot swap » concurrency pattern: one thread builds a fresh engine off-lock, then Swap makes it visible to every…
BoostContext<TPayload>	class	—	What a boost function is told about the search it is adjusting: the query, the ranking the inner engine produced before any boost, the caller’s payload, and the…
BoostedTextSearchEngine	class	IQueryCostProbe, ITextSearchEngine	A decorator engine that boosts or damps the matches of an inner engine after ranking, from a function of the candidate alone.
BoostedTextSearchEngine<TPayload>	class	IContextualSearchEngine, IQueryCostProbe, ITextSearchEngine	A decorator engine that boosts or damps the matches of an inner engine after ranking, with a boost that sees the search it is adjusting.
CheapestByCandidateCountEstimator	class	IQueryCostEstimator	Picks the engine with the smallest EstimateCandidateCount. Engines that do not implement IQueryCostProbe cannot be costed and are only used when no probed…
ClassificationResult	record	—	A category predicted for a piece of text, with its estimated probability.
CompressedHit	record	—	One compressed hit of SearchCompressed: the ranked document’s id and score, its text reduced to the query-relevant content, and how much of the source that…
ContextualBoost<TPayload>	delegate	—	A boost that depends on the search as a whole, not only on the candidate in hand: it sees the query, the pre-boost ranking, and the caller’s payload.
DamerauLevenshteinTermCorrector	class	ITermCorrector	A corrector based on Damerau-Levenshtein distance (insertion, deletion, substitution and adjacent transposition), walked over the vocabulary. Distance 1 and 2…
DetailedSearchResult	record	—	A ranked document whose final score is broken down by the source signal that produced it.
DocumentHierarchy	class	—	A validated parent/child forest over searchable document ids — the « hierarchical metadata » seam. Declare the tree your chunks belong to (a chunk under a…
EmbeddingProviderExtensions	class	—	Backwards-compatible call shape: embeds text as a Passage.
EmbeddingUse	enum	—	The role a piece of text plays when it is embedded. Several models encode queries and documents differently by construction (e.g. E5 prefixes queries with…
EngineRetrievalMetrics	record	—	Aggregated measurements for one engine.
EntityTriplet	record	—	One fact of a knowledge graph: a subject and an object linked by a predicate — the subject-verb-object shape relation extraction produces (« acme » → « supplies…
FacetBucket	record	—	Value counts for one facet field.
FacetValue	record	—	A distinct field value and how many counted documents carry it.
FacetedSearchResult	record	—	Ranked results plus one bucket per requested facet field.
FieldScopedPhrase	record	—	A double-quoted phrase restricted to a single field: title:”ssh tunnel” asks for the phrase at consecutive positions inside that field.
FieldScopedTerm	record	—	One query term restricted to a single field: title:ssh is new FieldScopedTerm(“title”, “ssh”).
FieldType	enum	—	The shape of a document field, used by MetadataFilter to interpret the stored string value when no explicit parsing is requested.
GraphAwareSearchEngine	class	IGraphSearchEngine, ITextSearchEngine	A decorator that layers a IKnowledgeGraphBridge over an existing engine: searching and writing are forwarded unchanged, and the IGraphSearchEngine surface adds…
GraphContext	record	—	The answer a IKnowledgeGraphBridge returns for a sub-graph query around a set of entities: the local facts that touch them, plus the community summaries the…
GraphHydratedResults	record	—	The outcome of SearchWithGraphFacts: the ordinary ranked page plus the facts the IKnowledgeGraphBridge returned for the query’s entities.
HierarchicalTextSearchEngine	class	IStructureAwareSearchEngine, ITextSearchEngine	A decorator that layers a DocumentHierarchy over an existing engine: searches and writes are forwarded unchanged — flat retrieval still finds the precise node —…
InMemoryRetrievalMetrics	class	IRetrievalMetrics	The default IRetrievalMetrics: in-memory counters grouped by engine and stage, with no external dependency. Enough to assert on in tests and to back a…
IndexRetrievalMetrics	record	—	The latest observed size of one engine’s index.
KeyboardAwareTermCorrector	class	ITermCorrector	A spell corrector that weights substitutions by keyboard proximity: swapping two adjacent keys costs less than swapping two distant ones, and transpositions of…
KeyboardLayout	enum	—	Keyboard layouts supported by KeyboardAwareTermCorrector.
KeyboardLayouts	class	—	Key positions for the supported layouts.
MetadataFilter	record	—	A declarative, single-field predicate over Fields — the raw material of structured filtering (e.g. category = article, year > 2023, tags containing nlp).
MetadataFilterOperator	enum	—	Comparison applied by a MetadataFilter.
MoreLikeThis	class	—	Builds a query from one already-indexed document and re-runs it through an engine, so the caller gets the documents most like it — the classic “more like this”…
NumericFacetBucket	record	—	The numeric facet counts for one requested field: its ranges, in requested order, plus the values that could not be read as numbers.
NumericFacetRange	record	—	One numeric bucket: how many counted documents fall between From and To — in [From, To), or in [From, To] when UpperInclusive is set.
NumericFacetRequest	record	—	One numeric facet over Fields: either explicit boundaries, or a bucket count whose boundaries are derived from the match set — a histogram.
ParsedQuery	record	—	The structured form of a raw query: free terms, quoted phrase constraints, expansion atoms, and the combined literal term list a scorer starts from.
PassthroughTermCorrector	class	ITermCorrector	A corrector that returns the term unchanged — the safe default that never alters a query.
QueryExpansion	record	—	A free-text query atom carrying a search-time expansion operator: the base term (already tokenized to exactly one term) is matched against the index vocabulary…
QueryExpansionKind	enum	—	How a query atom expands against the index vocabulary.
QueryFeatures	enum	—	The optional LexiSharp query-syntax features a raw query can carry. Engines declare which ones they interpret; using an unsupported one is a hard error rather…
QueryParser	class	—	Splits a raw query on double quotes before tokenization — the tokenizer treats “ as an ordinary separator and would destroy the delimiters — and recognizes the…
QueryRoute	record	—	The decision an IQueryRouter returns for a query: which route to run, and how confident the router is.
QuerySyntax	class	—	Detects and validates the LexiSharp query syntax a raw query carries, so an engine that does not interpret a feature fails fast instead of silently treating the…
QueryTermOperator	enum	—	What a query atom asks of the documents that contain it.
RawQuerySegments	record	—	The raw, un-tokenized quote-level split of a query: text outside quotes and the literal interior of each quoted segment. SQL backends forward these strings to…
RerankedTextSearchEngine	class	IQueryCostProbe, ITextSearchEngine	A decorator engine that re-ranks the matches of an inner engine before they are returned.
RetrievalAgreement	enum	—	How much the sources of a federated search agree about one document.
RetrievalAgreementAnalyzer	class	—	Classifies each document of a federated page by how much its sources agree, from the per-source scores in Contributions.
RetrievalAgreementReport	record	—	One document’s agreement across the sources that produced the page, with the per-source strengths it was derived from.
RetrievalLog	delegate	—	Sink for the events a RetrievalTelemetry produces.
RetrievalLogEvent	record struct	—	One observability event emitted by a search engine. Immutable and allocation-free to build, so a sink can consume it without the engine paying more than the…
RetrievalLogEvents	class	—	Stable event names carried by Event.
RetrievalLogLevel	enum	—	Severity of a RetrievalLogEvent.
RetrievalMetricsSnapshot	record	—	A copy of every counter an InMemoryRetrievalMetrics holds.
RetrievalTelemetry	class	—	The opt-in bundle an engine holds to report what it did: a RetrievalLog sink, an IRetrievalMetrics collector, or both. Pass it to an engine’s telemetry…
RoutedEngine	record	—	A named engine candidate for a RoutedSearchEngine.
RoutedSearchEngine	class	IDetailedSearchEngine, IExplainableSearchEngine, IFacetedSearchEngine, IQueryCostProbe, ITextSearchEngine	Opt-in decorator that forwards each query to one of several pre-filled engines, chosen by an IQueryCostEstimator — with CheapestByCandidateCountEstimator the…
RoutingSearchEngine	class	ITextSearchEngine	Opt-in decorator that asks an IQueryRouter which SearchRoute to run each query on, then runs that route’s engine with the route’s filters merged into the…
ScoreBoost	record struct	—	A signed score adjustment: a multiplicative factor plus an additive offset. Both can go up or down, so positive and negative boosts are expressed the same way.
ScoreExplanation	record	—	A transparent breakdown of why a document received its score for a query: global corpus figures, per-term contributions and the scorer’s parameter values.
SearchCostStage	record struct	—	One pipeline stage’s share of what a single query cost: what it processed, and how long it took.
SearchCosts	class	—	The per-query cost sheet a caller opts into: what the query was made of, and what each pipeline stage it went through processed and cost.
SearchDocument	record	—	A single unit of text that can be indexed and searched.
SearchOptions	record	—	Controls how a search is performed and how results are returned.
SearchResult	record struct	—	A document ranked against a query.
SearchRoute	record	—	A named route a RoutingSearchEngine can run: an engine plus optional metadata filters AND-ed onto the caller’s own filters when this route is selected.
SearchTrace	class	—	Collects a TraceStep per document per stage, so a ranking can be explained end to end: which engine scored a document, what each merger contributed, what the…
SparseEmbeddingProviderExtensions	class	—	Backwards-compatible call shape: embeds text as a Passage.
StageRetrievalMetrics	record	—	Aggregated measurements for one pipeline stage of one engine.
SyncCrossEncoderScorerAdapter	class	IAsyncCrossEncoderScorer	Adapts a synchronous ICrossEncoderScorer to IAsyncCrossEncoderScorer by running it on the calling thread inside a completed task. Useful for reusing an…
SyncRerankerAdapter	class	IAsyncReranker	Adapts a synchronous IReranker to IAsyncReranker by running it on the calling thread inside a completed task.
SynchronizedTextSearchEngine	class	IDetailedSearchEngine, IDisposable, IExplainableSearchEngine, IFacetedSearchEngine, ITextSearchEngine	Wraps an ITextSearchEngine so that concurrent searches and concurrent mutation stop being the caller’s problem: any number of Search (and the capability calls…
TermContribution	record	—	The contribution of a single query term to a document’s score, as broken down by an IScoreExplainer.
TextEvidence	record	—	A category with an unnormalized weight, produced by a source that is not a trained classifier.
TextFields	class	—	Field naming rules shared by the index and the field-aware scorers.
TextIndexStatistics	record	—	A read-only snapshot of the corpus statistics of an ITextIndex — useful for observability, dashboards and corpus introspection.
TokenEmbeddingProviderExtensions	class	—	Backwards-compatible call shape: embeds text as a Passage.
TraceStage	enum	—	The pipeline stage a TraceStep was recorded at.
TraceStep	record struct	—	One recorded step of a document’s journey through the search pipeline: the stage it was at, the score going in, the score coming out, and a stage-specific…
WeightedToken	record	—	A token paired with a fractional relevance weight, for classifiers that can consume pre-weightened tokens (e.g. a token corrected by a spell-corrector should…

LexiSharp.Diagnostics

Type	Kind	Contracts	What it is
RetrievalDiagnostics	class	IDisposable, IRetrievalMetrics	Turns the IRetrievalMetrics seam into Meter instruments and ActivitySource spans, using only types that ship with .NET.

LexiSharp.Embeddings

Type	Kind	Contracts	What it is
HashingEmbeddingProvider	class	IEmbeddingProvider, ITokenEmbeddingProvider	A deterministic, dependency-free embedding provider built on the hashing trick (feature hashing): every token is hashed to one dimension of a fixed-size vector…

LexiSharp.Expansion

Type	Kind	Contracts	What it is
CorrectingTextSearchEngine	class	IQueryCostProbe, ITextSearchEngine	A decorator engine that corrects query terms against the index vocabulary before searching — the consuming side of ITermCorrector, proving the seam is usable…
ExpandedTerm	record struct	—	A single additional term injected into the index for a document, alongside its source text tokenization. The term is indexed as a regular inverted-list member…
ExpandingTextSearchEngine	class	IDetailedSearchEngine, IExplainableSearchEngine, IFacetedSearchEngine, IQueryCostProbe, ITextSearchEngine	A decorator engine that widens the query with related terms before delegating to the inner engine — the query-side counterpart of ExpansionTextIndex, which…
ExpansionQueryTransformer	class	IQueryTransformer	A model-free IQueryTransformer that returns two variants — the caller’s query and the same query widened with the terms an ITermExpander derives — the…
ExpansionRanking	enum	—	Which statistic orders an input term’s candidate neighbours before the expansion budget applies.
FuzzyTextSearchEngine	class	IQueryCostProbe, ITextSearchEngine	A decorator engine that widens the query with vocabulary terms within a small edit distance of each query term — in-memory fuzzy matching, the counterpart of…
PmiTermExpander	class	ITermExpander	A corpus-derived ITermExpander: it learns which terms habitually appear in the same context window and expands a document with the top positively associated…
PmiTermExpanderOptions	class	—	Configuration of a PmiTermExpander: how the corpus is analyzed and how many expansion terms each document gets.
ReformulatingTextSearchEngine	class	IQueryCostProbe, ITextSearchEngine	A decorator engine that generates query variants from a term corrector’s uncertainty and fuses their results, instead of committing to a single correction or…
TransformingTextSearchEngine	class	IQueryCostProbe, ITextSearchEngine	A decorator engine that runs an IQueryTransformer on the raw query, searches the inner engine for every variant, and fuses the rankings — the query-side…

LexiSharp.Highlighting

Type	Kind	Contracts	What it is
HighlightOptions	record	—	Knobs for TextHighlighter.
HighlightSnippet	record	—	One highlighted excerpt: the source window [Start, Start + Length) with match tags already injected into Text.
TextHighlighter	class	—	Wraps query matches in the original text: full-string marking for short fields (titles, labels) or padded, word-snapped snippets for long documents.

LexiSharp.Hybrid

Type	Kind	Contracts	What it is
AsyncCrossEncoderReranker	class	IAsyncReranker	Asynchronous counterpart of CrossEncoderReranker: re-scores every candidate with a pairwise (cross-encoder) model via IAsyncCrossEncoderScorer and reorders the…
BayesianFusionMerger	class	IResultMerger	Merges per-engine results by calibrating each engine’s native score to a probability of relevance and combining them with the naive Bayes rule, instead of…
CascadeRerankOptions	record	—	Controls how a CascadeRerankPipeline passes candidates from stage to stage.
CascadeRerankPipeline	class	IReranker	Chains rerankers in cascade: each stage re-ranks the output of the previous one, and the shortlist may be trimmed between stages so only the strongest…
CombMNZResultMerger	class	IResultMerger	Blend strategy that multiplies each document’s CombSUM score by the number of engines that retrieved it — the classic CombMNZ.
CombSumResultMerger	class	IResultMerger	Blend strategy that sums the normalized per-engine scores of every document — the classic CombSUM of score-based fusion.
CrossEncoderReranker	class	IReranker	Precision-oriented reranker: re-scores every candidate with a pairwise (cross-encoder) model and reorders the list by the model’s judgment.
HybridTextSearchEngine	class	IDetailedSearchEngine, ITextSearchEngine	A federated search engine: queries a list of delegate engines, then merges their results into one coherent ordering through an IResultMerger.
MaxSimReranker	class	IReranker	Late-interaction reranker (ColBERT-style): scores each candidate with MaxSim between the token-level embeddings of the query and of the document, then reorders…
MaximalMarginalRelevanceReranker	class	IReranker	Diversity-preserving reranker: greedily re-orders candidates so that each next pick is both relevant to the query and different from what has already been…
MergeExplanation	record	—	The arithmetic a weighted merge applied: one weight and one divisor per source engine.
ReciprocalRankFusionMerger	class	IResultMerger	Merges ranked lists with Reciprocal Rank Fusion: each document accumulates the reciprocal of its rank (plus a constant k) across every engine that returned it.
RerankingResultMerger	class	IResultMerger	Default merger: rebuilds a single temporary index from the union of candidate documents returned by every engine, then re-scores every candidate with one shared…
VectorSimilarity	class	—	Pure vector math helpers used to compare embeddings. These are arithmetic utilities, not an embedding pipeline: vectors must already exist (see…
WeightedScoreResultMerger	class	IMergeExplainer, IResultMerger	Blends the native per-engine scores instead of re-ranking. Every engine’s scores are normalized to [0, 1] (divided by that engine’s maximum positive score),…

LexiSharp.Indexing

Type	Kind	Contracts	What it is
AverageLengthDivisor	enum	—	Which document count divides the corpus token count when AverageDocumentLength is computed.
ContextEnrichingIndex	class	ICandidateIndex, IFieldStatisticsIndex, IReadOnlyTextIndex, ITextIndex, IVocabularyIndex	An ITextIndex whose documents first pass through an IChunkContextEnricher: the enriched text is what gets tokenized and scored, while the original document the…
ExpansionTextIndex	class	ICandidateIndex, IFieldStatisticsIndex, IReadOnlyTextIndex, ITextIndex, IVocabularyIndex	An InMemoryTextIndex whose documents are additionally indexed under the weak terms returned by an ITermExpander — the « semantic lexical index ». The search…
FieldPrefixContextEnricher	class	IChunkContextEnricher	A model-free IChunkContextEnricher that prefixes a document’s text with the values of named Fields — the deterministic, offline cousin of a context-injection…
InMemoryTextIndex	class	ICandidateIndex, IFieldStatisticsIndex, IReadOnlyTextIndex, ITextIndex, IVocabularyIndex	In-memory inverted index: for each term, the documents containing it and the positions within each document. Exposes the corpus statistics required by scorers.
InMemoryVectorSearchEngine	class	IQueryCostProbe, IQuerySyntaxSupport, ITextSearchEngine	In-memory dense (vector) search engine: embeds documents with an IEmbeddingProvider and answers queries by cosine similarity over the whole corpus — a linear…
InMemoryVectorSearchOptions	record	—	Options for InMemoryVectorSearchEngine: how its vectors are stored, which is the only choice the engine makes on its own storage.
JournalFlushPolicy	enum	—	How long a write-ahead journal entry may sit unflushed.
JournalOptions	record	—	What a store’s write-ahead journal does: whether it exists, when it flushes, and how much it may let accumulate.
SegmentSourceMode	enum	—	How a store opens the segments already in its directory.
SegmentedTextIndex	class	ICandidateIndex, IDisposable, IReadOnlyTextIndex, ITextIndex, IVocabularyIndex	A search index that can be written to and read back from disk: immutable segments, a live overlay, and tombstones that survive until a merge reclaims them.
SparseIndexEntry	record	—	A document together with its stored sparse vector — the atomic unit of Export and Import, and what a persistence backend serializes to reload a corpus without…
SparseTextSearchEngine	class	IQuerySyntaxSupport, ITextSearchEngine	In-memory sparse search engine: indexes documents as their sparse learned embeddings (SPLADE, uniCOIL, …) and answers queries by dot-product scoring.
VectorIndexEntry	record	—	A document together with its stored dense embedding — the atomic unit of Export / Import, letting a corpus be reloaded without re-embedding it.
VectorIndexSnapshot	record	—	A whole vector index: its entries and the identity of the embedding pipeline that produced them. ExportSnapshot writes one and ImportSnapshot reads it, refusing…
VectorQuantization	enum	—	How InMemoryVectorSearchEngine stores the document vectors it scans.

LexiSharp.Keywords

Type	Kind	Contracts	What it is
Keyword	record	—	A keyword extracted from a text, with its extractor-specific relevance score (higher is better; the score scale depends on the extractor).
TextRankKeywordExtractor	class	IKeywordExtractor	Keyword extraction with the graph-based TextRank algorithm: terms are nodes, co-occurring terms within a sliding window are edges, and the PageRank of each node…
TfIdfKeywordExtractor	class	IKeywordExtractor	Keyword extraction by smoothed TF-IDF: a term’s score is its frequency inside the text, optionally discounted by how common the term is across a reference…

LexiSharp.Linguistics

Type	Kind	Contracts	What it is
FrenchStemmer	class	IStemmer	French stemmer implementing the Snowball French algorithm: an IStemmer that folds inflected French forms onto a common stem (continuer → continu, contractait →…
NormalizedTerm	record struct	—	One normalized term of a tokenized text: a slice of the text that produced it, or the string it had to be materialized into.
PorterStemmer	class	IStemmer	English stemmer implementing Porter’s suffix-stripping algorithm: an IStemmer that folds inflected English forms onto a common stem (caresses → caress, motoring…
StopWords	class	—	Built-in sets of low-information words that can be removed during tokenization.
SynonymMap	class	—	Raw synonym edges: one-way rewrites (Add) and bidirectional equivalence groups (AddEquivalent). Entries are plain text; the search engine tokenizes them with…
TokenSpan	record struct	—	A normalized term together with the half-open character range [Start, Start + Length) it occupies in the original text. Offsets are UTF-16 indices (string…
Tokenizer	class	ISpanTokenizer, ITokenizer	Splits raw text into normalized terms.
TokenizerOptions	record	—	Configuration knobs for the Tokenizer.
WordSegmentation	enum	—	How a character between two word characters affects the word, when WordSegmentation is not Flat.

LexiSharp.Ranking

Type	Kind	Contracts	What it is
Bm25Arithmetic	enum	—	The precision BM25’s arithmetic is carried out in.
Bm25FGridPoint	record	—	One evaluated BM25F configuration and the quality it achieved.
Bm25FParameterTuner	class	—	Finds the BM25F parameters (k1, b, and a weight per named field) that best satisfy a set of labeled validation queries, by brute-force grid search in two…
Bm25FParameters	record	—	BM25F tuning knobs: term-frequency saturation (k1), document-length normalization (b), and how much each field counts for.
Bm25FScorer	class	IFieldScopedScorer, IScoreExplainer, ITermOverlapScorer, ITextScorer	BM25F ranking: BM25 over a document split into weighted fields, so a term in a title can count for more than the same term in a body.
Bm25FTuningResult	record	—	The outcome of a Tune run: the best configuration found, the quality it achieved, and the full evaluation grid for inspection.
Bm25GridPoint	record	—	One evaluated (k1, b) combination and the quality it achieved.
Bm25LGridPoint	record	—	One evaluated BM25L configuration and the quality it achieved.
Bm25LParameterTuner	class	—	Finds the BM25L parameters (k1, b and the lower bound delta) that best satisfy a set of labeled validation queries, by brute-force grid search.
Bm25LScorer	class	IScoreExplainer, ITermOverlapScorer, ITextScorer	BM25L relevance: Okapi BM25 with a compressed term frequency in the numerator and a lower bound on it, which lengthens the reach of rare terms in long…
Bm25LTuningResult	record	—	The outcome of a Tune run: the best configuration found, the quality it achieved, and the full evaluation grid for inspection.
Bm25ParameterTuner	class	—	Finds the Okapi BM25 parameters (k1, b) that best satisfy a set of labeled validation queries, by brute-force grid search.
Bm25Parameters	record	—	Okapi BM25 tuning knobs: term-frequency saturation (k1) and document-length normalization (b).
Bm25PlusGridPoint	record	—	One evaluated BM25+ configuration and the quality it achieved.
Bm25PlusParameterTuner	class	—	Finds the BM25+ parameters (k1, b and the lower bound delta) that best satisfy a set of labeled validation queries, by brute-force grid search.
Bm25PlusScorer	class	IScoreExplainer, ITermOverlapScorer, ITextScorer	BM25+ relevance: Okapi BM25 with the hard term-presence cut replaced by a lower-bounded one, so a document that matches more distinct query terms is rewarded…
Bm25PlusTuningResult	record	—	The outcome of a Tune run: the best configuration found, the quality it achieved, and the full evaluation grid for inspection.
Bm25Scorer	class	IScoreExplainer, ITermOverlapScorer, ITextScorer	Okapi BM25 relevance — the standard non-semantic ranking model for full-text search. It blends term saturation (k1) with document-length normalization (b).
Bm25TuningResult	record	—	The outcome of a Tune run: the best parameter pair found, the quality it achieved and the full evaluation grid for inspection.
Bm25ValidationQuery	record	—	One labeled validation query of a Bm25ParameterTuner: a raw query together with the ids of the documents a good ranking should surface.
BooleanMatch	enum	—	How the modifiers of a BooleanScorer are combined.
BooleanScorer	class	IScoreExplainer, ITermOverlapScorer, ITextScorer	Pure boolean filter. A matching document gets a score of 1, a non-matching one 0. Because the RankedTextSearchEngine discards scores below MinimumScore, this…
CalibratedScoreConfidence	class	—	A learned confidence calibrator that maps raw scores to calibrated probabilities using isotonic regression or Platt scaling, with an optional Youden-derived…
CalibrationMethod	enum	—	The calibration method used by CalibratedScoreConfidence.
DocumentLengthRatioScorer	class	ITextScorer	A document’s length relative to the corpus average: |D| / avgdl, where both figures come from the index. One for an average document, above one for a long…
FeedbackAwareTextSearchEngine	class	IQueryCostProbe, ITextSearchEngine	A decorator engine that augments a primary search engine with a learned query → document feedback channel — one history for every search.
FeedbackAwareTextSearchEngine<TPayload>	class	IContextualSearchEngine, IQueryCostProbe, ITextSearchEngine	A decorator engine that augments a primary search engine with a learned query → document feedback channel, choosing the history per search from the caller’s…
NdcgGain	enum	—	How a relevance level is turned into a gain in NdcgAtK.
ProximityMode	enum	—	How ProximityReranker turns a term window into a score change.
ProximityReranker	class	IReranker	Re-ranks a candidate list by how close together the query terms actually occur in each document, refining the order a term-frequency scorer produced without…
QueryFeedbackHistory	class	—	A learned relevance feedback channel: it records which document a user chose for past queries, and uses those associations to boost matching documents in future…
QueryLikelihoodScorer	class	IScoreExplainer, ITermOverlapScorer, ITextScorer	Query likelihood retrieval model with Jelinek-Mercer smoothing, a probabilistic language-model alternative to BM25.
QueryTermWeighting	enum	—	How a scorer treats a term that occurs more than once in a query.
RankedTextSearchEngine	class	IExplainableSearchEngine, IFacetedSearchEngine, IQueryCostProbe, IQuerySyntaxSupport, ITextSearchEngine	The stock search engine: delegates the corpus storage to an IReadOnlyTextIndex, delegates the relevance math to an ITextScorer, and takes care of query…
RetrievalMetrics	class	—	Standard top-k retrieval metrics over a ranked list of document ids, used to evaluate (and tune) ranking quality against a validation set.
ScoreConfidence	class	—	Maps an ordered result set’s raw scores into calibrated confidences in [0,1] — the “is this really the answer?” gauge that raw lexical scores (BM25, TF-IDF,…
ScoreConfidenceMethod	enum	—	How an ordered result set’s raw scores are mapped into a calibrated per-result confidence in [0,1].
TfIdfScorer	class	IScoreExplainer, ITermOverlapScorer, ITextScorer	Classic TF-IDF relevance. A document matches when it contains at least one query term; rare terms are weighted more than common ones.
TieBreak	enum	—	How the engines order candidates whose scores are exactly equal.
TuningMetric	enum	—	The quality metric a Bm25ParameterTuner maximizes over its validation set.
WeightedCompositeScorer	class	ITextScorer	A scorer built from other scorers: the weighted sum of their scores, Σ wᵢ · componentᵢ, computed in one pass over one candidate set.
WindowBm25Scorer	class	ITermOverlapScorer, ITextScorer	Scores a document by its best window rather than by its whole text: BM25’s saturation computed over term frequencies counted inside a sliding window, and the…

LexiSharp.Similarity

Type	Kind	Contracts	What it is
LevenshteinDistance	class	—	Levenshtein edit distance between two strings: the minimum number of single-character insertions, deletions or substitutions to turn one into the other.
LexicalSimilarity	class	—	Pairwise lexical similarity measures over short strings — the in-memory answer to “are these two texts roughly the same?”: near-duplicate detection,…

LexiSharp.Sources

Type	Kind	Contracts	What it is
JsonDocumentLoadOptions	record	—	Knobs for JsonDocumentsLoader: which object property holds what.
JsonDocumentsLoader	class	—	Loads documents from a JSON array of objects, e.g. [{“id”: “1”, “text”: “hello”}].
LoadedDocument	record	—	A document loaded from an external source (a markdown file, a JSON record, a text file) and ready to be indexed — the bridge between your data and the search…
MarkdownLoadOptions	record	—	Knobs for MarkdownLoader.
MarkdownLoader	class	—	Loads markdown documents: the optional YAML front matter (—-delimited block at the top of the file) is parsed into structured fields, and the rest of the file…
TextFileLoadOptions	record	—	Knobs for TextFileLoader.
TextFileLoader	class	—	Loads plain text files from a directory, whole file content becoming the indexed text.

LexiSharp.MessagePack package

LexiSharp.MessagePack

Type	Kind	Contracts	What it is
MessagePackSparseIndexPersistence	class	—	Saves and reloads a SparseTextSearchEngine as MessagePack binary (LZ4-compressed): the documents themselves — id, text, fields, category — plus the learned…
MessagePackTextIndexPersistence	class	—	Saves and reloads an InMemoryTextIndex as MessagePack binary (LZ4-compressed): the documents themselves — id, text, fields, category — plus the tokenizer…

LexiSharp.Postgres package

LexiSharp.ParadeDB

Type	Kind	Contracts	What it is
ParadeDBOptions	record	—	Naming and behavior options for the ParadeDB-backed engine.
ParadeDBTextSearchEngine	class	IDisposable, IQuerySyntaxSupport, ITextSearchEngine	ParadeDB/PG_search-backed ITextSearchEngine: true BM25 ranking built on the Tantivy index of the pg_search extension. The search field is matched with the…

LexiSharp.Postgres

Type	Kind	Contracts	What it is
PostgresFieldWeight	enum	—	The four weight classes a tsvector lexeme can carry, which setweight writes and ts_rank_cd reads back through its weights array.
PostgresFuzzyOptions	record	—	Naming and behavior options for the PostgresFuzzySearchEngine.
PostgresFuzzySearchEngine	class	IDisposable, IQuerySyntaxSupport, ITextSearchEngine	PostgreSQL-backed ITextSearchEngine for approximate, typo-tolerant matching on top of the pg_trgm extension (trigram similarity) and fuzzystrmatch (levenshtein…
PostgresIndexOptions	record	—	Naming and behavior options for the PostgreSQL-backed index.
PostgresSchema	class	—	Shared DDL for the documents table. Used by the lexical (tsvector), vector (pgvector ANN) and ParadeDB (pg_search BM25) engines so they can all point at the…
PostgresSparseOptions	record	—	Naming, vocabulary and behavior options for the pgvector sparsevec index. The table layout is shared with the lexical engine so both engines can serve a single…
PostgresSparseSearchEngine	class	IDisposable, IQuerySyntaxSupport, ITextSearchEngine	PostgreSQL-backed learned-sparse search engine (pgvector sparsevec + HNSW): stores each document’s learned sparse vector (SPLADE-style term weights) in the…
PostgresTextSearchEngine	class	IDisposable, IQuerySyntaxSupport, ITextSearchEngine	PostgreSQL-backed ITextSearchEngine: lexical full-text search over a tsvector column with a GIN index, queried with websearch_to_tsquery and ranked with…
PostgresVectorOptions	record	—	Naming and behavior options for the embeddings-backed (pgvector) index. The table layout is shared with the lexical engine so both engines can serve a single…
PostgresVectorSearchEngine	class	IDetailedSearchEngine, IDisposable, IListableSearchEngine, IQuerySyntaxSupport, ITextSearchEngine	PostgreSQL-backed vector search engine (pgvector ANN): stores document embeddings in a shared documents table and answers queries with an approximate…
SparseDistance	enum	—	Distance operator driving the sparse ANN index and the query ordering.
TrgmIndexKind	enum	—	Which trigram index structures are created on the content field.
TrgmSearchMode	enum	—	The way approximate (trigram) search matches documents.
VectorDistance	enum	—	Distance operator driving the ANN index and the query ordering.
VectorIndexMethod	enum	—	Index access method used for the ANN index over the embedding column.

LexiSharp.AspNetCore package

LexiSharp.AspNetCore

Type	Kind	Contracts	What it is
LexiSharpIndexHealthCheck	class	IHealthCheck	Reports whether a search index is ready to serve traffic: non-empty, and optionally connected to its backing store and not too stale.
LexiSharpIndexHealthCheckOptions	class	—	Tuning for the registration helpers on LexiSharpObservabilityServiceCollectionExtensions.
LexiSharpObservabilityServiceCollectionExtensions	class	—	One-call registration of the LexiSharp observability pieces on an IServiceCollection.
LexiSharpSearchEndpoint	class	—	The testable heart of the search endpoint: an HTTP-agnostic function from an index and request parameters to an IResult. The minimal-API extension only wires…
LexiSharpSearchEndpointBuilder	class	—	Minimal-API wiring for the LexiSharpSearchEndpoint handler.
LexiSharpSearchError	record	—	The payload of a failed search response.
LexiSharpSearchHit<TDocument>	record	—	One hit of a LexiSharpSearchResponse.
LexiSharpSearchParameters	record	—	Query-string parameters of a LexiSharpSearchEndpoint request, bound by ASP.NET Core model binding. Mirror the knobs of LexiSharpQueryOptions.
LexiSharpSearchResponse<TDocument>	record	—	The payload of a successful search response.
RetrievalLogger	class	—	Bridges a RetrievalTelemetry onto ILogger, so an application that already has a logging pipeline gets retrieval events in it without the LexiSharp core taking a…

LexiSharp.Onnx package

LexiSharp.Onnx

Type	Kind	Contracts	What it is
BatchedCrossEncoderReranker	class	IReranker	The batched counterpart of CrossEncoderReranker: it re-scores the whole shortlist in one padded session call instead of one call per candidate, and otherwise…
BertWordPieceOptions	class	—	Normalization rules for BertWordPieceTokenizer. The BERT family does not agree on them: bert-base-uncased lowercases and strips accents, bert-base-cased does…
BertWordPieceTokenizer	class	IOnnxTokenizer	A BERT WordPiece tokenizer over a vocab.txt: the tokenizer most BERT-derived encoders and cross-encoders (including the MiniLM family) publish beside their ONNX…
BpeOnnxTokenizer	class	IOnnxTokenizer	A byte-level BPE tokenizer loaded from a HuggingFace tokenizer.json — the file most modern exports actually ship, where vocab.json and merges.txt are no longer…
BpeOnnxTokenizerOptions	class	—	Options for BpeOnnxTokenizer: the file carries the vocabulary and the merges, the caller states the byte-level and special-token conventions.
OnnxCrossEncoderOptions	class	—	Everything the pairwise scorer needs from the caller. As with the dense encoder, no default guesses a checkpoint: the export states its own tensor names, its…
OnnxCrossEncoderScorer	class	ICrossEncoderScorer, IDisposable	A pairwise relevance model backed by an ONNX Runtime session: the shipped implementation of ICrossEncoderScorer for a local cross-encoder, ready for…
OnnxPooling	enum	—	How a token-level model output is reduced to one vector per text.
OnnxScoreTransform	enum	—	How the selected cross-encoder logit is turned into a score.
OnnxTextEncoder	class	IBatchEmbeddingProvider, IDisposable, IEmbeddingProvider, IEmbeddingSignatureProvider	A dense text encoder backed by an ONNX Runtime session: the shipped implementation of IEmbeddingProvider when the model runs in-process rather than behind an…
OnnxTextEncoderOptions	class	—	Everything the dense encoder needs from the caller: the model file, the tokenizer and the graph’s own names. Nothing here has a default that guesses a…
OnnxTokenizedText	struct	—	A tokenized text ready to be bound to an ONNX graph: the token ids and the optional segment ids, with no padding applied. Padding and batching are the caller’s…
SentencePieceOnnxTokenizer	class	IOnnxTokenizer	A SentencePiece tokenizer over a checkpoint’s sentencepiece.bpe.model (or spiece.model): the tokenizer E5, multilingual MiniLM and most non-BERT encoders…
SentencePieceOnnxTokenizerOptions	class	—	Options for SentencePieceOnnxTokenizer: the model decides the pieces, the caller decides the special tokens and the id remapping.