Crate lance_context_core 

Core types for the lance-context storage layer.

Modules

serde

Structs

AbReport

Result of comparing the same query set across two dataset versions.

CompactionConfig

Configuration for background compaction.

CompactionMetrics

Metrics returned by compact_files.

CompactionStats

Statistics about compaction status and history.

Context

ContextEntry

ContextNamespace

Thin resolver for scoped context partitions under one namespace root.

ContextRecord

User-facing representation of a context entry written to storage.

ContextStore

Persistent Lance-backed context store.

ContextStoreOptions

Additional configuration when opening a ContextStore.

Distribution

Numeric distribution summary.

EvalConfig

Runtime configuration for an evaluation run.

EvalQuery

One labeled query: a vector and/or text channel plus its relevant records.

EvalQuerySet

A labeled query set, referenced by a stable id for reproducible reports.

EvalReport

A reproducible evaluation report: a manifest (query-set id, version, config) plus aggregate and per-query scores.

ExcludedCounts

Source records excluded during curation, by reason.

ExportConfig

Curation + export configuration.

ExportCounts

Record counts at each curation stage.

ExportManifest

Reproducible description of an export, written next to the JSONL output.

ExportStats

Auditable dataset statistics for one export output, written to <output_path>.stats.json.

LifecycleQueryOptions

Query-time controls for lifecycle-aware retrieval.

Message

A single chat message in a training example.

MetricScores

Aggregate or per-query retrieval-quality metrics, all in 0.0..=1.0.

PartitionInfo

Manifest row describing one resolved context partition.

PartitionSpec

Phase-1 partition specification for a context namespace.

Provenance

Where an exported example came from, for auditability.

QueryEval

Metrics and retrieved ids for a single query.

RankedCandidate

One ranked candidate in an N-way preference example.

RecordFilters

Filters applied to records before list pagination or search ranking.

RecordPatch

Mutable fields that can be patched without resupplying the payload.

Relationship

Directed relationship from this record to another graph node.

RelevanceLabel

A single relevance judgment: a relevant external_id and its grade.

RetrieveResult

Result returned from hybrid retrieval over context records.

RolloutExample

RL rollout example: a prompt plus one or more rewarded responses, with an optional group_id linking the N samples for one prompt.

RolloutResponse

One response within a rollout example.

SearchResult

Result returned from a vector similarity search.

SftExample

SFT example: an ordered message list (doubles as the rejection-sampling / Best-of-N target when filtered by min_reward).

Snapshot

SplitConfig

Reproducible train/eval split configuration.

SplitManifest

Records which side of a train/eval split an output is, and the parameters needed to reproduce the partition.

StateMetadata

Structured metadata captured alongside each context entry.

TokenStats

Token-length statistics and which source produced them.

UpdateResult

Result returned from partial record update operations.

UpsertResult

Result returned from insert-or-replace operations.

Enums

DistanceMetric

Distance metric used to rank candidates during vector search.

ExportTask

Which training shape to export.

GroupBy

How records are grouped into one training example.

IdIndexType

Type of scalar index on the id column.

MetadataFilter

Metadata matching operation for filtered retrieval.

PreferenceExample

Preference example in one of three forms, tagged by form.

PreferenceForm

Shape of a preference export (selected by the caller to match their labels).

RetrievalMode

Which retrieval API to evaluate.

Constants

EXPORT_SCHEMA_VERSION

Schema version stamped into every export manifest.

LIFECYCLE_ACTIVE

LIFECYCLE_CONTRADICTED

Type Aliases

PartitionSelector

Complete selector values for one context partition.