Class IngestionRecipe

Namespace

Qavren.Edge.Ingestion

Assembly

Qavren.Edge.Ingestion.dll

The tuple a chunk's vector is a pure function of (spec 9.2). When a document's stored recipe_hash differs from the run's, that document is dirty regardless of its content hash — which is what stops a MaxTokens change from leaving every old vector in place.

public sealed record IngestionRecipe : IEquatable<IngestionRecipe>

Inheritance

object

IngestionRecipe

Implements

IEquatable<IngestionRecipe>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Constructors

IngestionRecipe(int, string, int, string, string?, string?, string, int, int, string, int, ResolvedChunkOptions, string, string)

The tuple a chunk's vector is a pure function of (spec 9.2). When a document's stored recipe_hash differs from the run's, that document is dirty regardless of its content hash — which is what stops a MaxTokens change from leaving every old vector in place.

Fields

CurrentSchemaVersion

The version this build writes.

Properties

ChunkerId

The selected chunker.

ChunkerVersion

Its version.

Chunking

The frozen budget, field by field.

Dimensions

The collection's declared width.

DistanceFunction

The collection's distance function.

DocumentPrefix

A token RESERVE and a recipe input only; SP3 never writes it.

ExtractorFingerprint

The SELECTED extractor's "{Id}:{Version}", never the registry's: hashing the set would re-index an entire Markdown corpus because the PDF extractor's version was bumped. Because the selection is per document, the recipe a run freezes at startup carries the stand-in "*" in this one field, and the recipe actually written to and compared against recipe_hash is that baseline with the chosen extractor's fingerprint substituted. The baseline is what GetRecipeAsync, IngestionRunResult.RecipeHash and IngestionStatus.RecipeHash report, because it is the one recipe identity that does not depend on which file was next.

Hash

Thirty-two lowercase hex, 0x1F-separated and order-stable.

HashValue

The hash as a ContentHash, which is the persisted form.

ModelProfileId

SP2's lower-case preset id, verbatim.

Pooling

The model's pooling mode.

QueryPrefix

The query-side prefix, for completeness.

SchemaVersion

The recipe's own shape version. Bump when a field is added.

SpecialTokenOverhead

Tokens the encoder adds that CountTokens omits.

TokenizerId

The tokenizer's vocabulary identity.

TokenizerMaxSequenceLength

The tokenizer's ceiling.