Version: latest

`agrag.ingestion.stats.extraction.ExtractionStats`

Bases: BaseModel

Extraction-stage results.

Attributes:

chunks_processed (int) – Chunks the stage read. In llm mode this counts every chunk, including chunks skipped for having fewer than min_escalation_words words. EscalationReport.skipped_short gives that skipped count.
entities_extracted (int) – Entities the extractor returned.
relations_extracted (int) – Relations the extractor returned.
dropped (list[DroppedCount]) – Extractor discards counted by kind, label and reason. Always filled, whether candidates are on or off.
candidates_written (int) – Candidates this run wrote to the graph.
candidates_skipped (int) – Discarded items not written because a reviewer settled the key or both relation ends merged into one entity.
failures (list[StageFailure]) – Per-item failures, capped per call.
failures_total (int) – Failures recorded before capping.
failures_truncated (bool) – Whether failures was cut to the cap.

`candidates_skipped`

candidates_skipped: int = 0

candidates_written: int = 0

chunks_processed: int = 0

dropped: list[DroppedCount] = Field(default_factory=list)

entities_extracted: int = 0

failures: list[StageFailure] = Field(default_factory=list)

failures_total: int = 0

failures_truncated: bool = False

relations_extracted: int = 0