Insight Extractor Reference¶

API reference for the cross-metric insight extraction module.

from axion.reporting import (
    InsightExtractor,
    InsightPattern,
    InsightResult,
)

I

InsightExtractor

Bridges IssueExtractor output with EvidencePipeline for cross-metric pattern discovery and learning distillation.

P

InsightPattern

A discovered cluster enriched with cross-metric metadata — which metrics are involved, test case coverage, and confidence.

R

InsightResult

Complete result containing patterns, learnings, and the full pipeline result for advanced access.

InsightExtractor¶

axion.reporting.insight_extractor.InsightExtractor ¶

InsightExtractor(model_name: Optional[str] = None, llm=None, llm_provider: Optional[str] = None, method: ClusteringMethod = LLM, recurrence_threshold: int = 2, max_items: int = 50, min_category_size: int = 2, pipeline: Optional[EvidencePipeline] = None, **pipeline_kwargs)

Bridges IssueExtractor output with EvidencePipeline for cross-metric pattern discovery.

analyze `async` ¶

analyze(extraction_result: IssueExtractionResult) -> InsightResult

Analyze extracted issues for cross-metric patterns.

Parameters:

extraction_result (IssueExtractionResult) –

Output from IssueExtractor.extract_from_evaluation()

Returns:

InsightResult –

InsightResult with discovered patterns and learnings.

analyze_sync ¶

analyze_sync(extraction_result: IssueExtractionResult) -> InsightResult

Synchronous wrapper for analyze().

Data Classes¶

InsightPattern¶

axion.reporting.insight_extractor.InsightPattern `dataclass` ¶

InsightPattern(category: str, description: str, count: int, issue_ids: List[str], metrics_involved: List[str], is_cross_metric: bool, distinct_test_cases: int, examples: List[str], confidence: Optional[float] = None)

A cross-metric pattern discovered from evaluation issues.

InsightResult¶

axion.reporting.insight_extractor.InsightResult `dataclass` ¶

InsightResult(patterns: List[InsightPattern], learnings: List[LearningArtifact], total_issues_analyzed: int, clustering_method: ClusteringMethod, pipeline_result: PipelineResult)

Complete result from insight extraction.

Adapter Function¶

_issue_to_evidence¶

axion.reporting.insight_extractor._issue_to_evidence ¶

_issue_to_evidence(issue: ExtractedIssue) -> Optional[EvidenceItem]

Convert ExtractedIssue -> EvidenceItem. Returns None if no meaningful text.

Insight Extraction Guide Issue Extractor Reference Pattern Discovery