Repository navigation
v0.8.0
v0.8.0 (2026-01-30)
This release is published under the MIT License.
Bug Fixes
- Aggregate all warnings from topic modeling pipeline stages (
15fba05)
Collect warnings from all pipeline stages (text, LLM extraction, BERTopic, and fine-tuning) instead of just the text stage.
- Correct smart override selection logic for confidence filtering (
bf7f349)
Fix SelectSmartOverrideExtractor to skip extractions with confidence below threshold before checking meaningfulness. Ensures high-confidence extractions are preferred over those with missing or low confidence.
- Select highest confidence candidate when last extraction is not meaningful (
b2abacf)
When the last extraction lacks meaningful content, iterate through all previous candidates and select the one with the highest confidence score, preferring explicit confidence over missing confidence.
Co-Authored-By: Claude Haiku 4.5 noreply@anthropic.com
Documentation
- Document semantic release automation policy (
1682741)
Add section to AGENTS.md clarifying that semantic versioning, changelog generation, and PyPI publishing are fully automated in GitHub Actions and should not be handled locally.
Co-Authored-By: Claude Haiku 4.5 noreply@anthropic.com
- Update documentation for new features (
2018f88)
Update API documentation, extraction guide, user configuration, and README to reflect new extractors, analysis pipelines, and inference backends.
Features
- Add analysis pipeline infrastructure (
d740823)
Add base analysis pipeline framework with support for LLM-driven extraction and topic modeling. Includes schema validation and model abstractions.
- Add analysis run support to corpus (
d341c28)
Add analysis_runs_dir property and analysis_run_dir method. Add latest_extraction_run_reference helper for retrieving the most recent extraction run.
- Add analysis schema constants (
372d427)
Add ANALYSIS_SCHEMA_VERSION and ANALYSIS_RUNS_DIR_NAME constants needed by the analysis pipeline infrastructure.
- Add confidence scores to extraction output (
e3ac910)
Add optional confidence field (0.0-1.0) to ExtractedText and ExtractionStepOutput models to support extraction quality signals.
- Add extraction pipeline selection utilities (
cec306f)
Add SelectOverride and SelectSmartOverride extractors for intelligent extraction result selection in pipelines. Allows choosing extraction results based on media type or quality heuristics.
- Add inference backend abstraction (
4de0103)
Add composable inference backend configuration for components that support both local and API-based execution modes. Supports HuggingFace and OpenAI providers.
- Add provider configurations and extractor enhancements (
ac86e6e)
Add Deepgram and HuggingFace user configuration support. Update extractors with confidence score support and refined implementations for Docling, PaddleOCR, RapidOCR, and selection utilities.
- Add topic modeling analysis command and recipe file support (
6a60f57)
Add biblicus analyze topics command with recipe file support. Improve step spec parsing to handle nested structures (braces, brackets) in extraction pipeline configurations.
- Propagate confidence scores through extraction pipeline (
1616d45)
Thread confidence scores from ExtractedText through ExtractionStepResult to support extraction quality signals in the pipeline.
- Register new extractors in factory (
18c0280)
Register PaddleOCR, Docling, Deepgram, and selection utility extractors in the extractor factory.
Testing
- Add feature specs and steps for inference and selection utilities (
84205c3)
Add BDD feature specifications and step implementations for inference backend configuration and extraction result selection utilities.
- Add feature specs for analysis pipelines (
c98cded)
Add BDD specifications for analysis pipeline infrastructure including topic modeling analysis and schema validation.
- Update feature specs and test steps for new features (
636bd86)
Update test environment, step implementations, and test fixtures to support new extractors, analysis pipelines, inference backends, and configuration options.
Detailed Changes: v0.7.0...v0.8.0