v0.2.0-paperweave-upgrade

v0.2.0-paperweave-upgrade

This release marks the completed PaperWeave upgrade milestone for research-reading-agent.

Highlights

PaperWeave is the project's paper evidence weaving and hybrid retrieval evaluation system. This milestone consolidates the major optional capabilities developed during the 2026-06 upgrade cycle into a clean, tested, and auditable main branch.

Included capabilities

• Optional sentence-transformers embedding provider
• CI, LICENSE, and offline test isolation
• Citation faithfulness evaluation
• Optional LLM answer synthesis
• Optional SQLite vector store
• Optional Chroma vector store adapter
• Optional cross-encoder reranker
• Optional multi-step Agent execution
• Optional async workflow jobs

Design principles

• Default embedding provider remains hash
• Optional capabilities are disabled unless explicitly configured
• Main test suite remains offline, deterministic, and reproducible
• feature/upgrade-2026-06 was audited and intentionally not merged as a whole
• Legacy or conflicting implementations, such as provider=auto, max_steps=4, and old monolithic vector-store code, were intentionally left out

Validation

• Full pytest: 239 passed
• py_compile: passed
• Risk checks: passed
• Git status before release: clean
• Release commit: 0197db7 Merge optional async workflow jobs

Repository

https://github.com/mxz-dddd/research-reading-agent