Testing and Quality

Existing signals

The backend has focused tests for extraction, tabular chunking, hybrid retrieval, RRF/reranking, Qdrant/Meilisearch, project/history services, prompts, session memory, language, ingestion, and orchestration. Frontend and backend package scripts provide build/lint/test entry points.

Recommended matrix

Boundary	Cases
Ingestion	malformed PDFs, OCR fallback, tables, hierarchy, retries
Retrieval	dense/sparse fusion, reranking, project filters, no-hit fallback
Agent	multi-hop decomposition, prompt selection, citation parsing
Storage	MinIO/Postgres/index consistency and re-ingestion
Security	cross-project/private dossier access
Streaming	chunk boundaries, cancellation, debug events
UI	SSO, project switching, viewers, history, partial responses

The most important regression suite is a private-project leakage test across every retrieval and download endpoint.