Runbooks

One runbook per delivery phase, each covering how that subsystem is brought up, exercised, and verified — the operational counterpart to the architecture decisions.

Phase	Runbook	Covers
0	Environment bootstrap	venv, install, .env, diagnostics, the unified LLM adapter
1	Data ingestion	12 source types, chunking, embedding, Qdrant + BM25 indexing, DuckDB
2	RAG and memory	Hybrid retrieval, RRF fusion, re-ranking, scoped and episodic memory
3	Orchestrator	MCP tools, agent strategies, supervisor routing, Langfuse tracing
4	Canvas UI	React Flow canvas, live run state over Redis Streams and WebSockets
5	Fine-tuning	LoRA/QLoRA/DPO/RLHF jobs, evaluation, the model registry
6	Project import	Static analysis, AST metrics, LLM review, patch apply

Commands in these runbooks assume the source repository, which is private — see Source access.