schema · evergreen

Four hard things about code RAG

Vector search is the easy part: exact symbols, Turkish over English code, staleness, and saying no.

Vector search is the easy part of a codebase RAG. The four things that are actually hard decided every part of this system.

what breaks first

exact symbolshandleAuthCallback

Turkish → Englishprose vs. code

stalenessrepo moved on

saying "no"kNN never says no

Problem	What we did	Measured
Symbols	route symbol-shaped queries to BM25	1.0 / 1.0 recall@8 / MRR
Turkish questions	multilingual embedder, optional LLM descriptions	0.04 → 0.684 → 1.0 (subset)
Staleness	sha256 manifest diff on push	38 changed files in 23 s
No answer	three score bands + negative golden cases	abstain 0.846, false alarm 0.048

Every row is a decision with a number next to it. The rest of this garden is the story of those numbers — including the ones that changed our minds.