Coding agents receive only 32% of project rules; docs boost it to 70%

Segmently released the open-source context-recall benchmark, finding that coding agents capture only 32% of implicit project rules from code alone, rising to 70% with rule documentation plus retrieval.

2026-10-28 ~ 2026-10-29 · 2 related posts

• Only 32% of project rules reach Claude Code; docs lift it to 70% — open benchmark context-recall — AttitudeLatter7306 · 2026-10-28
• Only 32% of Project Rules Reach Claude Code; Docs Plus Retrieval Lifts It to 70% — AttitudeLatter7306 · 2026-10-29