Legibility

A system is legible when a capable engineer, or an AI agent, can look at a piece of behavior and understand why it works that way, without hunting down the one person who remembers.

Most codebases run fine and stay illegible at the same time. The tests pass. The service ships. Nobody can say why a particular retry limit exists, or why one customer segment gets a special code path. That gap between working and understood is where tribal knowledge does its quiet, load-bearing work, and where an agent handed the same task starts to guess.

The toolkit treats legibility as something you can score, not something you feel. A handful of short questions, about intent, history, dependency choices, and exceptions, turn that vague unease into a number you can compare across capabilities. Seetribal knowledgefor the failure mode legibility is meant to fix.