codebase-navigationlisted

# Codebase navigation The repository is never read whole. Every read carries a hypothesis it will confirm or kill, and exploration stops when the next file is unlikely to change the plan. The descent target is named file(s) and line region(s) — reading past that is scope creep. ## Evidence ranks Docs are evidence, not truth — they record what someone once intended. Code is evidence, not architecture — it shows what exists, not how it behaves when run. Runtime behavior beats both. A claim about the codebase inherits the rank of the evidence behind it. ## Hard rules - **No edits before the ownership path is located** — the named file(s) and line region(s) that own the behavior. The first edit then enters through a declared change mode (change-strategy). - **No architecture claims from docs or filenames alone.** Code or runtime evidence stands behind every claim (Layline doctrine: evidence over assertion). - **Build, run, and test commands are observed, never guessed.** Executed, exit status recorded — "it's probably `make test`" is a hypothesis, not a result. - **Git is a navigation instrument, not just storage.** Churn hotspots name the load-bearing files; `git log -S`/`-L` and `git bisect` localize when and where behavior changed faster than reading ever will. - **Harness tools first, shell as fallback.** Where the runtime provides dedicated search and read tools (Grep, Glob, ranged reads, read-only subagents), those replace the raw grep/sed equivalents — the recipes in