descope-sweeplisted

# De-scope sweep Codified as an architecture decision record (documentation-cascade). The pattern that keeps recurring: a de-scope correction lands cleanly on the primary surface (a top README, a repo description) and quietly survives everywhere else — ADRs, `.env.example`, KB stubs that get indexed and spoken by an agent, the source files that generate docs (portal source, docs-site bodies, diagram data), stale metric cells, and numbered ranges that don't grow. A full cross-repo audit found exactly this months after a dependency cut and a language port. The fix is a sweep that treats the long tail as the point, not an afterthought. ## How to run it 1. **Confirm the reconciliation target, not just the old term.** Get both sides from the user if they aren't already obvious: what's being searched for (e.g. "Kafka"), and what the *current truth* is to reconcile each hit against (e.g. "dropped, replaced by an in-process background task"). Don't guess the architecture decision. 2. **Sync before you grep.** Local clones can be stale — `git fetch`/`git pull` every repo before trusting what's on disk. For a repo with no local clone, search the remote directly rather than skipping it. 3. **Cast wide.** Grep every repo, but deliberately check these categories, not just READMEs — this is where past sweeps missed: - Top-level docs (README, CLAUDE.md) - ADRs / decision records (each repo's `decisions/`, the architecture repo's numbered decision docs) -