chatbi-maintenancelisted

# chatbi-maintenance Maintenance runbook for governed models and semantic definitions. A model change is delivered only after an impact manifest is generated and all affected assets are synced (DOC-001/004). ## 0. Tooling (read before running dbt) Before invoking `dbt run` / `dbt test` / any dbt command (Step 3 candidate verification, Step 4 sync gate), read `docs/org/data-warehouse-blueprint.md` § Tooling for the environment-specific dbt executable path and MySQL connection. dbt's Python runtime is separate from the harness `CHATBI_PYTHON` - on some hosts dbt requires a different Python (e.g. 3.12 vs 3.14 where `mashumaro` crashes), so an arbitrary `dbt` found on PATH may be the wrong one. Use the dbt named in the Tooling section, not one discovered on PATH. If the Tooling section is absent or does not name a dbt executable, STOP and ask the operator which dbt to use - do not guess from PATH (unconfirmed PATH entries are not execution authority, PORT-001 spirit). The Tooling section is operator guidance, not governed knowledge (DOC-001 does not apply to it). ## 1. Classify the change `change_kind` ∈ model/column/semantic/reference/Skill/downstream/eval; target is a logical alias. If the change is a protected action (`approve_metric`, `change_access_policy`, `production_publish`, `destructive_migration`), STOP and request human owner approval (SEM-003). Agent drafting is not approval. ## 2. Build the impact manifest `chatbi_harness.impact.build_impact_manifest(...)` ov