verbose live UI + tool-level SSE events + Groq default + regression tests

api/datasets/README.md

@@ -0,0 +1,23 @@
+# Datasets
+
+One subdirectory per dataset. The folder name is the dataset's identity —
+used as:
+
+- the `NVI_DATASET` env var value that selects it,
+- the Postgres schema where its tables live,
+- the SQLite filename in `ctrl/seed/data/<name>.sqlite` that the loader
+  reads from.
+
+## Adding a new dataset
+
+1. Place the source SQLite at `ctrl/seed/data/<name>.sqlite` (or adapt
+   `ctrl/seed/download.sh` to fetch it).
+2. Create `api/datasets/<name>/`:
+   - `schema_docs.yaml` — table and column descriptions.
+   - `metrics.yaml` — canonical SQL for business terms.
+3. Set `NVI_DATASET=<name>` and restart the api.
+4. `make seed` will load `<name>.sqlite` into Postgres schema `<name>`.
+
+The Plan/Analyses/Tools/Runtime code references no dataset name; only the
+loader (`api.tools.schema.load_schema`) and the engine (`api.tools.db`)
+consult the active-dataset setting.