Get the backend test suite green and run it in CI

.github/workflows/backend-tests.yml

@@ -0,0 +1,38 @@
+name: Backend tests
+
+on:
+  pull_request:
+    paths:
+      - "backend/**"
+      - ".github/workflows/backend-tests.yml"
+  push:
+    branches: [main]
+    paths:
+      - "backend/**"
+      - ".github/workflows/backend-tests.yml"
+
+permissions:
+  contents: read
+
+jobs:
+  pytest:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          lfs: false
+      - uses: astral-sh/setup-uv@v6
+        with:
+          version: "0.11.21"
+          enable-cache: true
+          cache-dependency-glob: backend/uv.lock
+      - name: Install locked Python dependencies
+        working-directory: backend
+        run: uv sync --frozen --python 3.13
+      # No warehouse.duckdb or DuckLake exists here: tests that read real data
+      # are marked `requires_warehouse` and skip (see tests/conftest.py).
+      # test_base.py still imports the deleted data_collection_OLD package; drop
+      # the --ignore once it is ported to data_collection.base.
+      - name: Run the backend test suite
+        working-directory: backend
+        run: uv run --frozen pytest -q --ignore=tests/test_base.py

CLAUDE.md

@@ -28,11 +28,11 @@ npm run format:check # Check formatting without modifying
 uvicorn api.main:app --reload --port 6767  # API server on localhost:6767
 pip install -r requirements.txt            # Install dependencies
 streamlit run Home.py                      # Legacy Streamlit interface (not primary)
-uv run sqlfluff lint query/sql build/sql   # Lint the Jinja SQL templates
-uv run sqlfluff fix query/sql build/sql    # Auto-fix SQL lint issues
+uv run sqlfluff lint query/sql data_cleaning/sql   # Lint the Jinja SQL templates
+uv run sqlfluff fix query/sql data_cleaning/sql    # Auto-fix SQL lint issues
 ```
 
-SQL templates under `backend/query/sql/` and `backend/build/sql/` are Jinja
+SQL templates under `backend/query/sql/` and `backend/data_cleaning/sql/` are Jinja
 templates rendered by `query/sql_render.py`, which also holds the
 general-purpose filter compiler (it emits `$N` prepared-statement placeholders
 plus a params list — execute with `DB.execute(sql, params)`). Lint-time
@@ -72,15 +72,15 @@ template context lives in `backend/.sqlfluff`.
 
 - `api/routes/get_routes/get_filters.py` — `GET /load/mapping/{dataset}/filters` — returns hierarchical filter tree
 - `api/routes/get_routes/get_wholedata.py` — full dataset fetch with optional filters
-- `api/routes/post_routes/post_census.py` — Census/ACS-5 data
+- `api/routes/post_routes/post_acs5_db.py` — Census ACS-5 tidy and time-series data (`/load/acs5-db/...`)
 - `api/routes/post_routes/post_zoning.py` — Zoning data
 
 **Response model:** `api/models/response_models.py` — `APIResponse` with `data`, `tableData`, and `metadata` fields.
 
 **Query layer (`query/`):**
 
 - `sql_render.py` — Jinja SQL rendering and the filter compiler (`sql_filter_block`, `compile_where`)
-- `timeseries_db.py` — DuckDB access to census time-series tables
+- `acs5.py` — ACS-5 tidy and time-series queries against `warehouse.duckdb`
 - `production_db.py` — lazy DuckDB connection
 
 **ETL census helpers:** `data_collection/census.py` (`tidy_census`, `split_name_col`).

backend/.sqlfluff

@@ -1,5 +1,5 @@
-# SQL linting for the Jinja SQL templates in query/sql/ and build/sql/.
-# Run from backend/:  uv run sqlfluff lint query/sql build/sql
+# SQL linting for the Jinja SQL templates in query/sql/ and data_cleaning/sql/.
+# Run from backend/:  uv run sqlfluff lint query/sql data_cleaning/sql
 #
 # The [templater:jinja:context] section supplies representative values for the
 # fragments that Python (sql_render.py) injects at runtime, so the linter

backend/api/routes/post_routes/__init__.py

@@ -1,7 +1,6 @@
 from .post_acs5_db import router as post_acs5_router
 from .post_ambulance import router as post_ambulance_router
 from .post_cdc import router as post_cdc_router
-from .post_census import router as post_census_router
 from .post_comparison import router as post_comparison_router
 from .post_export import router as post_export_router
 from .post_flood import router as post_flood_router
@@ -12,7 +11,6 @@
 
 all_post_routers = [
     post_zoning_router,
-    post_census_router,
     post_acs5_router,
     post_qcew_router,
     post_wastewater_router,

backend/data_cleaning/sql/zoning_empty_geom.sql

@@ -28,7 +28,7 @@ gaps AS (
         t.id,
         t.NAME,
         ST_DIFFERENCE(
-            ST_GeomFromWKB(geometry),
+            ST_GeomFromWKB(t.geometry),
             COALESCE(c.geom, ST_GEOMFROMTEXT('POLYGON EMPTY'))
         ) AS geom
     FROM town_boundaries AS t
@@ -41,7 +41,7 @@ exploded AS (
     SELECT
         g.id,
         g.NAME,
-        UNNEST(ST_DUMP(geom), recursive := true)  -- noqa: AL03
+        UNNEST(ST_DUMP(g.geom), recursive := true)  -- noqa: AL03
     FROM gaps AS g
     WHERE NOT ST_ISEMPTY(g.geom)
 ),

backend/data_cleaning/sql/zoning_rules.sql

@@ -9,4 +9,4 @@ FROM (
     INTO
     NAME col_name
     VALUE val
-)
+)

backend/query/sql/acs5/acs5_county_compare.sql

@@ -11,9 +11,9 @@ SELECT
     c.county,
     ST_ASGEOJSON(ST_GeomFromWKB(c.geometry)) AS geometry
 FROM {{ table }} AS a
+-- County rows in the tidy tables carry no geoid (it's only populated for
+-- towns); the county FIPS is what identifies them.
 LEFT JOIN vt_county_lines_geom AS c
-    -- County rows in the tidy tables carry no geoid (it's only populated for
-    -- towns); the county FIPS is what identifies them.
     ON a.county_fips = c.geoid
 
 {{ where_string }}

backend/query/sql/acs5/acs5_timeseries.sql

@@ -1,4 +1,4 @@
 SELECT *
 FROM {{ table }}
 {{ where_string }}
-ORDER BY year
+ORDER BY year

backend/query/sql/ambulance/ambulance_geo_query.sql

@@ -10,7 +10,7 @@ FROM (
             'type', 'Feature',
             'geometry', ST_AsGeoJSON(
                 ST_Simplify(ST_GeomFromWKB(g.geometry), 0.0001)
-                )::JSON,
+            )::JSON,
             'properties', json_object(
                 'Certification Level', i.cert_level,
                 'Acres', ROUND(g.shape__area, 2),
@@ -29,4 +29,4 @@ FROM (
     LEFT JOIN VCGI_ambulanceService_colors AS c
         ON i.cert_level = c.certification_level
     {{ join_filter_block }}
-) AS features
+) AS features

backend/query/sql/cdc/county_places.sql

@@ -1,3 +1,11 @@
+-- Filter PLACES before joining the county lines: the compiled WHERE clause
+-- names columns unqualified, and both tables have county, county_fips and geoid.
+WITH places AS (
+    SELECT *
+    FROM cdc_places_county
+    {{ where_string }}
+)
+
 SELECT
     p.geoid,
     p.measure,
@@ -6,10 +14,6 @@ SELECT
     ROUND(p.natl_pct * 100, 2) AS natl_pct,
     p.county,
     ST_ASGEOJSON(ST_GeomFromWKB(c.geometry)) AS geometry
-FROM cdc_places_county AS p
-LEFT JOIN (
-    SELECT * EXCLUDE(county)
-    FROM vt_county_lines_geom
- ) AS c
+FROM places AS p
+LEFT JOIN vt_county_lines_geom AS c
     ON p.geoid = c.geoid
-{{ where_string }}

backend/query/sql/wastewater/soil_suitability_geo_query.sql

@@ -8,9 +8,9 @@ FROM (
         json_object(
             'type', 'Feature',
             'geometry',
-                ST_AsGeoJSON(
-                    ST_Simplify(g.geometry, 0.0001)
-                )::JSON,
+            ST_AsGeoJSON(
+                ST_Simplify(g.geometry, 0.0001)
+            )::JSON,
             'properties', json_object(
                 'Suitability', i.suitability,
                 'Acres', ROUND(i.acres, 2),
@@ -29,4 +29,4 @@ FROM (
     LEFT JOIN VersoWastewater_soilSuitability_colors AS c
         ON i.suitability = c.soil_suitability
     {{ where_string }}
-) AS features;
+) AS features;

backend/query/sql/zoning/footprints.sql

@@ -1,12 +1,12 @@
 {{ cte_filter_block }}
-SELECT 
+SELECT
     b.object_id,
     b.town,
     b.county,
-    b.geoid, 
+    b.geoid,
     b.name,
     b.height_ft,
     b.building_type,
-    ST_AsGeoJSON(b.geometry) AS geometry,
+    ST_AsGeoJSON(b.geometry) AS geometry
 FROM VCGI_buildingFootprints_geom AS b
 {{ join_filter_block }}

backend/query/sql/zoning/unzoned.sql

@@ -1,6 +1,6 @@
 -- Towns' unzoned areas as a GeoJSON FeatureCollection, drawn as a grey layer
--- underneath the zoning districts. Built by build/sql/zoning_empty_geom.sql,
--- which already drops sliver-sized gaps.
+-- underneath the zoning districts. Built by
+-- data_cleaning/sql/zoning_empty_geom.sql, which already drops sliver-sized gaps.
 --
 -- This layer does not depend on the sidebar filters: it is the same "we have no
 -- data here" backdrop regardless of which districts are being shown.

backend/query/sql_render.py

@@ -4,9 +4,10 @@
 **Description**:
     Jinja SQL rendering plus the general-purpose filter compiler.
 
-    SQL files under `query/sql/` and `build/sql/` are Jinja templates, so they can
-    be linted with sqlfluff (`uv run sqlfluff lint query/sql build/sql` from
-    `backend/`; lint-time template context lives in `backend/.sqlfluff`).
+    SQL files under `query/sql/` and `data_cleaning/sql/` are Jinja templates, so
+    they can be linted with sqlfluff (`uv run sqlfluff lint query/sql
+    data_cleaning/sql` from `backend/`; lint-time template context lives in
+    `backend/.sqlfluff`).
 
     Python compiles FilterSource objects into SQL fragments (filter CTEs, joins,
     and WHERE clauses) and injects them through the Jinja context. This keeps the