sql-code-graph 1.34.2__tar.gz → 1.34.4__tar.gz

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sql-code-graph
-Version: 1.34.2
+Version: 1.34.4
 Summary: SQL code graph analyzer and lineage tracer
 Project-URL: Homepage, https://github.com/Warhorze/sql-code-graph
 Project-URL: Repository, https://github.com/Warhorze/sql-code-graph

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/docs/cli.md

@@ -438,6 +438,13 @@ sqlcg analyze impact [OPTIONS] TABLE
 
 Show all queries impacted by a table.
 
+One-hop "who reads this table" — the union of direct SELECTS_FROM consumers
+and consumers that read the table only via COLUMN_LINEAGE / STAR_SOURCE
+(CTE-wrapped / promoted-intermediate reads, e.g. the live DWH node
+``ba.all_rows_in_selection``). Bug #6 fix: brings ``impact`` to the same
+one-hop consumer completeness ``find_table_usages`` already has. Stays
+one-hop (NOT transitive — that is ``downstream``'s job).
+
 ### Options
 
 | Option | Type | Required | Repeatable | Default | Description |

sql_code_graph-1.34.4/plan/metrics/pr4_star_promote_after.json

@@ -0,0 +1,31 @@
+{"total_calls": 1231, "last_7d_calls": 1231, "index_runs": 5, "feedback_tp": 0, 
+"feedback_total": 0, "top_tools": [{"name": "execute_sql", "count": 412}, 
+{"name": "trace_column_lineage", "count": 326}, {"name": 
+"list_dialects_and_repos", "count": 106}], "execute_sql_ratio": 0.33, 
+"parse_quality": {"sqlglot": 5166, "scripting": 20, "dynamic_sql": 3}, 
+"coverage": {"catalogued_tables": 5800, "total_tables": 6581, "good_edges": 
+52673, "total_edges": 60821, "phantom_edges": 12241, "blindspot_tables": 628, 
+"good_edges_strict": 48140, "edge_health_strict_pct": 79.15, 
+"good_edges_scoped": 37460, "total_edges_scoped": 38009, 
+"edge_health_scoped_pct": 98.56, "phantom_confirmed": 11701, 
+"phantom_contradicted": 352, "phantom_unverified": 188, "top_blindspot_tables": 
+[{"table": "ba.wtfa_kpi_datum_klant", "bad_edges": 70}, {"table": 
+"ba.wtfs_voorraad_dagstand", "bad_edges": 55}, {"table": 
+"ba.wtfs_voorraad_zondagstand", "bad_edges": 55}, {"table": 
+"da.rtgmd_postnl_customer_usage", "bad_edges": 43}, {"table": 
+"da.rtgmd_postnl_athome_3s", "bad_edges": 33}, {"table": 
+"wtfe_kpi_voorraad_artikel_voorraadlocatie", "bad_edges": 32}, {"table": 
+"da.rtgmd_postnl_facturen", "bad_edges": 30}, {"table": "da.hthyb_pages", 
+"bad_edges": 23}, {"table": "da.rtgmd_dhb_extra_kosten", "bad_edges": 19}, 
+{"table": "da.hthyb_page_brandplacement", "bad_edges": 17}], 
+"blindspot_tables_for_80pct": 17, "files_indexed": 1335, "indexed_sha": 
+"fdf1b551a34601a6cf3ce1c8b9f76e27ce2753e6", "db_path": "/tmp/pr4_after.duckdb", 
+"index_timestamp": 1781457525.4581766, "degraded_parse_total": 5189, 
+"degraded_parse_queries": 20, "degraded_parse_by_dir": {"etl": {"degraded": 1, 
+"total": 2776}, "ddl": {"degraded": 19, "total": 2413}}, 
+"zero_edge_write_queries": 1063, "total_write_queries": 2352, 
+"cte_key_collisions": 0, "rescuable_unqualified_edges": 125, 
+"info_schema_has_column_rows": 100193, "cte_source_gap_writes": 0, 
+"resolvable_write_col_edges": 36936, "transitive_col_edges": 4184, 
+"qualify_failed_statements": 9, "tool_version_stale": false, 
+"indexed_tool_version": "1.34.2", "running_tool_version": "1.34.2"}}

sql_code_graph-1.34.4/plan/metrics/pr4_star_promote_before.json

@@ -0,0 +1,31 @@
+{"total_calls": 1231, "last_7d_calls": 1231, "index_runs": 5, "feedback_tp": 0, 
+"feedback_total": 0, "top_tools": [{"name": "execute_sql", "count": 412}, 
+{"name": "trace_column_lineage", "count": 326}, {"name": 
+"list_dialects_and_repos", "count": 106}], "execute_sql_ratio": 0.33, 
+"parse_quality": {"sqlglot": 5166, "scripting": 20, "dynamic_sql": 3}, 
+"coverage": {"catalogued_tables": 5796, "total_tables": 6581, "good_edges": 
+51336, "total_edges": 59484, "phantom_edges": 12241, "blindspot_tables": 628, 
+"good_edges_strict": 46495, "edge_health_strict_pct": 78.16, 
+"good_edges_scoped": 36842, "total_edges_scoped": 37619, 
+"edge_health_scoped_pct": 97.93, "phantom_confirmed": 11701, 
+"phantom_contradicted": 352, "phantom_unverified": 188, "top_blindspot_tables": 
+[{"table": "ia_analytics.ba_wtfa_artikel_formule_dagomzet", "bad_edges": 217}, 
+{"table": "ba.wtfa_kpi_datum_klant", "bad_edges": 70}, {"table": 
+"ba.wtfs_voorraad_dagstand", "bad_edges": 55}, {"table": 
+"ba.wtfs_voorraad_zondagstand", "bad_edges": 55}, {"table": 
+"da.rtgmd_postnl_customer_usage", "bad_edges": 43}, {"table": 
+"da.rtgmd_postnl_athome_3s", "bad_edges": 33}, {"table": 
+"wtfe_kpi_voorraad_artikel_voorraadlocatie", "bad_edges": 32}, {"table": 
+"da.rtgmd_postnl_facturen", "bad_edges": 30}, {"table": "da.hthyb_pages", 
+"bad_edges": 23}, {"table": "da.rtgmd_dhb_extra_kosten", "bad_edges": 19}], 
+"blindspot_tables_for_80pct": 15, "files_indexed": 1335, "indexed_sha": 
+"fdf1b551a34601a6cf3ce1c8b9f76e27ce2753e6", "db_path": "/tmp/pr4_before.duckdb",
+"index_timestamp": 1781457313.671681, "degraded_parse_total": 5189, 
+"degraded_parse_queries": 20, "degraded_parse_by_dir": {"etl": {"degraded": 1, 
+"total": 2776}, "ddl": {"degraded": 19, "total": 2413}}, 
+"zero_edge_write_queries": 1069, "total_write_queries": 2352, 
+"cte_key_collisions": 0, "rescuable_unqualified_edges": 125, 
+"info_schema_has_column_rows": 110257, "cte_source_gap_writes": 0, 
+"resolvable_write_col_edges": 35759, "transitive_col_edges": 4178, 
+"qualify_failed_statements": 9, "tool_version_stale": false, 
+"indexed_tool_version": "1.34.0", "running_tool_version": "1.34.0"}}

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "sql-code-graph"
-version = "1.34.2"
+version = "1.34.4"
 description = "SQL code graph analyzer and lineage tracer"
 readme = "README.md"
 requires-python = ">=3.12"

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/src/sqlcg/__init__.py

@@ -1,5 +1,5 @@
 """SQL Code Graph - SQL lineage and dependency analysis tool."""
 
-__version__ = "1.34.2"
+__version__ = "1.34.4"
 
 __all__ = ["__version__"]

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/src/sqlcg/cli/commands/analyze.py

@@ -8,7 +8,10 @@ import typer
 from rich.console import Console
 from rich.table import Table
 
-from sqlcg.core.queries import GET_TABLE_EXTERNAL_CONSUMERS_QUERY
+from sqlcg.core.queries import (
+    GET_TABLE_EXTERNAL_CONSUMERS_QUERY,
+    IMPACT_CONSUMERS_VIA_LINEAGE_QUERY,
+)
 from sqlcg.server.read_client import resolved_repo_root, run_read_routed
 
 if TYPE_CHECKING:
@@ -312,8 +315,16 @@ def impact(  # noqa: B008
     table: str = typer.Argument(..., help="Table name to analyze"),  # noqa: B008
     raw: bool = typer.Option(False, "--raw", help="Disable noise filtering on results"),  # noqa: B008
 ) -> None:
-    """Show all queries impacted by a table."""
-    results = run_read_routed(
+    """Show all queries impacted by a table.
+
+    One-hop "who reads this table" — the union of direct SELECTS_FROM consumers
+    and consumers that read the table only via COLUMN_LINEAGE / STAR_SOURCE
+    (CTE-wrapped / promoted-intermediate reads, e.g. the live DWH node
+    ``ba.all_rows_in_selection``). Bug #6 fix: brings ``impact`` to the same
+    one-hop consumer completeness ``find_table_usages`` already has. Stays
+    one-hop (NOT transitive — that is ``downstream``'s job).
+    """
+    direct = run_read_routed(
         "SELECT DISTINCT q.id AS id, q.kind AS kind, q.target_table AS target"
         ' FROM "SqlTable" t'
         ' JOIN "SELECTS_FROM" sf ON sf.dst_key = t.qualified'
@@ -321,6 +332,17 @@ def impact(  # noqa: B008
         " WHERE t.qualified = ? LIMIT 100",
         {"t": table},
     )
+    # Consumers reachable only via COLUMN_LINEAGE / STAR_SOURCE. Two "?" both
+    # bound to the qualified table name: _params_list returns list(values()) in
+    # insertion order, so the ordered dict {"t":..., "t2":...} binds correctly.
+    via = run_read_routed(
+        IMPACT_CONSUMERS_VIA_LINEAGE_QUERY,
+        {"t": table, "t2": table},
+    )
+    seen = {r["id"] for r in direct}
+    results = direct + [r for r in via if r["id"] not in seen]
+    # Preserve the existing 100-row bound on the merged consumer set.
+    results = results[:100]
     if not raw:
         from sqlcg.server.noise_filter import NoiseFilter
 

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/src/sqlcg/core/queries.py

@@ -72,3 +72,4 @@ GET_TABLES_EXTERNAL_CONSUMERS_BATCH_QUERY = _Q["GET_TABLES_EXTERNAL_CONSUMERS_BA
 COUNT_EXTERNAL_CONSUMERS_QUERY = _Q["COUNT_EXTERNAL_CONSUMERS"]
 GET_TABLE_READS_ADJACENCY_QUERY = _Q["GET_TABLE_READS_ADJACENCY"]
 GET_PRODUCER_TABLES_QUERY = _Q["GET_PRODUCER_TABLES"]
+IMPACT_CONSUMERS_VIA_LINEAGE_QUERY = _Q["IMPACT_CONSUMERS_VIA_LINEAGE"]

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/src/sqlcg/core/queries.sql

@@ -413,3 +413,26 @@ WHERE t.name = ?
 SELECT DISTINCT target_table AS producer_table
 FROM "SqlQuery"
 WHERE target_table <> '' AND target_table IS NOT NULL
+
+-- IMPACT_CONSUMERS_VIA_LINEAGE
+-- Bug #6: consumers that read a table only via COLUMN_LINEAGE or STAR_SOURCE
+-- (CTE-wrapped / promoted-intermediate reads that emit no top-level SELECTS_FROM
+-- edge, e.g. the live DWH cross-file-promoted node ba.all_rows_in_selection:
+-- kind='table', non-namespaced, 0 SELECTS_FROM in-edges, 55 COLUMN_LINEAGE
+-- out-edges). Returns the same (id, kind, target) shape as the SELECTS_FROM-only
+-- impact query in analyze.py:impact() so the two result sets merge + dedup on q.id.
+-- Both branches key on the QUALIFIED table name (impact() keys on t.qualified):
+-- COLUMN_LINEAGE branch -> src.table_qualified; STAR_SOURCE branch -> ss.dst_key.
+-- The STAR_SOURCE branch closes the STAR-only-consumer gap (FIND_TABLE_USAGES
+-- itself still lacks it — out of scope, separate follow-up).
+-- params: [table_qualified, table_qualified]  (same value bound to BOTH ?)
+SELECT DISTINCT q.id AS id, q.kind AS kind, q.target_table AS target
+FROM "SqlColumn" src
+JOIN "COLUMN_LINEAGE" cl ON cl.src_key = src.id
+JOIN "SqlQuery" q ON q.id = cl.query_id
+WHERE src.table_qualified = ?
+UNION
+SELECT DISTINCT q.id AS id, q.kind AS kind, q.target_table AS target
+FROM "STAR_SOURCE" ss
+JOIN "SqlQuery" q ON q.id = ss.src_key
+WHERE ss.dst_key = ?

{sql_code_graph-1.34.2 → sql_code_graph-1.34.4}/src/sqlcg/indexer/indexer.py

@@ -896,6 +896,20 @@ class Indexer:
         if profile:
             _t_catalog_end = time.perf_counter()
 
+        # Bug #4 (PR-4): re-run the star expansion AFTER the catalog is applied.
+        # The first pass at the top of this block expands STAR_SOURCE rows whose
+        # leaf table already had HAS_COLUMN rows (parsed DDL). A temp built as
+        # `CREATE TEMP t AS SELECT * FROM src` where `src` is catalogued ONLY via
+        # information_schema has NO columns at the first pass — its HAS_COLUMN rows
+        # arrive with the catalog above — so the temp (and any promote out of it)
+        # produced zero edges. The star-expansion SQL is idempotent
+        # (INSERT OR REPLACE) and chains both levels (src -> temp -> persisted) in a
+        # single pass once the leaf columns exist, so one extra invocation here is
+        # the full fix. Only re-run when the catalog actually added columns; a repo
+        # with no configured catalog pays nothing.
+        if catalog_result.get("columns_loaded", 0) > 0:
+            star_edges_expanded = self._expand_star_sources(db)
+
         # Post-catalog: resolve JOIN_COL_RESOLVE markers against the full HAS_COLUMN
         # catalog (Bug #5, PR-5 option b). MUST run after the catalog is applied so
         # information_schema columns are present; the suppressed sqlglot mis-bind

sql_code_graph-1.34.4/tests/integration/test_impact_consumer_parity.py

@@ -0,0 +1,350 @@
+"""Bug #6 gate: ``analyze impact <T>`` must not under-report consumers that
+``downstream`` / ``empty-impact`` surface.
+
+Governing plan: [bugfix_impact_consumer_union]
+([plan doc](plan/sprints/bugfix_impact_consumer_union.md), branch
+``plan/impact-consumer-union``).
+
+GATE OUTCOME: REPRODUCED at DWH scale. The bug is real: the live DWH node
+``ba.all_rows_in_selection`` is a *cross-file qualified-name-promoted* inner-CTE
+intermediate — ``kind='table'``, **non-namespaced** (no ``::``), **0
+SELECTS_FROM in-edges**, **55 COLUMN_LINEAGE out-edges**. ``analyze impact``
+joined only SELECTS_FROM, so it returned "No results" for that node while
+``empty-impact`` / ``downstream`` (which follow COLUMN_LINEAGE) found the real
+consumer.
+
+Why the earlier "NOT REPRODUCED" claim was wrong (PR #156): the disproven
+hypothesis was that *only namespaced CTE intermediates* (``kind='cte'``) ever
+have a COLUMN_LINEAGE/STAR read without a SELECTS_FROM edge. A from-scratch
+in-memory index of a nested-CTE INSERT…SELECT does indeed only emit a namespaced
+``kind='cte'`` gap node — but the live divergence arises from a **cross-file**
+promotion effect (a source-ref node defaults to ``kind='table'`` and is only
+upgraded to ``kind='cte'`` when the *same* qualified name is later confirmed a
+CTE destination in the same pass; the single-file case namespaces+confirms →
+``cte``). A unit/indexer fixture cannot reproduce the non-namespaced
+``kind='table'`` shape, so the authoritative repro below is **graph-layer
+seeded**: it writes the exact divergence shape directly, with no indexer. The
+nested-CTE indexer fixture is kept only as a *no-regression* case (it proves
+``impact`` on a physical source is unchanged), NOT as the bug repro.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from sqlcg.core.duckdb_backend import DuckDBBackend
+from sqlcg.core.queries import IMPACT_CONSUMERS_VIA_LINEAGE_QUERY
+from sqlcg.indexer.indexer import Indexer
+
+# The exact query the OLD `analyze impact <table>` ran (SELECTS_FROM only): a
+# single hop SqlTable -> SELECTS_FROM -> SqlQuery. Kept here as the negative
+# control — the seeded repro asserts this returns ∅ for the promoted node.
+_IMPACT_SQL = (
+    "SELECT DISTINCT q.id AS id, q.kind AS kind, q.target_table AS target"
+    ' FROM "SqlTable" t'
+    ' JOIN "SELECTS_FROM" sf ON sf.dst_key = t.qualified'
+    ' JOIN "SqlQuery" q ON q.id = sf.src_key'
+    " WHERE t.qualified = ? LIMIT 100"
+)
+
+# DDL shared by all parity cases.
+_DDL = (
+    "CREATE TABLE ba.src (col1 INT, col2 INT);\n"
+    "CREATE TABLE ba.tgt (col1 INT, col2 INT);\n"
+    "CREATE TABLE ba.main (col1 INT, id INT);\n"
+    "CREATE TABLE ba.lookup (x INT);\n"
+    "CREATE TABLE ba.filt (id INT);\n"
+)
+
+# (case-id, consumer-SQL, physical-target-table) — every shape the validation
+# sprint named, plus subquery / CTE / view / star variants.
+IMPACT_PARITY_CASES = [
+    ("plain_insert_select", "INSERT INTO ba.tgt SELECT col1, col2 FROM ba.src;\n", "ba.src"),
+    ("insert_select_star", "INSERT INTO ba.tgt SELECT * FROM ba.src;\n", "ba.src"),
+    (
+        "cte_wrapped_insert",
+        "INSERT INTO ba.tgt WITH c AS (SELECT col1, col2 FROM ba.src) SELECT col1, col2 FROM c;\n",
+        "ba.src",
+    ),
+    (
+        "cte_wrapped_create_view",
+        "CREATE VIEW ba.v AS WITH c AS (SELECT col1 FROM ba.src) SELECT col1 FROM c;\n",
+        "ba.src",
+    ),
+    (
+        "cte_wrapped_star",
+        "INSERT INTO ba.tgt WITH c AS (SELECT * FROM ba.src) SELECT * FROM c;\n",
+        "ba.src",
+    ),
+    (
+        "nested_cte_source_deep",
+        "INSERT INTO ba.tgt WITH a AS (SELECT col1 FROM ba.src), "
+        "b AS (SELECT col1 FROM a) SELECT col1 FROM b;\n",
+        "ba.src",
+    ),
+    (
+        "scalar_subquery_source",
+        "INSERT INTO ba.tgt SELECT col1, (SELECT max(x) FROM ba.lookup) FROM ba.main;\n",
+        "ba.lookup",
+    ),
+    (
+        "where_in_subquery_source",
+        "INSERT INTO ba.tgt SELECT col1 FROM ba.main WHERE id IN (SELECT id FROM ba.filt);\n",
+        "ba.filt",
+    ),
+]
+
+
+def _index(tmp_path, files: dict[str, str]) -> DuckDBBackend:
+    for name, content in files.items():
+        (tmp_path / name).write_text(content)
+    db = DuckDBBackend(":memory:")
+    db.init_schema()
+    Indexer().index_repo(tmp_path, dialect=None, db=db, use_git=False)
+    return db
+
+
+def _impact_consumers(db: DuckDBBackend, table: str) -> set[str]:
+    """OLD `impact`: direct SELECTS_FROM consumers only (the under-reporting query)."""
+    return {r["id"] for r in db.run_read(_IMPACT_SQL, {"t": table})}
+
+
+def _union_consumers(db: DuckDBBackend, table: str) -> set[str]:
+    """The full consumer set across SELECTS_FROM ∪ STAR_SOURCE ∪ COLUMN_LINEAGE.
+
+    This is the parity yardstick the fixed ``impact`` reaches.
+    """
+    sf = db.run_read(
+        'SELECT DISTINCT src_key AS q FROM "SELECTS_FROM" WHERE dst_key = ?', {"t": table}
+    )
+    ss = db.run_read(
+        'SELECT DISTINCT src_key AS q FROM "STAR_SOURCE" WHERE dst_key = ?', {"t": table}
+    )
+    cl = db.run_read(
+        'SELECT DISTINCT cl.query_id AS q FROM "COLUMN_LINEAGE" cl '
+        'JOIN "SqlColumn" s ON s.id = cl.src_key WHERE s.table_qualified = ?',
+        {"t": table},
+    )
+    return {r["q"] for r in sf} | {r["q"] for r in ss} | {r["q"] for r in cl}
+
+
+def _merged_impact_consumers(db: DuckDBBackend, table: str) -> list[dict]:
+    """The NEW `impact` consumer set: direct (SELECTS_FROM) merged with via-lineage
+    (COLUMN_LINEAGE ∪ STAR_SOURCE), deduped on q.id — exactly what
+    ``analyze.py:impact()`` builds. Two ``?`` both bound to the qualified name via
+    an ordered 2-entry dict (``_params_list`` returns ``list(values())``).
+    """
+    direct = db.run_read(_IMPACT_SQL, {"t": table})
+    via = db.run_read(IMPACT_CONSUMERS_VIA_LINEAGE_QUERY, {"t": table, "t2": table})
+    seen = {r["id"] for r in direct}
+    return (direct + [r for r in via if r["id"] not in seen])[:100]
+
+
+def _seed_promoted_intermediate(db: DuckDBBackend) -> tuple[str, str]:
+    """Seed the live DWH divergence shape DIRECTLY (no indexer).
+
+    Writes:
+      * a non-namespaced ``kind='table'`` SqlTable (``ba.promoted_intermediate``);
+      * a SqlColumn whose ``table_qualified`` is that table;
+      * a consumer SqlQuery (INSERT INTO ba.consumer);
+      * a COLUMN_LINEAGE edge column -> consumer-column with ``query_id`` = consumer;
+      * NO SELECTS_FROM in-edge to the seeded table.
+
+    Returns ``(table_qualified, consumer_query_id)``. This is the only shape that
+    reproduces Bug #6 (cross-file promotion); an indexer cannot emit it.
+    """
+    table = "ba.promoted_intermediate"
+    consumer_qid = "q:ba.consumer:0"
+    src_col = f"{table}.dn_functie_naam"
+    dst_col = "ba.consumer.dn_functie_naam"
+
+    db.upsert_node(
+        "SqlTable",
+        table,
+        {"catalog": None, "db": "ba", "name": "promoted_intermediate", "kind": "table"},
+    )
+    db.upsert_node(
+        "SqlColumn",
+        src_col,
+        {
+            "catalog": None,
+            "db": "ba",
+            "table_name": "promoted_intermediate",
+            "col_name": "dn_functie_naam",
+            "table_qualified": table,
+        },
+    )
+    db.upsert_node(
+        "SqlColumn",
+        dst_col,
+        {
+            "catalog": None,
+            "db": "ba",
+            "table_name": "consumer",
+            "col_name": "dn_functie_naam",
+            "table_qualified": "ba.consumer",
+        },
+    )
+    db.upsert_node(
+        "SqlQuery",
+        consumer_qid,
+        {
+            "file_path": "consumer.sql",
+            "statement_index": 0,
+            "sql": "INSERT INTO ba.consumer SELECT dn_functie_naam FROM ba.promoted_intermediate",
+            "kind": "INSERT_SELECT",
+            "target_table": "ba.consumer",
+            "parse_failed": False,
+            "start_line": 1,
+        },
+    )
+    db.upsert_edge(
+        "SqlColumn",
+        src_col,
+        "SqlColumn",
+        dst_col,
+        "COLUMN_LINEAGE",
+        {"transform": "CTE_PROJECTION", "confidence": 1.0, "query_id": consumer_qid},
+    )
+    return table, consumer_qid
+
+
+@pytest.mark.parametrize(
+    ("case_id", "consumer_sql", "target"),
+    IMPACT_PARITY_CASES,
+    ids=[c[0] for c in IMPACT_PARITY_CASES],
+)
+def test_impact_matches_full_consumer_set_for_physical_target(
+    tmp_path, case_id, consumer_sql, target
+):
+    """`analyze impact <physical table>` covers the full SELECTS_FROM ∪ STAR ∪
+    COLUMN_LINEAGE consumer union (the union is a superset, never drops the
+    physical consumer).
+
+    Guards the union fix's superset property
+    ([plan doc](plan/sprints/bugfix_impact_consumer_union.md), A3): for a physical
+    table the merged ``impact`` set equals the full union — it neither under-reports
+    nor adds spurious rows.
+    """
+    db = _index(tmp_path, {"ddl.sql": _DDL, "etl.sql": consumer_sql})
+    try:
+        merged = {r["id"] for r in _merged_impact_consumers(db, target)}
+        union = _union_consumers(db, target)
+        assert merged, f"[{case_id}] impact found no consumer for {target}"
+        # Merged impact must equal the full union for a physical target.
+        assert union - merged == set(), (
+            f"[{case_id}] merged impact UNDER-reports for {target}: "
+            f"union has {sorted(union)}, merged has {sorted(merged)}, "
+            f"missed {sorted(union - merged)}"
+        )
+        assert merged - union == set(), (
+            f"[{case_id}] merged impact OVER-reports for {target}: extra {sorted(merged - union)}"
+        )
+    finally:
+        db.close()
+
+
+def test_impact_surfaces_lineage_only_consumer_of_promoted_intermediate(tmp_path):
+    """The fixed ``impact`` surfaces a consumer reachable ONLY via COLUMN_LINEAGE
+    (no SELECTS_FROM edge) — the live ``ba.all_rows_in_selection`` shape.
+
+    Authoritative Bug #6 repro, GRAPH-LAYER SEEDED
+    ([plan doc](plan/sprints/bugfix_impact_consumer_union.md), A1). FAILS pre-fix
+    (the OLD SELECTS_FROM-only query returns ∅), PASSES post-fix (the union
+    surfaces the consumer). The negative control (``_impact_consumers`` == ∅) is
+    asserted in the same test so the repro cannot silently pass for the wrong
+    reason.
+    """
+    db = DuckDBBackend(":memory:")
+    db.init_schema()
+    try:
+        table, consumer_qid = _seed_promoted_intermediate(db)
+
+        # Negative control: the OLD SELECTS_FROM-only query finds nothing (the bug).
+        old = _impact_consumers(db, table)
+        assert old == set(), (
+            f"negative control broken: the SELECTS_FROM-only impact query should "
+            f"return ∅ for a promoted intermediate with no SELECTS_FROM in-edge, "
+            f"got {sorted(old)}"
+        )
+
+        # The fix: the merged (direct ∪ via-lineage) impact surfaces the consumer.
+        merged = {r["id"] for r in _merged_impact_consumers(db, table)}
+        assert consumer_qid in merged, (
+            f"fixed impact must surface the lineage-only consumer {consumer_qid}; "
+            f"got {sorted(merged)}"
+        )
+    finally:
+        db.close()
+
+
+def test_impact_physical_source_unchanged_by_union(tmp_path):
+    """``impact`` on a physical source of a nested-CTE INSERT…SELECT is UNCHANGED
+    by the union (direct == merged).
+
+    No-regression case ([plan doc](plan/sprints/bugfix_impact_consumer_union.md),
+    A2). This nested-CTE indexer fixture does NOT reproduce Bug #6 (it emits only a
+    namespaced ``kind='cte'`` gap node, and ``ba.src`` already has a SELECTS_FROM
+    edge) — the live divergence needs cross-file promotion, covered by the seeded
+    test above. Here it proves the union adds/drops nothing for a physical source.
+    """
+    db = _index(
+        tmp_path,
+        {
+            "ddl.sql": "CREATE TABLE ba.src (col1 INT, col2 INT);\n"
+            "CREATE TABLE ba.tgt (col1 INT, col2 INT);\n",
+            "etl.sql": "INSERT INTO ba.tgt WITH outer_c AS ("
+            "  WITH all_rows AS (SELECT col1, col2 FROM ba.src) SELECT col1, col2 FROM all_rows"
+            ") SELECT col1, col2 FROM outer_c;\n",
+        },
+    )
+    try:
+        direct = _impact_consumers(db, "ba.src")
+        merged = {r["id"] for r in _merged_impact_consumers(db, "ba.src")}
+        assert direct, "physical source ba.src must have a direct SELECTS_FROM consumer"
+        assert direct == merged, (
+            f"union changed the consumer set for physical source ba.src: "
+            f"direct {sorted(direct)} != merged {sorted(merged)}"
+        )
+    finally:
+        db.close()
+
+
+def test_impact_dedups_consumer_reachable_via_both_edges(tmp_path):
+    """A consumer reachable via BOTH SELECTS_FROM and COLUMN_LINEAGE is listed
+    exactly once (dedup on q.id).
+
+    Guards A4 ([plan doc](plan/sprints/bugfix_impact_consumer_union.md)): the merge
+    must not double-count. A plain ``INSERT … SELECT … FROM ba.src`` yields a
+    consumer reachable both ways; assert it appears once.
+    """
+    db = _index(
+        tmp_path,
+        {"ddl.sql": _DDL, "etl.sql": "INSERT INTO ba.tgt SELECT col1 FROM ba.src;\n"},
+    )
+    try:
+        rows = _merged_impact_consumers(db, "ba.src")
+        ids = [r["id"] for r in rows]
+        assert ids, "expected at least one consumer for ba.src"
+        assert len(ids) == len(set(ids)), f"duplicate consumer rows in impact: {ids}"
+    finally:
+        db.close()
+
+
+def test_impact_result_schema_has_id_and_kind(tmp_path):
+    """The merged ``impact`` rows still expose ``id`` and ``kind`` keys (output
+    schema unchanged).
+
+    Guards A5 ([plan doc](plan/sprints/bugfix_impact_consumer_union.md)).
+    """
+    db = _index(
+        tmp_path,
+        {"ddl.sql": _DDL, "etl.sql": "INSERT INTO ba.tgt SELECT col1 FROM ba.src;\n"},
+    )
+    try:
+        rows = _merged_impact_consumers(db, "ba.src")
+        assert rows, "expected at least one consumer row"
+        for r in rows:
+            assert "id" in r and "kind" in r, f"missing id/kind in row keys: {sorted(r)}"
+    finally:
+        db.close()