feat(eeio): derive y_nab from Adom and scaled_q

bedrock/transform/eeio/derived_cornerstone.py

@@ -46,7 +46,6 @@
     load_2017_V_usa,
     load_2017_value_added_usa,
     load_2017_Ytot_usa,
-    load_summary_Uimp_usa,
 )
 from bedrock.transform.allocation.derived import derive_E_usa
 from bedrock.transform.eeio.cornerstone_bea_intermediates import (
@@ -94,12 +93,11 @@
 )
 from bedrock.utils.math.disaggregation import disaggregate_vector
 from bedrock.utils.math.formulas import (
+    backcompute_y_from_A_and_q,
     compute_q,
     compute_Unorm_matrix,
     compute_Vnorm_matrix,
     compute_x,
-    compute_y_for_national_accounting_balance,
-    compute_y_imp,
 )
 from bedrock.utils.math.handle_negatives import (
     handle_negative_matrix_values,
@@ -127,9 +125,6 @@
     USA_2017_FINAL_DEMAND_IMPORT_CODE,
     USA_2017_FINAL_DEMAND_PERSONAL_CONSUMPTION_EXPENDITURE_CODE,
 )
-from bedrock.utils.taxonomy.bea.v2017_industry_summary import (
-    USA_2017_SUMMARY_INDUSTRY_CODES,
-)
 from bedrock.utils.taxonomy.bea_v2017_to_ceda_v7_helpers import (
     get_bea_v2017_summary_to_cornerstone_corresp_df,
 )
@@ -904,40 +899,17 @@ def derive_cornerstone_Y_and_trade_scaled() -> SingleRegionYtotAndTradeVectorSet
 
 @functools.cache
 def derive_cornerstone_y_nab() -> pd.Series[float]:
-    """Y for national accounting balance, year-scaled."""
-    cfg = get_usa_config()
-    detail_2017 = derive_cornerstone_Ytot_matrix_set()
-
-    y_nab_2017 = compute_y_for_national_accounting_balance(
-        y_tot=detail_2017.ytot,
-        y_imp=compute_y_imp(
-            imports=detail_2017.imports,
-            Uimp=derive_cornerstone_U_set().Uimp,
-        ),
-        exports=detail_2017.exports,
-    )
-
-    summary_Y = derive_summary_Ytot_usa_matrix_set(cfg.usa_io_data_year)
-    y_nab_summary = compute_y_for_national_accounting_balance(
-        y_tot=summary_Y.ytot,
-        y_imp=compute_y_imp(
-            imports=summary_Y.imports,
-            Uimp=load_summary_Uimp_usa(cfg.usa_io_data_year).loc[
-                USA_2017_SUMMARY_INDUSTRY_CODES, USA_2017_SUMMARY_INDUSTRY_CODES
-            ],
-        ),
-        exports=summary_Y.exports,
-    )
+    """National-accounting final demand consistent with scaled ``Adom`` and ``q``.
 
-    y_nab_scaled = _disaggregate_and_inflate_vector(
-        base=y_nab_summary,
-        weight=y_nab_2017,
-        corresp_df=get_bea_v2017_summary_to_cornerstone_corresp_df(),
-        original_year=cfg.usa_io_data_year,
-        target_year=cfg.model_base_year,
-    )
+    Enforces row balance ``q = Adom @ diag(q) + y_nab`` using the same
+    ``derive_cornerstone_Aq_scaled`` object whose ``scaled_q`` is snapshotted.
+    Any future change to ``scaled_q`` in that path propagates to ``y_nab``.
 
-    return handle_negative_vector_values(y_nab_scaled)
+    Negative values are retained so ``q ≈ L_dom @ y_nab`` holds numerically;
+    clipping would break the domestic Leontief identity.
+    """
+    aq = derive_cornerstone_Aq_scaled()
+    return backcompute_y_from_A_and_q(A=aq.Adom, q=aq.scaled_q)
 
 
 def derive_cornerstone_ydom_and_yimp() -> SingleRegionYVectorSet:

bedrock/utils/math/formulas.py

@@ -254,7 +254,7 @@ def backcompute_q_from_L_and_y(
     return (L @ np.diag(y)).sum(axis=1)
 
 
-def backcompute_y_from_q_and_Aq(
+def backcompute_y_from_A_and_q(
     *, A: pd.DataFrame, q: pd.Series[float]
 ) -> pd.Series[float]:
     return q - A.multiply(q, axis=1).sum(axis=1)

bedrock/utils/validation/__tests__/test_eeio_diagnostics.py

@@ -353,28 +353,27 @@ def test_compare_Uset_y_dom_and_q_usa(
 
 @pytest.mark.eeio_integration
 @pytest.mark.parametrize(
-    "modelType, use_domestic",
-    [
-        ("Commodity", False),
-    ],
-)  # TODO: add ("Commodity", True) test and industry parameters when Industry models become available [("Industry", False), ("Industry", True)]
-@pytest.mark.parametrize(
-    "pipeline",
+    "modelType, use_domestic, pipeline",
     [
+        ("Commodity", True, "cornerstone"),
         pytest.param(
-            "ceda",
+            "Commodity",
+            False,
+            "cornerstone",
             marks=pytest.mark.xfail(
-                reason="CEDA: scaled q≠L_total·y_total for ~298 commodity sectors (total Leontief identity).",
+                reason="Cornerstone total L·y still uses ytot/trade, not y_nab.",
             ),
         ),
         pytest.param(
-            "cornerstone",
+            "Commodity",
+            False,
+            "ceda",
             marks=pytest.mark.xfail(
-                reason="Cornerstone: scaled q≠L_total·y_total for 323 sectors; A-matrix vs Y-inflation path mismatch.",
+                reason="CEDA: scaled q≠L_total·y_total for ~298 commodity sectors (total Leontief identity).",
             ),
         ),
     ],
-)
+)  # TODO: add industry parameters when Industry models become available
 def test_compare_output_and_L_y(
     modelType: str,
     use_domestic: bool,
@@ -396,23 +395,17 @@ def test_compare_output_and_L_y(
             y = y_set.ytot + y_set.exports - y_set.imports
             L = formulas.compute_L_matrix(A=Aq.Adom + Aq.Aimp)
     else:
-        # Cornerstone total L·y uses year-scaled Adom+Aimp and scaled_q from
-        # derive_cornerstone_Aq_scaled, compared to L @ y where y = y_tot +
-        # exports − imports from derive_cornerstone_Y_and_trade_scaled (summary-
-        # disaggregated and inflated to model year). 323 of 392 commodities fail
-        # at 1% tolerance—a systemic scale mismatch between the A-matrix scaling
-        # pipeline and the Y/trade inflation path (see zero-weight disaggregation
-        # warning during Y scaling), not an isolated wiring error in this test.
         # Cornerstone scales A and q to model year; CEDA branch stays in 2017 detail.
         Aq = derive_cornerstone_Aq_scaled()
         # Output must match Aq scaling (scaled_q), not derive_cornerstone_q() from V.
         output = Aq.scaled_q if modelType == "Commodity" else derive_cornerstone_x()
         if use_domestic:
-            # Precomputed scaled y_nab (same path as derive_y_for_national_accounting_balance_usa).
+            # y_nab from backcompute_y_from_A_and_q(Adom, scaled_q); unclipped.
             y = derive_cornerstone_y_nab()
             L = formulas.compute_L_matrix(A=Aq.Adom)
         else:
-            # Total y from summary-disaggregated inflated trade, not raw Ytot_matrix_set.
+            # Total L·y still uses y from derive_cornerstone_Y_and_trade_scaled
+            # (summary-disaggregated BEA Y/trade), not IO-balanced y_nab.
             y_trade = derive_cornerstone_Y_and_trade_scaled()
             y = y_trade.ytot + y_trade.exports - y_trade.imports
             L = formulas.compute_L_matrix(A=Aq.Adom + Aq.Aimp)