Stream reform validation simulations

packages/populace-build/src/populace/build/us_runtime/reform_validation.py

@@ -365,14 +365,10 @@ def reform_validation_payload(
     targets = in_sample_targets or {}
     baseline: Any = None
     baseline_totals: dict[tuple[int, str], float] = {}
-    parameter_reform_sims: dict[str, Any] = {}
     obbba_specs = tuple(spec for spec in specs if _is_obbba_spec(spec))
     obbba_pre_baseline_changes = _merged_parameter_changes(obbba_specs)
     obbba_stacked: dict[str, tuple[float, float, float]] | None = None
 
-    def parameter_changes_key(changes: dict[str, Any]) -> str:
-        return json.dumps(changes, sort_keys=True, separators=(",", ":"))
-
     def baseline_total(measure: str, at_period: int) -> float:
         nonlocal baseline
         if baseline is None:
@@ -383,11 +379,8 @@ def baseline_total(measure: str, at_period: int) -> float:
         return baseline_totals[key]
 
     def simulation_for_parameter_changes(changes: dict[str, Any]) -> Any:
-        key = parameter_changes_key(changes)
-        if key not in parameter_reform_sims:
-            reform = None if not changes else _build_parameter_reform(changes)
-            parameter_reform_sims[key] = simulate(reform)  # type: ignore[misc]
-        return parameter_reform_sims[key]
+        reform = None if not changes else _build_parameter_reform(changes)
+        return simulate(reform)  # type: ignore[misc]
 
     def stacked_obbba_effects() -> dict[str, tuple[float, float, float]]:
         """Score the OBBBA provisions *stacked* in their JCX-35-25 order.
@@ -411,9 +404,7 @@ def stacked_obbba_effects() -> dict[str, tuple[float, float, float]]:
             return {}
         measures = {(spec.budget_measure, spec.period) for spec in obbba_specs}
         if len(measures) != 1:
-            return {
-                spec.id: _isolated_obbba_effect(spec) for spec in obbba_specs
-            }
+            return {spec.id: _isolated_obbba_effect(spec) for spec in obbba_specs}
         measure, period = next(iter(measures))
         # state 0: pre-OBBBA (every provision reverted).
         prev_total = _weighted_total(

packages/populace-build/tests/test_reform_validation.py

@@ -213,12 +213,66 @@ def simulate(reform):
     assert rows["obbba_b"]["populace"]["reform_total"] == pytest.approx(960.0)
     assert rows["obbba_b"]["populace"]["budget_effect"] == pytest.approx(60.0)
     # Stacked line effects telescope to the true total OBBBA effect.
-    total = sum(
-        rows[i]["populace"]["budget_effect"] for i in ("obbba_a", "obbba_b")
-    )
+    total = sum(rows[i]["populace"]["budget_effect"] for i in ("obbba_a", "obbba_b"))
     assert total == pytest.approx(960.0 - 1_000.0)
 
 
+def test_obbba_stacked_scoring_releases_intermediate_simulations(monkeypatch):
+    specs = tuple(
+        ReformValidationSpec(
+            id=f"obbba_{name}",
+            name=f"OBBBA {name}",
+            category="OBBBA",
+            in_sample=False,
+            period=2026,
+            jct_score=None,
+            jct_window="FY2026",
+            jct_source="JCX",
+            jct_source_url="",
+            parameter_changes={
+                f"gov.example.{name}": {"2026-01-01.2026-12-31": 0},
+            },
+            effect_direction="baseline_minus_reform",
+        )
+        for name in ("a", "b", "c")
+    )
+    monkeypatch.setattr(
+        reform_validation_module,
+        "_build_parameter_reform",
+        lambda changes: frozenset(changes),
+    )
+
+    live_simulations = 0
+
+    class _TrackedSim(_FakeSim):
+        def __init__(self, total: float) -> None:
+            nonlocal live_simulations
+            assert live_simulations == 0
+            live_simulations += 1
+            super().__init__({"income_tax": total})
+
+        def __del__(self) -> None:
+            nonlocal live_simulations
+            live_simulations -= 1
+
+    def simulate(reform):
+        totals = {
+            frozenset({"gov.example.a", "gov.example.b", "gov.example.c"}): 1000.0,
+            frozenset({"gov.example.b", "gov.example.c"}): 900.0,
+            frozenset({"gov.example.c"}): 960.0,
+            None: 970.0,
+        }
+        return _TrackedSim(totals[reform])
+
+    payload = reform_validation_payload(specs, period=2026, simulate=simulate)
+    assert [row["id"] for row in payload["reforms"]] == [
+        "obbba_a",
+        "obbba_b",
+        "obbba_c",
+    ]
+    assert live_simulations == 0
+
+
 def test_shipped_obbba_config_is_out_of_sample_counterfactual():
     specs = out_of_sample_reform_specs(period=2026)
     assert {s.id for s in specs} >= {"obbba_no_tax_on_tips", "obbba_no_tax_on_overtime"}
@@ -247,9 +301,9 @@ def test_itemized_benefit_limit_counterfactual_keeps_pease():
     )
     paths = set(spec.parameter_changes or {})
     assert paths == {"gov.irs.deductions.itemized.limitation.obbb.applies"}, paths
-    assert (
-        "gov.irs.deductions.itemized.limitation.applies" not in paths
-    ), "must not disable the whole limitation (drops present-law Pease)"
+    assert "gov.irs.deductions.itemized.limitation.applies" not in paths, (
+        "must not disable the whole limitation (drops present-law Pease)"
+    )
 
 
 def test_shipped_tax_expenditure_specs_neutralize_big_provisions():