Measure a branch's behavioural responses against baseline policy, with its own cache

changelog.d/fix-lsr-branch-baseline-measurements.fixed.md

@@ -0,0 +1,3 @@
+Measure a branch's or clone's labor-supply and capital-gains responses against baseline policy applied to its own inputs, and give each simulation its own measurement cache. A branch of a reform simulation made before anything was calculated (and, on policyengine-core before #587, such a clone) measured against a new branch under the reform policy and reported no response, and a clone given other inputs reused its source's measurements.
+Preserve zero responses without a reform, make capital-gains marginal tax rates independent of calculation order while reusing household measurements, and discard incompatible holders when applying baseline policy to measurement branches.
+Forget deleted pre-response input records on Core 3.32.8 so fully calculated clones remeasure changed earnings, while retaining explicit overrides and the baseline household's inputs.

policyengine_us/spm.py

@@ -83,6 +83,12 @@
 
 COUNTY_FIPS_PATTERN = re.compile(r"[0-9]{5}")
 
+# Where a simulation keeps its behavioural-response measurements, by period
+# (``variables/gov/simulation/behavioral_response_measurements.py``). They are
+# cached formula output kept outside the holders, so the simulation's clone,
+# branch and invalidation paths below handle them explicitly.
+BEHAVIORAL_RESPONSE_CACHE_ATTR = "_behavioral_response_measurements"
+
 COUNTY_INPUT_FIX = (
     'send county_fips as a five-digit string (for example "06037"), or select '
     'geography_kind="national" in the spm configuration'
@@ -711,7 +717,16 @@ def _isolate_parameter_tracing(self):
             branch._isolate_parameter_tracing()
 
     def get_branch(self, name="branch", clone_system=False):
+        creating = name != self.branch_name and name not in self.branches
         branch = super().get_branch(name, clone_system)
+        if creating:
+            # A branch starts from what this simulation has calculated, its
+            # behavioural-response measurements included, but keeps its own
+            # copy (``clone`` emptied it): what either measures afterwards, the
+            # other does not see.
+            measurements = self.__dict__.get(BEHAVIORAL_RESPONSE_CACHE_ATTR)
+            if measurements is not None:
+                branch.__dict__[BEHAVIORAL_RESPONSE_CACHE_ATTR] = dict(measurements)
         # Core names the branch and hands it this simulation's tracer after
         # cloning, so re-prime the branch's root with what it ended up holding.
         branch._isolate_parameter_tracing()
@@ -908,8 +923,20 @@ def clone(self, debug=False, trace=False, clone_tax_benefit_system=True):
         cloned._rebind_holders()
         self._rebind_method_aliases(cloned)
         cloned._isolate_parameter_tracing()
+        # Core copies the instance dictionary by reference, which shared this
+        # simulation's measurements dictionary with the clone: a clone given
+        # other inputs reused them, and what it measured reached this
+        # simulation. A clone measures its own (``get_branch`` seeds a branch).
+        cloned.__dict__.pop(BEHAVIORAL_RESPONSE_CACHE_ATTR, None)
         return cloned
 
+    def _invalidate_all_caches(self):
+        # Core purges cached formula output here (``apply_reform``,
+        # ``subsample``); the measurements are formula output kept outside the
+        # holders.
+        self.__dict__.pop(BEHAVIORAL_RESPONSE_CACHE_ATTR, None)
+        super()._invalidate_all_caches()
+
     def _rebind_method_aliases(self, cloned):
         """Point copied bound-method aliases at the clone, not the original.
 

policyengine_us/system.py

@@ -35,6 +35,7 @@
     uprate_parameters,
 )
 from .tools.default_uprating import add_default_uprating
+from .tools.period_branch import delete_input_values
 from .tools.section_911 import validate_section_911_batch_inputs
 from .tools.per_capita_uprating import (
     add_per_capita_parameters_for_parameter_uprating,
@@ -264,27 +265,27 @@ def __init__(self, *args, **kwargs):
         for known_period in employment_income.get_known_periods():
             array = employment_income.get_array(known_period)
             self.set_input("employment_income_before_lsr", known_period, array)
-            employment_income.delete_arrays(known_period)
+            delete_input_values(employment_income, known_period)
 
         self_employment_income = self.get_holder("self_employment_income")
         for known_period in self_employment_income.get_known_periods():
             array = self_employment_income.get_array(known_period)
             self.set_input("self_employment_income_before_lsr", known_period, array)
-            self_employment_income.delete_arrays(known_period)
+            delete_input_values(self_employment_income, known_period)
 
         sstb_self_employment_income = self.get_holder("sstb_self_employment_income")
         for known_period in sstb_self_employment_income.get_known_periods():
             array = sstb_self_employment_income.get_array(known_period)
             self.set_input(
                 "sstb_self_employment_income_before_lsr", known_period, array
             )
-            sstb_self_employment_income.delete_arrays(known_period)
+            delete_input_values(sstb_self_employment_income, known_period)
 
         weekly_hours = self.get_holder("weekly_hours_worked")
         for known_period in weekly_hours.get_known_periods():
             array = weekly_hours.get_array(known_period)
             self.set_input("weekly_hours_worked_before_lsr", known_period, array)
-            weekly_hours.delete_arrays(known_period)
+            delete_input_values(weekly_hours, known_period)
 
         # Capital gains responses
 
@@ -294,7 +295,7 @@ def __init__(self, *args, **kwargs):
             self.set_input(
                 "long_term_capital_gains_before_response", known_period, array
             )
-            cg_holder.delete_arrays(known_period)
+            delete_input_values(cg_holder, known_period)
 
         # Geography backfill: state_code_str-only input -> state_code enum.
         _backfill_state_code_from_str(self)
@@ -524,27 +525,27 @@ def __init__(self, *args, **kwargs):
         for known_period in employment_income.get_known_periods():
             array = employment_income.get_array(known_period)
             self.set_input("employment_income_before_lsr", known_period, array)
-            employment_income.delete_arrays(known_period)
+            delete_input_values(employment_income, known_period)
 
         self_employment_income = self.get_holder("self_employment_income")
         for known_period in self_employment_income.get_known_periods():
             array = self_employment_income.get_array(known_period)
             self.set_input("self_employment_income_before_lsr", known_period, array)
-            self_employment_income.delete_arrays(known_period)
+            delete_input_values(self_employment_income, known_period)
 
         sstb_self_employment_income = self.get_holder("sstb_self_employment_income")
         for known_period in sstb_self_employment_income.get_known_periods():
             array = sstb_self_employment_income.get_array(known_period)
             self.set_input(
                 "sstb_self_employment_income_before_lsr", known_period, array
             )
-            sstb_self_employment_income.delete_arrays(known_period)
+            delete_input_values(sstb_self_employment_income, known_period)
 
         weekly_hours = self.get_holder("weekly_hours_worked")
         for known_period in weekly_hours.get_known_periods():
             array = weekly_hours.get_array(known_period)
             self.set_input("weekly_hours_worked_before_lsr", known_period, array)
-            weekly_hours.delete_arrays(known_period)
+            delete_input_values(weekly_hours, known_period)
 
         # Capital gains responses
 
@@ -554,7 +555,7 @@ def __init__(self, *args, **kwargs):
             self.set_input(
                 "long_term_capital_gains_before_response", known_period, array
             )
-            cg_holder.delete_arrays(known_period)
+            delete_input_values(cg_holder, known_period)
 
         # Geography backfill: state_code_str-only input -> state_code enum.
         # Datasets carry state_fips, so this only fires for situations that