Fix state_pension_type classifying all pensioners as BASIC (#1618)

The formula used `sp.new_state_pension.active.values_list[0]` to find
the instant the New State Pension was switched on (2016-01-01 under
current law). policyengine-core auto-extrapolates values_list into the
far future in newest→oldest order, so `[0]` was returning a 2040s entry
instead of the activation instant. That made `years_since_instant`
negative, pushing `male_age` / `female_age` down to ~51, so every
pensioner was `over_age` → classified `BASIC` and nobody received the
New State Pension.

Walks the list oldest→newest and takes the first `True` value to get
the real activation instant. Adds baseline YAML tests covering both
BASIC and NEW classification on either side of the NSP activation.

Verified against the latest enhanced FRS dataset:

- BASIC pensioners: 12.06M → 7.36M weighted
- NEW pensioners: 0.00M → 4.71M weighted
- state_pension aggregate 2025: £116.2bn → £127.5bn

Remaining gap to OBR (~£140bn) is additional State Pension coverage,
outside this PR's scope.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: MaxGhenis

changelog.d/state-pension-type-nsp-activation.fixed.md

@@ -0,0 +1 @@
+- Fix `state_pension_type` incorrectly classifying every pensioner as receiving the pre-2016 basic State Pension. The formula used `values_list[0]` to find when the New State Pension activated, but policyengine-core auto-extrapolates the parameter into the far future, so `[0]` was returning a 2040s entry instead of the 2016 activation date. Walks the list oldest-first to find the real activation instant, so post-2016 retirees are now correctly classified as `NEW`. Raises the modelled 2025 state pension aggregate from about £116bn to about £127bn.

policyengine_uk/tests/policy/baseline/gov/dwp/state_pension_type.yaml

@@ -0,0 +1,39 @@
+- name: Below State Pension age is NONE
+  period: 2025
+  input:
+    age: 60
+    is_male: true
+  output:
+    state_pension_type: NONE
+
+- name: Male reaching SPA after NSP activation (age 70 in 2025, born 1955) gets NEW
+  period: 2025
+  input:
+    age: 70
+    is_male: true
+  output:
+    state_pension_type: NEW
+
+- name: Male reaching SPA before NSP activation (age 80 in 2025, born 1945) gets BASIC
+  period: 2025
+  input:
+    age: 80
+    is_male: true
+  output:
+    state_pension_type: BASIC
+
+- name: Female reaching SPA after NSP activation (age 70 in 2025, born 1955) gets NEW
+  period: 2025
+  input:
+    age: 70
+    is_male: false
+  output:
+    state_pension_type: NEW
+
+- name: Female reaching SPA before NSP activation (age 80 in 2025, born 1945) gets BASIC
+  period: 2025
+  input:
+    age: 80
+    is_male: false
+  output:
+    state_pension_type: BASIC

policyengine_uk/variables/gov/dwp/state_pension_type.py

@@ -18,14 +18,26 @@ class state_pension_type(Variable):
     def formula(person, period, parameters):
         sp = parameters.gov.dwp.state_pension
         male = person("is_male", period)
-        last_entry = sp.new_state_pension.active.values_list[0]
         is_sp_age = person("is_SP_age", period)
-        if not last_entry:
+
+        # Find the instant the New State Pension was first switched on.
+        # values_list is ordered newest→oldest and PolicyEngine-core
+        # auto-extrapolates it into the far future, so values_list[0] is a
+        # future year's extrapolation rather than the activation date. Walk
+        # oldest→newest and take the first `True` value to get the real
+        # activation instant (e.g. 2016-01-01 under current law).
+        activation_entry = None
+        for entry in reversed(sp.new_state_pension.active.values_list):
+            if entry.value:
+                activation_entry = entry
+                break
+
+        if activation_entry is None:
             values_if_sp_age = where(
                 is_sp_age, StatePensionType.BASIC, StatePensionType.NONE
             )
         else:
-            instant = last_entry.instant_str
+            instant = activation_entry.instant_str
             years_since_instant = period.start.year - int(instant[:4])
             male_age = sp.age.male(instant) + years_since_instant
             female_age = sp.age.female(instant) + years_since_instant