MaxGhenis · Oct 24, 2026 · May 27, 2026 · Oct 8, 2026 · Oct 12, 2026 · Oct 12, 2026
diff --git a/changelog.d/1398-child-poverty-tcl-page-corrections.changed.md b/changelog.d/1398-child-poverty-tcl-page-corrections.changed.md
@@ -0,0 +1 @@
+Corrected the two-child limit child poverty validation page (`docs/book/validation/child-poverty-tcl.md`): its snippet counted absolute low income (`in_poverty_ahc`) against DWP and Resolution Foundation benchmarks that measure relative low income. Re-run on policyengine-uk 2.148.3 with the certified Microcosm UK 2024-25 release, using relative low income and HBAI dependent children, removing the limit lifts about 548,000 children after housing costs and 460,000 before in 2029-30, at a cost of £3.21bn: above both benchmarks on headcount. The page also corrects the Resolution Foundation report's title and forecast vintage and the reason for the cost's sign.
diff --git a/docs/book/validation/child-poverty-tcl.md b/docs/book/validation/child-poverty-tcl.md
@@ -13,22 +13,38 @@ Tracks [#1398](https://github.com/PolicyEngine/policyengine-uk/issues/1398).
 
 ## Results
 
-| Source | Children lifted from poverty (2029-30) | Treasury cost (2029-30) |
+| Source | Children lifted from relative low income (2029-30) | Treasury cost (2029-30) |
 |--------|----------------------------------------:|------------------------:|
 | [DWP poverty impact assessment][gov-2cl] (children); [Budget 2025, Table 1.2][budget-2025] (cost, OBR-certified) | 450,000 (AHC); 300,000 (BHC) | £3.1 bn |
 | [Resolution Foundation *No half measures*][rf-nhm] | 480,000 (AHC) | £3.5 bn |
-| **PolicyEngine UK, Microcosm UK 2024-25 (AHC)** | **~463,000** | **£3.26 bn** |
-| **PolicyEngine UK, Microcosm UK 2024-25 (BHC)** | **~336,000** | (same) |
-| PolicyEngine UK, Enhanced FRS 2022-23 (AHC), May 2026 | ~501,000 | £3.40 bn |
-| PolicyEngine UK, Enhanced FRS 2022-23 (BHC), May 2026 | ~415,000 | (same) |
-
-The October 2026 run uses the Microcosm UK 2024-25 national release
-(SHA-256 `aa31bdf6…`) on policyengine-uk 2.123.7. On that dataset,
-PolicyEngine's after-housing-costs headcount sits between the DWP and
-Resolution Foundation figures, its before-housing-costs headcount is
-above DWP's, and its cost sits between the Budget 2025 and Resolution
-Foundation figures. The May 2026 Enhanced FRS run, kept for comparison,
-was above both on headcount.
+| **PolicyEngine UK, Microcosm UK 2024-25 (AHC)** | **~548,000** | **£3.21 bn** |
+| **PolicyEngine UK, Microcosm UK 2024-25 (BHC)** | **~460,000** | (same) |
+
+All the figures are relative low income: below 60% of the contemporary
+median. The PolicyEngine rows count HBAI dependent children
+(`is_hbai_dependent_child`), as DWP does, and were run on 23 October 2026
+on policyengine-uk 2.148.3 (main at `c65bced64`) with the certified
+Microcosm UK 2024-25 national release (`microcosm-uk-2024-25-national-20261019T083323Z-6cba5d71`, SHA-256
+`810ba980…`). On that run PolicyEngine's headcounts are above DWP's and the
+Resolution Foundation's, and its cost sits between the Budget 2025 and
+Resolution Foundation figures.
+
+On absolute low income the same run lifts ~549,000 children after housing
+costs and ~325,000 before, against DWP's 250,000 on both measures (Tables 3
+and 4). These are not like for like: PolicyEngine's absolute line is
+60% of the FYE 2025 median held constant in real terms, the reference year
+HBAI adopted in March 2026, while DWP's note predates that change and cites
+the FYE 2024 HBAI methodology, whose line is 60% of the FYE 2011 median.
+
+Earlier versions of this page reported ~463,000 (AHC) and ~336,000 (BHC)
+children and £3.26 bn on an earlier Microcosm UK release (SHA-256
+`aa31bdf6…`, policyengine-uk 2.123.7), and ~501,000, ~415,000 and £3.40 bn
+on the Enhanced FRS 2022-23 in May 2026. The snippet those versions showed
+computed only the after-housing-costs figure, using `in_poverty_ahc`
+(absolute low income) and counting children with `is_child` (everyone under
+18), so at least those figures are not comparable with the relative
+benchmarks above; the code for the before-housing-costs figures was not
+shown.
 
 ## How the estimate was produced
 
@@ -37,7 +53,7 @@ from policyengine_uk import Microsimulation
 
 # Local path to the Microcosm UK 2024-25 national release file.
 DATASET = "microcosm_uk_2024_25.h5"
-YEAR = 2029
+YEAR = 2029  # fiscal year 2029-30
 
 # Current law baseline (TCL removed from 2026-04-06 per Budget 2025)
 sim_baseline = Microsimulation(dataset=DATASET)
@@ -52,51 +68,61 @@ sim_counter = Microsimulation(
     },
 )
 
-is_child = sim_baseline.calculate("is_child", YEAR)
-in_poverty_baseline = sim_baseline.calculate(
-    "in_poverty_ahc", YEAR, map_to="person"
-).astype(float)
-in_poverty_counter = sim_counter.calculate(
-    "in_poverty_ahc", YEAR, map_to="person"
-).astype(float)
+# HBAI dependent children, as in DWP's statistics.
+is_child = sim_baseline.calculate("is_hbai_dependent_child", YEAR)
+
+for measure in ["in_relative_poverty_ahc", "in_relative_poverty_bhc"]:
+    baseline = sim_baseline.calculate(measure, YEAR, map_to="person")
+    counter = sim_counter.calculate(measure, YEAR, map_to="person")
+    lifted = ((counter.astype(float) - baseline.astype(float)) * is_child).sum()
+    print(f"{measure} children lifted: {lifted / 1e3:.0f}k")
 
-children_lifted = ((in_poverty_counter - in_poverty_baseline) * is_child).sum()
 fiscal_cost = (
     sim_counter.calculate("gov_balance", YEAR).sum()
     - sim_baseline.calculate("gov_balance", YEAR).sum()
 )
-
-print(f"AHC children lifted: {children_lifted / 1e3:.0f}k")
-print(f"Treasury cost:       £{fiscal_cost / 1e9:.2f}bn")
+print(f"Treasury cost: £{fiscal_cost / 1e9:.2f}bn")
 ```
 
 This is a *reverse* comparison: PolicyEngine's baseline already includes
 the TCL removal (the parameter flips to infinity from 2026-04-06), so the
 counterfactual *re-imposes* the limit at 2 children to compute the
 2029-30 effect. The signs cancel out: the children-lifted figure is
 positive (baseline has fewer in poverty than counterfactual) and the
-Treasury cost is positive (baseline collects less revenue than the
-counterfactual would).
+Treasury cost is positive (the baseline pays more Universal Credit, so its
+`gov_balance`, tax less spending, is lower than the counterfactual's).
 
 ## Why the estimates differ
 
-Three plausible drivers of the remaining differences, and of the move
-between the Enhanced FRS and Microcosm runs:
+Three plausible drivers of the remaining differences:
 
-1. **Nowcasting methodology**. RF and DWP/OBR use somewhat different
-   earnings, rent, and benefit-uprating paths through 2029-30. See the
+1. **Nowcasting methodology**. RF projects incomes with its own nowcast,
+   with private rents rising in line with average earnings after a
+   12-month lag and its own after-housing-costs deflator; DWP uses its
+   static Policy Simulation Model on the FRS for FYE 2022 to 2024 with its
+   caseload forecasts. See the
    [PolicyEngine vs RF comparison](../assumptions/nowcasting-comparison.md)
    and the [RF methodology detail](../assumptions/rf-nowcasting-methodology.md)
-   for the specific divergence points.
-2. **Take-up assumptions**. PolicyEngine applies the same Universal
-   Credit take-up (`would_claim_uc`, set in the data) to families gaining
-   entitlement from the removal as to existing claimants. A source that
-   assumes lower take-up among newly entitled families would show a
-   smaller headcount and cost.
-3. **Data and calibration vintage**. The DWP, OBR and RF estimates were
-   finalised against November 2025 assumptions; PolicyEngine's parameters tick
-   over with each EFO release. The Microcosm release is built from FRS
-   2024-25 rather than 2022-23, with take-up anchored on reported receipt.
+   for where PolicyEngine's nowcasting differs from RF's.
+2. **Take-up assumptions**. PolicyEngine decides Universal Credit claims
+   with flags set in the data, which do not respond to the reform: a family
+   claims if `would_claim_uc` is true, or if DWP has moved it off a closed
+   legacy benefit it reports (tax credits, Income Support or income-based
+   Jobseeker's Allowance; `legacy_benefits_closed`, which leaves out
+   families sent to Pension Credit instead) and
+   `would_claim_uc_at_legacy_closure` is true.
+   The data set `would_claim_uc` for every family that reports Universal
+   Credit and draw it at a residual rate for the rest, so newly entitled
+   families claim at that residual rate unless they reported a closed
+   legacy benefit. A
+   source that assumes lower take-up among newly entitled families would
+   show a smaller headcount and cost.
+3. **Data and calibration vintage**. The DWP and OBR estimates were
+   finalised against the OBR's November 2025 forecast, and RF's
+   (30 October 2025) against the OBR's March 2025 earnings forecast and the
+   Bank of England's August 2025 inflation forecast; PolicyEngine's
+   parameters tick over with each EFO release. The Microcosm release is
+   built from FRS 2024-25, with take-up anchored on reported receipt.
 
 ## How to keep this validation alive
 
@@ -110,8 +136,8 @@ between the Enhanced FRS and Microcosm runs:
 
 - HM Treasury, [Budget 2025][budget-2025], Table 1.2: the OBR-certified cost of removing the two-child limit (£3.1 billion in 2029-30).
 - OBR, [Economic and Fiscal Outlook November 2025][obr-efo].
-- Resolution Foundation, [No half measures: ending the two-child limit][rf-nhm].
-- DWP, [Removing the two-child limit on Universal Credit: impact on low income poverty levels in the United Kingdom][gov-2cl], Tables 1 and 2 (financial year ending 2030).
+- Resolution Foundation, [No half measures: setting child poverty on a downward course at the Autumn Budget][rf-nhm] (30 October 2025).
+- DWP, [Removing the two-child limit on Universal Credit: impact on low income poverty levels in the United Kingdom][gov-2cl], Tables 1 and 2 (relative low income) and 3 and 4 (absolute low income), financial year ending 2030.
 - Issue: [#1398](https://github.com/PolicyEngine/policyengine-uk/issues/1398). Related: [#1171](https://github.com/PolicyEngine/policyengine-uk/issues/1171), [#1388](https://github.com/PolicyEngine/policyengine-uk/issues/1388).
 
 [obr-efo]: https://obr.uk/efo/economic-and-fiscal-outlook-november-2025/
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		Corrected the two-child limit child poverty validation page (`docs/book/validation/child-poverty-tcl.md`): its snippet counted absolute low income (`in_poverty_ahc`) against DWP and Resolution Foundation benchmarks that measure relative low income. Re-run on policyengine-uk 2.148.3 with the certified Microcosm UK 2024-25 release, using relative low income and HBAI dependent children, removing the limit lifts about 548,000 children after housing costs and 460,000 before in 2029-30, at a cost of £3.21bn: above both benchmarks on headcount. The page also corrects the Resolution Foundation report's title and forecast vintage and the reason for the cost's sign.