[vahid-ahmadi]

Summary

Adds docs/book/validation/two-child-limit-full-rollout.md, a page on reproducing published two-child limit estimates that assume full roll-out and report absolute child poverty, such as the IFS Green Budget 2025 comment Options for reforming the two-child limit. An external replication using raw FRS 2024-25 found PolicyEngine UK's children lifted out of absolute poverty well short of the IFS figures. Most of that comes from settings that are easy to miss:

• the baseline already removes the limit from 6 April 2026, so the comparable year is 2025-26;
• current law protects children born before 6 April 2017, so full roll-out needs the UC and CTC born_before dates moved (this alone roughly doubles the headcount);
• in_poverty_ahc/_bhc are the HBAI absolute flags, now on the FYE 2025 reference line, while estimates published before March 2026 use the FYE 2011 line (the page shows a parameter override);
• children should be counted at person level with is_hbai_dependent_child;
• full UC take-up is a set_input on would_claim_uc.

With full roll-out and full take-up, cost (£3.58bn) and families gaining (778k) match the IFS (about £3.6bn, over 800,000 households), but children lifted are 25-30% short (455k against 630k for abolition; 293k against 420k for a three-child limit, AHC on the FYE 2011-based line). The page does not isolate the cause. It lists plausible contributors: the release has about 21% fewer children in 3+ child families in relative low income than HBAI FYE 2025 (filed as PolicyEngine/microcosm#1371), the snippet approximates the FYE 2011-based line, it omits the statutory exceptions, and its full take-up setting is not the IFS reported-take-up adjustment.

It complements child-poverty-tcl.md (2029-30, DWP/OBR/RF benchmarks), which #1724 is revising, and does not touch that file.

Evidence

The snippet on the page was run end to end on policyengine-uk 2.147.2 with Microcosm release microcosm-uk-2024-25-national-20261023T075039Z-a133f37a (SHA-256 verified); the results table is its output. The current-law row's affected count (467k benefit units, 1.64m children) is checked against DWP's April 2025 two-child limit statistics (453,600 households, 1.61m children).

Reproducer: n/a: documentation only, no model source or dependency change
axiom: n/a: documentation only

🤖 Generated with Claude Code

[vahid-ahmadi]

853b588 fixes the Lint failure (ruff format on the page's Python snippet) and replaces the microcosm#ISSUE placeholder with PolicyEngine/microcosm#1371, in the page and the body.

🤖 Generated with Claude Code

[juaristi22]

Reviewed at 853b588a29e69566152bd09b19ad3aea2c3228bb. One P2 finding is attached inline: the final comparison attributes the remaining poverty gap entirely to data without isolating that cause.

Independent primary-source checks support the quoted benchmark values. The actual published Python helpers also passed model/API assertions on a small synthetic H5, including weighting, full take-up and both reform comparisons. The private national totals were not reproduced; the synthetic checks do not establish the claimed causal explanation.

[juaristi22]

[P2] Qualify the attribution of the remaining poverty gap to data

The evidence does not isolate data as the cause of the remaining poverty gap. A shortfall in the marginal number of large-family children in relative low income cannot identify how many cross an absolute threshold after reform. This snippet also uses an approximate old-reference line, omits statutory exceptions and has not matched the IFS reported-take-up adjustment. Describe the family-size mismatch as a plausible contributor and retain those other explanations, or show a controlled comparison before ruling them out.

[vahid-ahmadi]

@juaristi22 agreed, the page claimed more than the evidence shows. That was our overreach. 0ae9428 rewrites the bullet:

• It now says the checks do not isolate the cause of the remaining 25-30% gap, and that attributing it to any one factor would need a controlled comparison.
• The family-size shortfall (microcosm#1371) is listed as a plausible contributor, with your caveat that a relative low income shortfall cannot show how many children cross an absolute line after the reform.
• It keeps the other explanations alongside it: the approximate FYE 2011-based line, the omitted statutory exceptions, and full take-up as would_claim_uc set to true for everyone, which is not the IFS reported-take-up adjustment.

The PR body now says the same. The push needed the PR out of the merge queue, so I dequeued it, pushed and re-queued it (now at the back of the queue).

🤖 Generated with Claude Code