[vahid-ahmadi]

Fixes #1126

What changes

take_up_contract.json changes the maximum_extended_childcare_hours_usage entry. The entry was frozen by adjudication; it is now sourced from DfE:

	Before	After
mean	15.019	31.141
sd	4.972	4.972 (unchanged; no source identifies it)
clip	[0, 30]	[0, 30]
status	frozen_by_adjudication (#834)	sourced_administrative_derived
clipped mean	15.02 h/week	28.54 h/week

The derivation and both DfE sources are recorded in the entry. #834 froze the draw "pending evidence on hours actually used". That evidence:

• DfE, 2026 to 2027 early years national funding formulae: technical note: 341,988.09 additional-hours PTEs for 3- and 4-year-olds, from the January 2025 censuses. One PTE is 15 hours a week for 38 weeks.
• DfE, Education provision: children under 5, 2025: 379,000 3- and 4-year-olds registered for the working parent entitlement in January 2025.
• 15 + 15 × 341,988.09 / 379,000 = 28.535 weekly hours. With the sd and bounds held, E[clip(N(μ, 4.972), 0, 30)] = 28.535 gives μ = 31.141.

The engine variable caps a family's total weekly extended hours, so the anchor is the total (universal plus additional). Only the 3-4 cohort is used as the anchor, because DfE's under-2 and 2-year-old PTEs are partly forecast.

Effect: offline check, no rebuild

On the certified national release (f9d1922c, policyengine-uk 2.102.3), I inverted the release's own draws through N(15.019, 4.972) to recover the identity-keyed uniforms, mapped them through N(31.141, 4.972), clipped to [0, 30] and set the result as input. A rebuild with this contract would produce the same draws. Results for 2026-27 (£bn):

	Before	After	DfE
Free early-years entitlements, total	5.81	9.30	9.9
Under 2	1.44	2.58	3.0
2-year-olds	1.50	2.58	2.7
3- and 4-year-olds	2.87	4.14	4.2
3-4 additional-hours PTEs	~0	359k	342.0k

The remaining £0.6bn sits on the policyengine-uk side and is out of scope here. It is made up of hourly rates 1-6% below DfE's 2026-27 averages (+£0.21bn) and under-1s (whole-year ages), offset by 4-year-olds in reception who are still funded. The universal-hours switch-off, policyengine-uk#1930, adds +£0.11bn on the old draw and has essentially no effect on the new one.

Tests

• New test_extended_hours_usage_reproduces_dfe_additional_hours_ptes (engine-free UK). It checks the entry is sourced and that its clipped mean reproduces the DfE-implied 28.535 hours to within 0.01.
• uv run pytest on tests/engine_free/uk and tests/engine_free/shared, plus ruff check, run locally. The stochastic-tool tests use their own fake contract, so they are unaffected.
• The contract digest is not pinned in any committed file.

Not in this PR (needs a decision or a build)

• Refit the take-up rates. tools/fit_uk_childcare_takeup.py fits the four childcare rates using the hours draw, and its extended-rate row is bilinear with the hours through the engine's exclusions. It needs the engine and the spine, so it was not re-run here. The landed rates should be refit, or the decision to keep them recorded, before the next UK build.
• No release is rebuilt or published. The next UK national build picks up the new draw.
• Review requested on whether a sourced derivation can replace the Port post-freeze UK childcare/TFC spine inputs and calibration contract #834 freeze. The freeze recorded a reviewed decision. If lifting it needs an adjudication record in experiments/ or the tracking issue, I will add one.

🤖 Generated with Claude Code

[juaristi22]

Superseded by #1299 (merged 16 October), which carries this PR's commit with your authorship as 88e355305 (the same change as b91ef34f8), with the take-up fitter and the refit on top. Closing without merging. Thanks @vahid-ahmadi.