UK local release build 1: let a leg of deferred cells stand without a control, and renew the savings-interest diagnostic (#1301)

[juaristi22]

The two changes the first local release build (#1301) needs on top of main 58b0f24. Build 1 runs from this PR's head, so merge it with a merge commit, never squash or rebase: the release records the build's commit, and it must stay in main's history (the runbook asks the same of the sign-off PR).

1. Let a leg of deferred cells stand without a control (d8b913d73)

#1307 merged at e119867af, one commit before its fix, 89cc6b9bb, reached the PR, so main does not have it. On main, the build's target compilation (uk.full.target_compilation, that is load_uk_full_target_inputs then compile_uk_full_target_surface) refuses the local surface:

cross-grain inconsistency 'exact:uk.household.count:eae2efd5da4f:region_over_la' has unparented lower-grain leg(s) ['E12000001', 'W92000004']

• Why. The North East and Wales £2m local-authority cells are all deferred by the 400-home floor. The small-cell rule adds deferred cells to the reconciliation frame, so that they take their share of a region control. With those two region rows signed out (485008810), their legs have no control, and the unparented-leg check stops the compile. Nothing in CI compiles the full surface on the feed.
• The fix is UK local release: publish a K=15, 150,000-household dense dataset in at most two builds (#1301) #1307's follow-up commit, unchanged; only its changelog line is merged with UK local surface before the Budget build: high-value homes OBR leg, per-vehicle fuel flag, Northern Ireland energy share and housekeeping (#1278) #1306's rewording.
  • The local surface passes the reconciliation the area rows that bind nothing (its frame-only rows).
  • A leg on which every open row is one of them, for targets licensed empty there, needs no control. It is receipted under deferred_legs_without_control instead of refused.
  • These are still refused as orphans: a bound row on such a leg, a frame-only row without the licence, and a row of a target the leg is not licensed for. The Scottish authority of an England-only target is still refused too.
• See the follow-up on #1307.

2. Renew the savings-interest diagnostic exclusion (d0f0fafec)

This is #1308's R0 (0610d6356), unchanged in the register: ons.savings_interest_income stays a measured diagnostic and moves from 2026-10-22 to 2027-01-08 (renewed by juaristi22 on 2026-10-16). It carries the one-line test update #1308's re-review asked for, since test_packaged_exclusions_load pins the entries approved on 2026-10-16.

• The re-review recommends shipping R0 alone.
• Without it, every dense sign-off would hit a hard stop on 22 October:
  • the build-2 gate clock range ends there;
  • the assembly deadline (build 2, its verification and the assembly) falls the same day;
  • the national line could not calibrate after it.
• The register change is solve-changing, so it has to be in build 1's commit for build 2 to reuse build 1's solve.
• Port the uk-data wave to policyengine-uk 2.134.4: stamp duty, Move to UC, national home values (sixth #1095 PR) #1308 drops the commit when it rebases, because the change is identical.

Checks

These ran locally at this head; CI runs the full suites.

• Tests. The cross-grain, UK ledger-target and off-surface-cell test files pass. So does every engine-free test file that reads the measure-exclusion register: country spec, sign-off draft, full-build CLI and preparation, full targets, measure simulation, national graph, rowwise national role, solve scope, support floor and UC family.
• Target compilation. On main plus the fix, against the pinned abf0921 feed, the build's target load and compile pass. deferred_legs_without_control names exactly two legs, E12000001 and W92000004, for contract:microcosm.high_value_homes.over_2m, and nothing else.
• Problem support. The bound high-value-homes local-authority cells were checked against uk.full.problem's zero-support refusal, which stopped the 16 October dense build on cells the 400-home floor now defers. The check used the public Land Registry value knots and the OA ladder. Every bound cell has a constituency whose value distribution reaches the threshold.

No figure in this PR comes from licensed microdata.

🤖 Generated with Claude Code