UK local release builds: deferred-cell legs, the savings-interest renewal and the export surface's high-value support columns (#1301)

changelog.d/1301-uk-export-surface-high-value-support.fixed.md

@@ -0,0 +1 @@
+The UK export-surface gate's allow-list carries the high-value homes support split's three household columns (`high_value_support_copies`, `high_value_support_copy_index` and `household_is_high_value_support_copy`, microcosm#1240), beside the CGT support split's precedent, in `gates.json` and `UK_ALLOWED_EXTRA_EXPORT_COLUMNS` alike. They are the stage's declared coverage-manifest outputs. Without them every UK build since microcosm#1270 fails `uk_export_surface` at the calibrated gates, and the first dense release build (microcosm#1301) did; that gate has no gate-only register, so the fix changes the solve's key and the next build re-solves. The full gate battery's policy, gates-manifest and spec digests and the release-cut part's digests are re-pinned; the spine, calibration-seam and dense digests do not move.

changelog.d/1301-uk-local-solve-registers.changed.md

@@ -4,7 +4,7 @@ The registers that change the UK solve are in place before the first local relea
 
 **High-value homes.** The family's region rows control it on the local line, so:
 - `tools/generate_uk_local_target_references.py` defers its constituency level whole (`high_value_homes_constituency_grain_deferred`, `defer_if_compiles`), after the English band-H precedent, with a roster guard of 575 England and Wales constituencies;
-- a new signed small-cell rule, `hvcts_la_household_floor` in `uk/local_small_cell_rule.json` (prefix `microcosm.high_value_homes.`, floor 400 homes, about two household weights at 150,000 households, expires 2027-04-01), defers the local-authority cells under the floor and keeps 192, 101 and 57 cells over £1m, £1.5m and £2m, whose deferred homes still count in their region rows. The two region rows below the floor, `microcosm.high_value_homes.over_2m_by_region` for the North East (`E12000001`) and Wales (`W92000004`), are signed out of the region fan-out by two row exclusions in `uk/target_reference_signed_exclusions.json` (María's ruling of 17 October 2026): their published stock is about one average household weight each at the ruled 150,000 households (an expectation from the stock, not a count of frame records), so their fit would be noise, and a public gate-only sign-off would narrow an estimate expected to rest on fewer than 10 records. A row-level signed exclusion can now name a geography cell (`row: {"dimension": "geography_id", ...}`). The same areas' £1m and £1.5m rows, which count those homes, stay bound;
+- a new signed small-cell rule, `hvcts_la_household_floor` in `uk/local_small_cell_rule.json` (prefix `microcosm.high_value_homes.`, floor 400 homes, about two household weights at 150,000 households, expires 2027-04-01), defers the local-authority cells under the floor and keeps 192, 101 and 57 cells over £1m, £1.5m and £2m, whose deferred homes still count in their region rows. The two region rows below the floor, `microcosm.high_value_homes.over_2m_by_region` for the North East (`E12000001`) and Wales (`W92000004`), are signed out of the region fan-out by two row exclusions in `uk/target_reference_signed_exclusions.json` (María's ruling of 17 October 2026): their published stock is about one average household weight each at the ruled 150,000 households (an expectation from the stock, not a count of frame records), so their fit would be noise, and a public gate-only sign-off would narrow an estimate expected to rest on fewer than 10 records. A row-level signed exclusion can now name a geography cell (`row: {"dimension": "geography_id", ...}`). The same areas' £1m and £1.5m rows, which count those homes, stay bound. Their £2m authority cells are all deferred, so the cross-grain reconciliation now lets a leg that holds only the small-cell rule's deferred cells, of targets licensed empty there, stand without a control, receipted under `deferred_legs_without_control`; a bound row on such a leg is still refused as an orphan;
 - the `high_value_home_stock_derived_estimate` adjudication records where the family binds, and that the seat rank correlations, now out of sample, decide its renewal.
 
 **Islands.** Orkney (`local_authority/S12000023`) and Shetland (`local_authority/S12000027`) are pre-signed gate-only exclusions of the area-support floor in `uk/local_area_support_exclusions.json`, scoped to the local line and expiring 2027-01-16. Their cells stay bound.

changelog.d/866-savings-interest-diagnostic-renewal.changed.md

@@ -0,0 +1 @@
+The `ons.savings_interest_income` measure exclusion is renewed to 2027-01-08 (microcosm#866; renewed by juaristi22 on 2026-10-16): the ONS D.41 interest row stays a measured diagnostic and is never fitted, as ruling R3 of 2026-09-22 set, because its concept differs from the model's cash interest. Without the renewal it lapsed on 2026-10-22, and no national calibration could run after that date.

packages/microcosm-build/src/microcosm/build/cross_grain.py

@@ -284,13 +284,19 @@ def _reconcile_cross_grain_surface_with_control_receipts(
     *,
     licensed_empty_legs: Mapping[str, frozenset[str]] | None = None,
     check_factor_bounds: bool = True,
+    frame_only_positions: frozenset[int] = frozenset(),
 ) -> tuple[
     pd.DataFrame,
     list[dict[str, Any]],
     list[dict[str, Any]],
     list[dict[str, Any]],
 ]:
-    """Return a copied surface with every detected lower grain rescaled."""
+    """Return a copied surface with every detected lower grain rescaled.
+
+    ``frame_only_positions`` are rows that join the frame to take a share of
+    a control and bind nothing (a small-cell rule's deferred cells): a leg
+    holding only those, for targets licensed empty on it, needs no control.
+    """
 
     columns = _surface_columns(local_frame)
     _validate_rule(rule)
@@ -313,6 +319,7 @@ def _reconcile_cross_grain_surface_with_control_receipts(
     unparented_rows: dict[tuple[str, str], dict[int, str]] = {}
     first_pair_id: dict[tuple[str, str], str] = {}
     absent_middle_tier_legs: list[dict[str, Any]] = []
+    deferred_legs_without_control: list[dict[str, Any]] = []
     pairs_per_key: dict[tuple[str, str], int] = {}
     for group in materialized_groups:
         key = (group.inconsistency_id.rsplit(":", 1)[0], group.lower_grain)
@@ -386,12 +393,24 @@ def _reconcile_cross_grain_surface_with_control_receipts(
         if last_pair_for_key and group.lower_grain not in control_grains:
             # No farther control can still claim these leaf rows: refuse now,
             # ahead of the empty-leg licence check, as the standing rule does.
+            # A leg only licensed-empty targets hold carries deferred cells
+            # alone and needs no control (_wholly_deferred_legs).
+            open_rows = {
+                position: leg
+                for position, leg in pending.items()
+                if position not in already and leg not in assigned_controls
+            }
             unparented = sorted(
-                {
-                    leg
-                    for position, leg in pending.items()
-                    if position not in already and leg not in assigned_controls
-                }
+                set(open_rows.values())
+                - set(
+                    _wholly_deferred_legs(
+                        open_rows,
+                        reconciled,
+                        columns,
+                        licensed_empty_legs,
+                        frame_only_positions,
+                    )
+                )
             )
             if unparented:
                 raise ValueError(
@@ -649,27 +668,78 @@ def _reconcile_cross_grain_surface_with_control_receipts(
             # A control-tier row with no control above it is the top of its
             # own leg, not an orphan.
             continue
-        still = {
-            leg
+        open_rows = {
+            position: leg
             for position, leg in pending.items()
             if position not in claimed[(identity, lower_grain)]
         }
+        deferred = _wholly_deferred_legs(
+            open_rows, reconciled, columns, licensed_empty_legs, frame_only_positions
+        )
+        still = set(open_rows.values()) - set(deferred)
         if still:
             raise ValueError(
                 f"cross-grain inconsistency "
                 f"{first_pair_id[(identity, lower_grain)]!r} has "
                 f"unparented lower-grain leg(s) {sorted(still)}."
             )
+        deferred_legs_without_control.extend(
+            {
+                "inconsistency_id": first_pair_id[(identity, lower_grain)],
+                "lower_grain": lower_grain,
+                "leg": leg,
+                "lower_target_ids": list(target_ids),
+            }
+            for leg, target_ids in sorted(deferred.items())
+        )
     return (
         reconciled,
         receipts,
         empty_legs_licensed,
         controls_without_lower_rows,
         delegated_legs,
         absent_middle_tier_legs,
+        deferred_legs_without_control,
     )
 
 
+def _wholly_deferred_legs(
+    rows: Mapping[int, str],
+    frame: pd.DataFrame,
+    columns: Mapping[str, str],
+    licensed_empty_legs: Mapping[str, frozenset[str]] | None,
+    frame_only_positions: frozenset[int],
+) -> dict[str, tuple[str, ...]]:
+    """The legs of ``rows`` that hold deferred cells alone.
+
+    A leg qualifies when every row on it is frame-only (it binds nothing and
+    joins the frame only to take its share of a control) and every target on
+    it is licensed empty there, so no cell of it binds on the leg. With no
+    control above them such rows have nothing to reconcile, and the leg is
+    no orphan (microcosm#1301: a region row signed out below the small-cell
+    floor leaves its wholly deferred authorities without one). A bound row,
+    or a row of a target the leg is not licensed for, keeps it an orphan.
+    """
+
+    targets_by_leg: dict[str, set[str]] = {}
+    bound_legs: set[str] = set()
+    for position, leg in rows.items():
+        targets_by_leg.setdefault(leg, set()).add(
+            str(frame.iloc[position][columns["target_id"]])
+        )
+        if position not in frame_only_positions:
+            bound_legs.add(leg)
+    return {
+        leg: tuple(sorted(target_ids))
+        for leg, target_ids in targets_by_leg.items()
+        if leg not in bound_legs
+        and all(
+            leg in _licensed_legs_for_target(target_id, licensed_empty_legs)
+            for target_id in target_ids
+        )
+    }
+
+
 def reconcile_cross_grain_surface(
     local_frame: pd.DataFrame,
     groups: Iterable[CrossGrainInconsistency],
@@ -699,12 +769,16 @@ def apply_cross_grain_reconciliation(
     reviewed_unbound_higher_targets: Mapping[str, Mapping[str, object]] | None = None,
     licensed_empty_legs: Mapping[str, frozenset[str]] | None = None,
     check_factor_bounds: bool = True,
+    frame_only_positions: Iterable[int] = (),
 ) -> tuple[pd.DataFrame, dict[str, Any]]:
     """Detect, reconcile, and return the always-present pass receipt.
 
     ``check_factor_bounds=False`` records each bounded bridge's factor without
     refusing it, for a pass that is not the caller's last (a surface whose
     off-surface cells join a second pass is bounded on that one).
+    ``frame_only_positions`` names the rows that bind nothing (deferred
+    cells); a leg holding only those needs no control
+    (``deferred_legs_without_control``).
     """
 
     bound = tuple(str(target_id) for target_id in bound_higher_targets)
@@ -728,12 +802,14 @@ def apply_cross_grain_reconciliation(
         controls_without_lower_rows,
         delegated_legs,
         absent_middle_tier_legs,
+        deferred_legs_without_control,
     ) = _reconcile_cross_grain_surface_with_control_receipts(
         local_frame,
         groups,
         rule,
         licensed_empty_legs=licensed_empty_legs,
         check_factor_bounds=check_factor_bounds,
+        frame_only_positions=frozenset(int(p) for p in frame_only_positions),
     )
     receipt = {
         "bound_higher_targets": list(bound),
@@ -745,6 +821,7 @@ def apply_cross_grain_reconciliation(
         "controls_without_lower_rows": controls_without_lower_rows,
         "delegated_legs": delegated_legs,
         "absent_middle_tier_legs": absent_middle_tier_legs,
+        "deferred_legs_without_control": deferred_legs_without_control,
         "absence": (
             None
             if groups

packages/microcosm-build/src/microcosm/build/uk/calibration_measure_exclusions.json

@@ -285,9 +285,9 @@
       "reason": "Concept, not a weight question: the ONS UKEA HAXV row is households' (S.14) D.41 interest resources in the national accounts (FISIM-adjusted, including the unincorporated-business part of the sector; GBP 95.5bn at 2025) while the model's savings_interest_income is cash interest, so the row is unreachable by design (-82 percent on the v20 national candidate). From 2026-09-22 the fitting anchor for interest is the SPI 2023-24 Table 3.7 interest rows by total-income band, uprated to the calibration year (hmrc.spi.savings_interest_income.*, PolicyEngine/chronicle#280 lane); this row is measured on every evaluation as the D.41 diagnostic and never fitted. Retire only with a declared D.41-to-cash bridge.",
       "tracking": "microcosm#866",
       "approved_by": "juaristi22",
-      "adjudication": "PolicyEngine/microcosm#1006 (issue comment 5812206367, the register rulings María gave in session: ruling R3 of 2026-09-22, the SPI interest rows bound and the ONS row a measured diagnostic; income-anchor lane, PolicyEngine/chronicle#280); previously microcosm#762 A16 (2026-09-03).",
-      "approved_on": "2026-09-22",
-      "expires_on": "2026-10-22"
+      "adjudication": "PolicyEngine/microcosm#1006 (issue comment 5812206367, the register rulings María gave in session: ruling R3 of 2026-09-22, the SPI interest rows bound and the ONS row a measured diagnostic; income-anchor lane, PolicyEngine/chronicle#280); previously microcosm#762 A16 (2026-09-03); renewed on 2026-10-16 to 2027-01-08 by juaristi22 in session (folded into the sixth microcosm#1095 ports PR): the gap is one of concept, not of weights, so the row stays a measured diagnostic until a D.41-to-cash bridge is declared.",
+      "approved_on": "2026-10-16",
+      "expires_on": "2027-01-08"
     },
     {
       "name": "obr.housing_benefit",

packages/microcosm-build/src/microcosm/build/uk/gates.json

@@ -953,6 +953,9 @@
           "household.cash_isa",
           "household.cgt_support_copies",
           "household.cgt_support_copy_index",
+          "household.high_value_support_copies",
+          "household.high_value_support_copy_index",
+          "household.household_is_high_value_support_copy",
           "household.household_is_cgt_residential_clone",
           "household.cgt_residential_clone_index",
           "household.household_clone_index",

packages/microcosm-build/src/microcosm/build/uk_runtime/ledger_targets.py

@@ -1997,13 +1997,17 @@ def apply_uk_cross_grain_reconciliation(
     extra_bridges: Sequence[CrossGrainBridge] = (),
     extra_signatures: Mapping[str, Mapping[str, Any]] | None = None,
     check_factor_bounds: bool = True,
+    frame_only_positions: Iterable[int] = (),
 ) -> tuple[pd.DataFrame, dict[str, Any]]:
     """Apply the standing UK rule to a bound mixed-grain target surface.
 
     ``extra_bridges`` and ``extra_signatures`` carry the summed fan-out
     controls the joint surface adds for its declared fan-out sum bridges.
     ``check_factor_bounds=False`` records the bridges' factor bounds without
     refusing, for a pass the caller reruns with more rows.
+    ``frame_only_positions`` names the rows that bind nothing (the small-cell
+    rule's deferred cells), so a wholly deferred leg with no control above it
+    is receipted rather than refused (microcosm#1301).
 
     Increment #762 may extend the grains in the UK rowwise solve only through
     this front door, so detection, reconciliation, and the manifest receipt
@@ -2040,6 +2044,7 @@ def apply_uk_cross_grain_reconciliation(
         reviewed_unbound_higher_targets=reviewed_unbound_higher_targets,
         licensed_empty_legs=licences,
         check_factor_bounds=check_factor_bounds,
+        frame_only_positions=frame_only_positions,
     )
 
 
@@ -2741,6 +2746,14 @@ def reconcile(
             columns=["grain", "geography_id", "target_id", "value", "_output_position"],
         )
         raw = frame[["grain", "geography_id", "target_id", "value"]]
+        # The area rows that bind nothing (deferred and off-surface cells): a
+        # leg holding only deferred ones needs no control (microcosm#1301).
+        frame_only = [
+            position
+            for position, row in enumerate(reconciliation_rows)
+            if row["_output_position"] is None
+            and row["grain"] in level_to_area_type.values()
+        ]
         result, pass_receipt = apply_uk_cross_grain_reconciliation(
             raw,
             bound_control_ids,
@@ -2750,6 +2763,7 @@ def reconcile(
             extra_bridges=fanout_bridges,
             extra_signatures=fanout_signatures,
             check_factor_bounds=check_factor_bounds,
+            frame_only_positions=frame_only,
         )
         return frame, raw, result, pass_receipt
 

packages/microcosm-build/src/microcosm/build/uk_runtime/terminal_gates.py

@@ -182,6 +182,13 @@ def __post_init__(self) -> None:
     # #1045: the CGT support split's family copy count.
     "household.cgt_support_copies",
     "household.cgt_support_copy_index",
+    # microcosm#1240: the high-value homes support split's family copy count,
+    # index and flag, the CGT split's precedent. The stage's coverage-manifest
+    # outputs; the first dense release build failed this gate on them
+    # (microcosm#1301).
+    "household.high_value_support_copies",
+    "household.high_value_support_copy_index",
+    "household.household_is_high_value_support_copy",
     # #1063: the residential split's arm flag and index.
     "household.household_is_cgt_residential_clone",
     "household.cgt_residential_clone_index",