juaristi22 · Sep 16, 2026 · Sep 15, 2026 · Sep 15, 2026 · Sep 15, 2026 · Sep 16, 2026
diff --git a/changelog.d/929-household-composition-signature.fixed.md b/changelog.d/929-household-composition-signature.fixed.md
@@ -0,0 +1 @@
+Declare the household-type partition on the measurement of the ten `ons.household_composition.*` targets (#929). Since #906 every K02000001 row is a candidate control for any lower-grain row with the same exact measurement signature; on the pre-42254801 frame the composition cells carried the bare `uk.household.count` signature and collided with the council-tax stock totals on the rowwise local surface ("two different control values at grain 'country'"). #906's 42254801 filters the totals by band from the other side; both declarations stand, the filter mirrors the binding and the exact cover is pinned by test. The binding is unchanged; only the cross-grain grouping key moves.
diff --git a/changelog.d/929-uk-council-tax-taxbase.changed.md b/changelog.d/929-uk-council-tax-taxbase.changed.md
@@ -0,0 +1,3 @@
+Re-pin the UK Chronicle consumer feed to `df35af7` (#929, from `474a0ae`): the rows are `chronicle.consumer_fact.v3` (chronicle #267), which carries the publisher's `geography.name` on every fact, so the local target surface compiles again on the schema-8 hierarchy without a catalog entry per constituency or authority (closes #920); `build.chronicle_epoch` declares the v3 identity beside v2. The artifact also carries chronicle #264's council taxbase packages (MHCLG CTB lines by band for England's 296 billing authorities, StatsWales CT1 dwellings by band A–I for Wales, CTAXBASE chargeable dwellings for Scotland's 32 councils; 266,390 rows, +124,990). No existing reference on either surface resolves a different fact or value: the national and local registers regenerate with only the feed identity, the hierarchy catalog header and the fact-supplied geography names moving; the nine vendored resources and the local census restate the pin.
+
+Re-base the UK council tax stock family on the councils' taxbase returns (#929, closes the family's part of #736's council-tax lines). The VOA valuation-list stock counted exempt, empty and second-home dwellings as households (England +6.3 %, Wales +6.6 % over the household rows on the 10 September dense rebuild), the source of the family's uniform negative bias and of 33 of the release-blocking `uk_local_target_fit` failures. Three source-named families replace the 17 `voa.council_tax_stock*` targets on one measurement signature: `mhclg.council_tax_stock.*` binds MHCLG CTB line 7 (chargeable dwellings after the disabled-relief adjustment; band A adds the A- column) minus line 11 (second homes) minus line 15 (empty dwellings) for the 296 English billing authorities, with the nine English region controls composed from the authority rows (a consumer rollup, since MHCLG prints no region row and Chronicle computes no sums); `welshgov.council_tax_stock.*` binds StatsWales CT1 a1 − h7 − h8 by band A–I for the 22 Welsh authorities and the Wales row on the 2025-26 return; `scotgov.council_tax_stock.by_area.*` binds CTAXBASE chargeable dwellings for the 32 councils beside the existing country rows. New shared machinery: a `linear_combination` value operation (signed, n-ary, each operand a sum over its selector with `expected_member_count`, formula stamped in `ledger_value_formula`); geography fan-out cells composed from lower-grain member facts (`composed_from_level`, the hierarchy labels them from the geography catalog); per-target area rosters (`area_scope`) for publications that cover one nation; and publisher code aliases on the crosswalk (`code_aliases`: Barnsley and Sheffield, recoded E08000038/E08000039 in the April 2025 LAD layer, bind their 2025 MHCLG and PIPR rows under the roster codes). Band I joins the local metric roster (Wales only); the empty-leg licences derived from the local membership honour each target's `area_scope`, so a nation's authority family is licensed outright on the other nations' legs. National surface 603 → 613 references; local surface 20,430 → 20,885 (176 Welsh A–H, 22 Welsh I, 255 Scottish, 2 PIPR cells), the 296 English band-H cells still deferred on spine support. Retired: `council_tax_voa_scotland_absent`, `council_tax_ni_domestic_rates`, `council_tax_wales_country_control_absent` (its premise was stale: Chronicle carried the Wales row in every pin), `council_tax_city_of_london_band_a_suppressed`; the `voa_dwellings_vs_household_frame` fence is re-adjudicated onto the taxbase basis. The 55k report's "frozen vs recomputed +3 to +12 %" on the VOA rows was a comparison defect (`frozen_vs_recomputed` summed the Wales rollup into the England-pinned target); it now keys on bound rows only and refuses duplicates.
diff --git a/docs/uk-chronicle-feed-repin.md b/docs/uk-chronicle-feed-repin.md
@@ -19,8 +19,11 @@ regeneration tests accept a `CHRONICLE_UK_FACTS` override and skip only when
 neither is present.
 
 Verify both SHA-256 digests and the manifest's `facts_sha256`, row count, and
-schema version, then update `uk/chronicle_feed.json` and regenerate, in this
-order:
+schema version, then update `uk/chronicle_feed.json` and regenerate. Vendor
+the per-concern fact resources (step 5) before the national generator: the
+generator refuses to compile against a vendored resource whose feed identity
+differs from the pin, so on a fresh pin step 5 has to run first. The rest go in
+this order:
 
 1. the national references and membership with
    `tools/generate_uk_target_references.py` (pass the stable
@@ -60,4 +63,8 @@ and energy packages and unified the national and local pins into this one
 declaration; the `474a0ae` re-pin (#904, chronicle #263) moved the rows to
 `chronicle.consumer_fact.v2`, which carries the dimension and value labels the
 schema-8 target hierarchy completes from (141,400 rows, including chronicle #260's
-Universal Credit packages).
+Universal Credit packages); the `df35af7` re-pin (#929, chronicle #264 and #267)
+moved the rows to `chronicle.consumer_fact.v3`, which adds the publisher's
+`geography.name` to every fact (the label the hierarchy needs for constituencies
+and local authorities, microcosm#920) and brought the council taxbase packages
+for England, Wales and Scotland (266,390 rows).
diff --git a/experiments/355-uk-dataset-size-receipts.md b/experiments/355-uk-dataset-size-receipts.md
@@ -569,6 +569,13 @@ rows agree within 1%. This is the population-dependent-measure physics the plan
 the pool's contributions; the engine on 55k rows resolves the council-tax band imputation and the private
 school flag differently.
 
+> **Erratum (2026-09-15, microcosm#929).** The seven VOA band rows were not diverging: `frozen_vs_recomputed` keyed the
+> incumbent-surface rows by `contract_target_id` and summed the Wales regional rollup into the England-pinned target
+> (band A: recomputed − frozen = 248,027 = the Wales band-A rollup to the unit; B 284,108, C 294,555 likewise). Without
+> Wales, frozen and recomputed agree exactly. The comparison now keys on bound rows only (#929), and the same evaluation
+> on the #929 candidate carries no VOA row over 1 %: max divergence 12.4 % → 6.2 %, the two remaining rows being
+> `isc.private_school_students` and `ons.land.corporate_land_value` as before.
+
 **50 downstream (uk-candidate-eval under 2.94.0; "incumbent" = R17's dense H5 on spine-m and the old feed, so
 input differences ride along):**
 
@@ -930,3 +937,7 @@ loan balance). **#731 scorecard (2.97.0):** population Q50f 69.97 / Q50 58.43 /
 0.371 / 0.376. **T6 (uk-data venv, eFRS 1.57.3 as the comparison file):** Q50 autumn budget 807 ok / 485 flag (closer 25/37),
 RF UC 10/17 (closer 6/10), childcare 17/16 (closer 7/17), all "review"; Q50f's replays ran after 22:33Z (results in the
 evaluation page's T6 tables).
+
+> **Erratum (2026-09-15, microcosm#929).** The "+3 to +12 %" on the VOA council-tax bands is the same keying artifact
+> as above (the Wales rollup summed into the England-pinned target); corrected, no VOA row diverges and the maximum
+> divergence is 6.2 % (`isc.private_school_students`); see the erratum under the D2 frozen-versus-recomputed paragraph.
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		Declare the household-type partition on the measurement of the ten `ons.household_composition.*` targets (#929). Since #906 every K02000001 row is a candidate control for any lower-grain row with the same exact measurement signature; on the pre-42254801 frame the composition cells carried the bare `uk.household.count` signature and collided with the council-tax stock totals on the rowwise local surface ("two different control values at grain 'country'"). #906's 42254801 filters the totals by band from the other side; both declarations stand, the filter mirrors the binding and the exact cover is pinned by test. The binding is unchanged; only the cross-grain grouping key moves.
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,3 @@
		Re-pin the UK Chronicle consumer feed to `df35af7` (#929, from `474a0ae`): the rows are `chronicle.consumer_fact.v3` (chronicle #267), which carries the publisher's `geography.name` on every fact, so the local target surface compiles again on the schema-8 hierarchy without a catalog entry per constituency or authority (closes #920); `build.chronicle_epoch` declares the v3 identity beside v2. The artifact also carries chronicle #264's council taxbase packages (MHCLG CTB lines by band for England's 296 billing authorities, StatsWales CT1 dwellings by band A–I for Wales, CTAXBASE chargeable dwellings for Scotland's 32 councils; 266,390 rows, +124,990). No existing reference on either surface resolves a different fact or value: the national and local registers regenerate with only the feed identity, the hierarchy catalog header and the fact-supplied geography names moving; the nine vendored resources and the local census restate the pin.

		Re-base the UK council tax stock family on the councils' taxbase returns (#929, closes the family's part of #736's council-tax lines). The VOA valuation-list stock counted exempt, empty and second-home dwellings as households (England +6.3 %, Wales +6.6 % over the household rows on the 10 September dense rebuild), the source of the family's uniform negative bias and of 33 of the release-blocking `uk_local_target_fit` failures. Three source-named families replace the 17 `voa.council_tax_stock` targets on one measurement signature: `mhclg.council_tax_stock.` binds MHCLG CTB line 7 (chargeable dwellings after the disabled-relief adjustment; band A adds the A- column) minus line 11 (second homes) minus line 15 (empty dwellings) for the 296 English billing authorities, with the nine English region controls composed from the authority rows (a consumer rollup, since MHCLG prints no region row and Chronicle computes no sums); `welshgov.council_tax_stock.` binds StatsWales CT1 a1 − h7 − h8 by band A–I for the 22 Welsh authorities and the Wales row on the 2025-26 return; `scotgov.council_tax_stock.by_area.` binds CTAXBASE chargeable dwellings for the 32 councils beside the existing country rows. New shared machinery: a `linear_combination` value operation (signed, n-ary, each operand a sum over its selector with `expected_member_count`, formula stamped in `ledger_value_formula`); geography fan-out cells composed from lower-grain member facts (`composed_from_level`, the hierarchy labels them from the geography catalog); per-target area rosters (`area_scope`) for publications that cover one nation; and publisher code aliases on the crosswalk (`code_aliases`: Barnsley and Sheffield, recoded E08000038/E08000039 in the April 2025 LAD layer, bind their 2025 MHCLG and PIPR rows under the roster codes). Band I joins the local metric roster (Wales only); the empty-leg licences derived from the local membership honour each target's `area_scope`, so a nation's authority family is licensed outright on the other nations' legs. National surface 603 → 613 references; local surface 20,430 → 20,885 (176 Welsh A–H, 22 Welsh I, 255 Scottish, 2 PIPR cells), the 296 English band-H cells still deferred on spine support. Retired: `council_tax_voa_scotland_absent`, `council_tax_ni_domestic_rates`, `council_tax_wales_country_control_absent` (its premise was stale: Chronicle carried the Wales row in every pin), `council_tax_city_of_london_band_a_suppressed`; the `voa_dwellings_vs_household_frame` fence is re-adjudicated onto the taxbase basis. The 55k report's "frozen vs recomputed +3 to +12 %" on the VOA rows was a comparison defect (`frozen_vs_recomputed` summed the Wales rollup into the England-pinned target); it now keys on bound rows only and refuses duplicates.