rurality-by-version

Date: 2026-09-21. Status: implemented.

Decision. The history table carries the Scottish Government Urban Rural Classification of all nine published versions, 2003-2004 to 2022, for every postcode life: per version a 6-fold and an 8-fold code and a status, 27 nullable columns appended to the 162 of postcode_simd_wide_v1, which are unchanged and in the same order. The schema becomes postcode_simd_wide_v2. Each life's own grid reference is placed in that version's pinned polygons (core/rurality.py). A point in no polygon is null with status outside_polygons, with no nearest-polygon fallback. A point on an edge between two classes is null with ambiguous_polygons. A post-office box, a large user linked NO LINKP, is null in every version with status po_box, because NRS puts its grid reference at the sorting or delivery office. The two 2022 columns NRS publishes are kept exactly as supplied; the derived 2022 columns sit beside them. The main table is unchanged: SSPL allocates from output-area centroids and keeps one life per postcode, so it has no per-life point.

Reason. The directory carries one classification, the 2022 one, so a historical address was described by today's categories while its data zone and SIMD were read at the right vintage. About one life in nine has a different 6-fold class in a version before 2020 than the 2022 code it carries. Every version's polygons are still published, so one postcode file and one method give every version without archived lookups. The status is stored per row because a null can mean three different things and a count in the build report cannot say which rows. Keeping the published columns beside the derived ones leaves the validation permanently visible rather than overwriting a publisher's value.

Evidence. Gate fixed before the run at 99.5% for current small users and 99% for other cohorts over 1,000 lives: point-in-polygon on the 2022 polygons reproduces NRS's published 2022 codes on 247,770 of 247,773 lives, 100% of current lives, and all three misses are deleted lives outside every polygon, none a different class. In the built table the derived 2022 codes equal the published ones on all 216,045 placed rows; 31,725 boxes are withheld; outside_polygons is 3 to 34 per version except 767 non-box lives in 2005-2006, whose coastline is drawn tighter. The original 162 columns keep fingerprint 59369357e44e45a5, now pinned as wide_v1_columns_fingerprint; the main table keeps 9515021d99099b43. Readback places every point again from the saved grid references, and a test changes a 2005-2006 code, a 2013-2014 code and a 2022 status after writing and requires each to be caught. Plan and feasibility result: docs/plans/Rurality_By_Version_Plan.md.

Columns (27): urbanrural2003_2004_6fold, urbanrural2003_2004_8fold, urbanrural2003_2004_status, urbanrural2005_2006_6fold, urbanrural2005_2006_8fold, urbanrural2005_2006_status, urbanrural2007_2008_6fold, urbanrural2007_2008_8fold, urbanrural2007_2008_status, urbanrural2009_2010_6fold, urbanrural2009_2010_8fold, urbanrural2009_2010_status, urbanrural2011_2012_6fold, urbanrural2011_2012_8fold, urbanrural2011_2012_status, urbanrural2013_2014_6fold, urbanrural2013_2014_8fold, urbanrural2013_2014_status, urbanrural2016_6fold, urbanrural2016_8fold, urbanrural2016_status, urbanrural2020_6fold, urbanrural2020_8fold, urbanrural2020_status, urbanrural2022_6fold, urbanrural2022_8fold, urbanrural2022_status