chore(regions): re-sync the polities snapshot to whep-polities 2830fb7

[lbm364dl]

What was wrong

Not what #745 says, and that is the main finding.

#745 compares the shipped crosswalk's upstream_map rows (245 over 197 areas)
against eduaguilera/whep-polities@main's published map (281 over 228) and
concludes the snapshot is stale, so ~31 areas resolve through whep's
Rest-of-World bucket instead of upstream's mapping.

The map is not stale. Over the areas the crosswalk actually consumes, the
committed rows and the current upstream map agree on every
(area_code, year_start, year_end, polity_code) tuple: 245 against 245, zero
rows on either side alone. The 31 areas without an upstream_map row are the
FABIO Rest-of-World fold, which outranks the map deliberately. It is
documented at data-raw/table_mappings.R:242 ("ONE DELIBERATE EXCEPTION … 31
map-covered areas carry fabio_code == 999 … Letting the map win there would
move every Rest-of-World figure"), pinned by
test_polity_faostat_map.R ("the FABIO Rest-of-World fold still outranks the
map"), and printed by the build itself:

! 31 reporting area codes are named by the upstream FAOSTAT map but kept on the
  FABIO Rest-of-World fold.
i Areas: 5, 6, 17, 42, 47, 61, 64, 65, 69, 85, 87, 88, 125, 135, 140, 142, 153,
  154, …, 240, and 299.
i Lifting the fold is tracked separately and is out of scope here.

Those are exactly the 31 areas #745 lists. Lifting the fold is #717/#740, not a
re-sync, and it is a science decision with large value movement — the opposite
of what this PR does.

What is real is drift in the two other upstream artifacts. The snapshot was
taken at 09d7eaf; upstream main is 22 commits ahead, and the polities
database and the label alias map both moved.

Evidence it reproduced BEFORE the change

At the base commit (ee4ded6d), against upstream main at 2830fb7:

	rows	areas
pinned faostat_area_polity_map.csv	281	228
polity_area_crosswalk, mapping_source == "upstream_map"	245	197
tuple-level disagreement over the 197 consumed areas	0	0
the 31 remaining areas, all mapping_source == "fabio_row_fold", all fabio_code == 999, all polity_code == "ROW-1850-2025"		31

So the reported gap reproduces exactly, and its cause is the fold, not staleness.

The drift that is real, 09d7eaf → 2830fb7:

• faostat_area_polity_map.csv — 16 registry spans corrected over 15 areas.
  All 15 are Rest-of-World folded, so none reaches the crosswalk join.
• polities_database.gpkg — 4 new polities, 4 wrong cow_code values, polygon
  metadata for 8 polities, 6 predecessor/successor edges.
• label_alias_map.csv — 1 new alias, 3 corrected year_start bounds.

What I changed

Regenerated data-raw/table_mappings.R against a pinned upstream revision,
not the live checkout (which sits on a feature branch with a dirty tree):

whep-polities main @ 2830fb735bcfaaead5e9465324dec9ce9f9b2fba
  "Merge pull request #205 from eduaguilera/fix/aliases-before-target"

git -C ~/whep-polities cat-file -p <rev>:data/final/polities_database.gpkg   > /tmp/v.gpkg
git -C ~/whep-polities cat-file -p <rev>:data/final/faostat_area_polity_map.csv > /tmp/v_map.csv
git -C ~/whep-polities cat-file -p <rev>:data/final/label_alias_map.csv      > /tmp/v_alias.csv
WHEP_POLITIES_GPKG=/tmp/v.gpkg WHEP_POLITIES_FAOSTAT_MAP=/tmp/v_map.csv \
  WHEP_POLITIES_LABEL_ALIAS_MAP=/tmp/v_alias.csv \
  Rscript --vanilla data-raw/table_mappings.R

dataset	before	after
polities	753 rows, 715 geometries	757 rows, 725 geometries
polity_label_aliases	902 rows	903 rows
polity_area_crosswalk	595 rows	595 rows, every routing column identical
gleam_geographic_hierarchy	3 of 204 unresolved	204 of 204 resolved
items_cbs, items_prod		value-identical, git checkout --ed

data/livestock_coefs.rda had to be rebuilt too, and that was not a guess —
test_territorial_identity.R:196 failed the moment polities refreshed,
which is precisely what its "FRESHNESS BACKSTOP FOR THE GLEAM TABLE" comment
says it exists to do. Rebuilt from the same GLEAM workbook: 45 objects, one
changed, three cells in it (ATF, SGS, WLF from NA to
ATF-1800-2025 / SGS-1800-2025 / WLF-1800-2025, upstream
whep-polities#187). Its sibling assertion at :244 pinned those three as
permanently unresolvable and becomes the empty set.

Three comments that the re-sync makes factually wrong are corrected in place
(the retained-polity count, the map's two year-boundary exceptions upstream has
fixed, and the GLEAM NA note).

How I verified

• Both generators re-run from the pinned revision, every .rda diffed with
  all.equal() per dataset; the two that came back value-identical were
  reverted so the diff shows no spurious binaries.
• .area_year_polity_conflicts(): 0 rows after the re-sync (0 before too).
• fix(regions): stop the prefix expansion awarding a polity outside the area's fold #743's invariant holds: 0 crosswalk rows award a polity another area owns
  outside its fold. Its guard still passes.
• polity_mapping_provenance() is bit-identical before and after —
  46,816 (area, year) pairs, 34,065 upstream / 10,912 whep_bucket /
  1,622 whep_prefix / 217 unresolved. That is the direct measurement that
  the re-sync moves nothing from a guess to upstream_map, i.e. polity_area_crosswalk is built from a stale upstream map: 31 areas upstream now maps resolve through whep's Rest-of-World bucket instead #745's
  predicted 3.37% shift does not happen.
• Prefix fallback still exactly 351, 901-906.
• The GLEAM backstop is load-bearing, proved by reintroducing the defect:
  restore the stale livestock_coefs.rda next to the fresh polities →
  test_territorial_identity.R:196 fails; restore the rebuilt one → passes.
• Gates: air format . clean · devtools::document() · devtools::test()
  FAIL 0 | PASS 7190 · lintr::lint_package(...) no lints ·
  rcmdcheck 0 errors, 0 warnings, 0 notes (needs
  _R_CHECK_FORCE_SUGGESTS_=false locally: archive and RSQLite are not
  installed on this machine) · pkgdown: every man/*.Rd topic present in
  _pkgdown.yml.

Moves published values

It does not, and that is measured, not assumed.

A full real get_primary_production() (1850-2023, warm pins) was driven before
and after:

	before	after
rows	6,310,390	6,310,390
all 12 published columns		identical, every one
keys added / removed		0 / 0
(area, year) pairs whose polity_area_code, reporting_polity_code or reporting_polity_has_geometry moved		0
Σ value, all 8 units (tonnes, ha, heads, LU, slaughtered_heads, t_ha, t_head, t_LU)		delta 0.0 in each

all.equal() on the whole object reports one difference, in the non-published
.cb_extracts attribute. That is run-to-run nondeterminism in the pin read,
not this change: a third control build with the same data reproduces the
same attribute difference, while the published columns stay identical, and the
sorted contents hash equal across all three runs.

Why it cannot move: the crosswalk's routing columns (area_code,
polity_area_code, reporting_polity_code, polity_code,
polity_start_year, polity_end_year, mapping_source, mapping_status,
map_year_*, fabio_code, polity_type, has_geometry, …) are all
bit-identical; only cow_code (3 cells) and polygon_status (1 cell) changed,
and neither is emitted anywhere. The corrected map spans are all on ROW-folded
areas. The alias changes are on the fao1952, mitchell and unscoped sources,
while the only in-package caller of resolve_polity_label() asks for
source = "trade-sources".

The user-visible change is therefore confined to the tables themselves:
polities gains 4 rows and 10 geometries, 4 cow_code values are corrected,
and gleam_geographic_hierarchy resolves 3 territories it previously left
NA. NEWS.md records it.

Classification: mechanical. It adopts upstream's corrections verbatim and
changes no whep decision. The science decision in this area — lifting the
Rest-of-World fold — is deliberately not taken here.

What I deliberately did not do

• Did not lift the FABIO Rest-of-World fold. That is the actual cause of
  polity_area_crosswalk is built from a stale upstream map: 31 areas upstream now maps resolve through whep's Rest-of-World bucket instead #745's 31 areas, it moves every Rest-of-World figure, and it belongs to
  A promoted Rest-of-World member still carries the ROW aggregate polity: polity_type = "aggregate", continent = "World" #717/288 of 595 polity_area_crosswalk rows have no upstream authority: whep invents them by ISO3-prefix match #740 with its own before/after. Left standing, and the comment saying so
  is still accurate.
• Did not touch R/polities.R or R/polity_folds.R (owned by the A back-cast row reports mapping_status = "matched" for a polity that was not alive that year, in 41.5% of cells — and the gap diagnostic cannot see it #763
  worker) or R/build_production.R (The pre-1962 back-cast measures hectares on present-day borders: make the area half historical, as the yield half already is #761). One consequence worth handing over:
  resolve_polity_label()'s roxygen in R/polities.R still says the upstream
  database "is ahead of the polities snapshot this package ships (740 rows
  upstream against 603 here)" and that "225 of the 869 published aliases …
  point at one of 115 codes get_polity_geometries() cannot yet return a row
  for". Measured after this re-sync: 903 aliases, 757 polities, 0 aliases
  pointing at a missing code. That paragraph was already stale before this PR
  (the build has a hard guard making it 0 by construction) and needs one edit
  in a file I do not own.
• Did not re-run data-raw/harmonization_tables.R. No regions.csv or
  items_*.csv changed.
• Did not add a new test. The two assertions that matter already existed and
  both fired: the fold pin refutes polity_area_crosswalk is built from a stale upstream map: 31 areas upstream now maps resolve through whep's Rest-of-World bucket instead #745's premise on its own, and the GLEAM
  freshness backstop caught the one genuinely stale table. Adding a third
  restatement would not catch anything they miss.
• Did not close A promoted Rest-of-World member still carries the ROW aggregate polity: polity_type = "aggregate", continent = "World" #717. polity_area_crosswalk is built from a stale upstream map: 31 areas upstream now maps resolve through whep's Rest-of-World bucket instead #745 speculates it may close with this; it does not —
  A promoted Rest-of-World member still carries the ROW aggregate polity: polity_type = "aggregate", continent = "World" #717 is the fold, which is untouched.

Closes #745.
Part of the polity migration epic #458.

🤖 Generated with Claude Code