The crosswalk cannot say whether a row answers for a reporting area or for the bucket it folds into

[lbm364dl]

Split out of #741, which fixed the half of this that was actionable now.

The gap

polity_area_crosswalk rows are keyed on (area_code, polity_code), and
.add_polity_columns_dt() is called with two different kinds of code in that
first slot:

• a raw FAOSTAT reporting area code -- R/build_cbs.R:766, 892, 1393, 2425,
  R/build_trade.R:191, 201;
• a bucket code, i.e. polity_area_code -- .bucket_area_labels()
  (R/polity_folds.R:585), which is where every published
  reporting_polity_code is actually decided.

Nothing in the table says which of the two a row is meant to answer, so one row
has to serve both. (238, ETH-1952-1993) is the case:

• as a bucket row it is right. FAOSTAT area 62 (Ethiopia PDR) folds into
  bucket 238, so bucket 238 sums Ethiopia's 1961-1992 data, and this row is the
  bucket's only period before 1993. Deleting it sends 176 of the 352
  (area, year) pairs for areas 62/238 to ETH-1993-2025 as an out-of-span
  stand-in -- measured in The prefix expansion hands an area the polity another area owns upstream, because it only tests each area against its own map span #741, which is why that PR kept it.
• as an area row it is not. Reporting area 238 does not exist before 1993;
  upstream's map gives it 1993-2024 only.

#741 removed the mirror row (62, ETH-1993-2025) precisely because it has no
bucket reading -- nothing folds into bucket 62 -- but the asymmetric rule it
shipped is a workaround for the missing distinction, not the distinction itself.

Scope

Small and static. Only three areas fold into a bucket that is not their own code
(62 into 238, 276 and 277 into 206), and ETH is the only ISO3 prefix family the
upstream map splits across two reporting areas, so exactly one row is
currently ambiguous.

Not urgent, and why

The two readings happen to give the same answer today. Measured over all 266
area/bucket codes x 1850-2025 (46,816 pairs) and confirmed on a real
6,310,390-row get_primary_production(), every pre-1993 Ethiopia row reaches
reporting_polity_code through the bucket route, and ETH-1952-1993 is the
right answer for it. A raw area-238 row dated before 1993 -- which FAOSTAT never
emits -- would get the same code, correctly by extent but silently as a period
hit rather than as the stand-in it is. So this is labelling precision, not a
wrong number.

What a fix would look like

Give the crosswalk a column naming the key space a row serves (key_role in
{"area", "bucket"}, or two tables), have .add_polity_columns_dt() select on
it, and mark (238, ETH-1952-1993) bucket-only. That is a published schema
change with more than one defensible shape, which is why it is filed rather than
folded into #741.

Part of the polity migration epic #458.

[lbm364dl]

One more thing PR for #741 surfaced and deliberately left alone, filed here so it is not lost.

.bucket_year_polity_conflicts() resolves every member area of a bucket by year and flags the years the members disagree. It deliberately does not restrict a member to the years it actually reports -- the comment says so, because that is what makes bucket 206 ambiguous in all 65 reported years rather than only the 15 its periods overlap in.

With #741 merged that detector reports 98 rows instead of 65: bucket 238 joins it for 1993-2025, because dissolved area 62 stands in with ETH-1952-1993 while area 238 is a period hit on ETH-1993-2025. That is honest -- the row #741 removed was manufacturing the agreement -- and it cannot reach published output, since area 62 contributes no rows after 1992 and .bucket_area_labels() labels a bucket from the bucket code.

Restricting a member to its reported years would remove bucket 238's 33 rows and also shrink bucket 206's 65 to roughly the 15 overlap years, which silently redefines an already-filed finding (#414). So it belongs with the schema question above, not in a bug fix.