Translate old-vintage CD targets to current district geography

[MaxGhenis]

Problem

Populace can now compile SOI congressional-district target rows, but the current SOI CD source is an older congressional geography vintage. A real candidate support H5 built from the current CD-expanded flow has 436 raw CD geoids, but they are the old 117th-district set:

• extra old keys: CA-53, IL-18, MI-14, NY-27, OH-16, PA-18, WV-03
• missing current policyengine.py keys: CO-08, FL-28, MT-02, NC-14, OR-06, TX-37, TX-38

The incumbent policyengine.py release manifest uses the current 436-key district artifact set. Populace should not publish district artifacts or hard-gate CD targets while the support/target surface mixes vintages.

Proposed approach

Handle this in Ledger/target aging rather than in one-off Populace target code:

1. Store geography vintage metadata on regional facts, e.g. source_geography_level=congressional_district, source_geography_vintage=117th_congress, target_geography_vintage=118th_congress.
2. Add a generic geography-vintage translation step to target aging. For old CD targets, translate source CD rows to current CD rows through a block-level crosswalk using population weights initially.
3. Conserve parent totals exactly. For every source vintage target set, translated current-CD rows should sum to the same state and national totals before any period uprating.
4. Emit lineage metadata on translated targets: source record ids, crosswalk basis, source/target geography vintages, and reconciliation residuals.
5. Use current-boundary support/artifacts, likely assigned from ACS/Census current CD population or age distributions. Older SOI CD fiscal rows become translated current-vintage targets or diagnostics, not direct support keys.
6. Leave room for improved fiscal bases later, e.g. ACS income/tax proxy weights instead of pure population for fiscal targets.

Acceptance tests

• A fixture with old-vintage CA-53 plus the current 436-district support surface materializes no CA-53 target row.
• Translated targets include the current missing keys where appropriate, e.g. CO-08, FL-28, MT-02, NC-14, OR-06, TX-37, TX-38.
• State and national totals are conserved exactly after vintage translation and before period uprating.
• Target metadata records the source and target geography vintages and crosswalk basis.
• Release artifact generation fails if a calibrated H5 still contains stale or mixed CD artifact keys.

[MaxGhenis]

Reopening after validating the merged crosswalk support against the current CD-expanded Ledger facts. PR #207/#208 added the runtime mechanics and provenance guard, but the actual current-surface materialization is still short of the issue acceptance criteria.

Validation artifact: /Users/maxghenis/.codex-worktrees/populace-cd-calibration-run-20260627/out/cd-vintage-crosswalk-20260627/cd117_to_cd119_population_crosswalk.csv (SHA-256 a61553933a585d23365bef0602328ff347a0ae00f35e8499516f0eef069ba17d).

Compiler result with that crosswalk:

• all materialized CD specs are 5001900US and 119th_congress; no 5001700US rows remain
• 23,912 CD specs over 427 unique current districts
• missing target geos: 5001900US0200, 5001900US1000, 5001900US1198, 5001900US3001, 5001900US3002, 5001900US3800, 5001900US4600, 5001900US5000, 5001900US5600

Root cause: the SOI CD Ledger fact artifact has no old source rows for AK, DE, DC, MT, ND, SD, VT, or WY; Montana then becomes two missing current targets after reapportionment. We need an explicit state-total proxy / state-to-CD allocation path for source-missing at-large states before this can be treated as full current-CD parity.