S2 — Coverage contract: assert expected cell count + log cell counts (C-34)

[Polichinel]

Epic: #51 · risk-register C-34 (Tier 2) · foundational — S4 builds on this

Problem

Nothing asserts how many cells the FAO pipeline should process. The coverage decision lives in views-models (config_queryset.py:REGION), the cell-set in views-datafactory, and the consequences here — but unfao.py has no counterpart check. A wrong or upstream-changed region delivers partial coverage to FAO with no signal.

Context

• _validate (unfao.py:147-172) checks only null-ness of the 9 metadata columns; never row/cell counts.
• The enricher already computes + logs the unmapped count (enrichment.py:117-125) — but there is no expected total to compare against, and nothing is asserted.
• Current REGION = "africa_me_legacy" (~13,110 cells); the land_gaul switch (~64,736) is imminent (views-models#127).

Work

• Establish an expected-cell-count source keyed by region (config value or a small region→count map) so the manager knows the contract for the configured region.
• In _read and _validate: log cell/row counts (historical + forecast) — fetched, enriched, unmapped.
• In _validate (or a new _validate_coverage): assert completeness — distinct enriched cells == expected for the region; raise loud on under/over.
• Keep the existing null-gate; this adds a count-gate alongside it.

Acceptance criteria

• The configured region's expected cell count is available to the manager (no inline magic numbers).
• _read/_validate log cell counts; expected≠actual raises ValueError with a clear message.
• Unit tests: correct count passes; under-coverage and over-coverage each raise.

Implementation notes

• Surface the enricher's unmapped accounting (return value/attribute) so _validate can assert on it, not just log.
• Keep the count source declarative so S4 can pin land_gaul = 64,736.

Files

views_postprocessing/unfao/managers/unfao.py (_read, _validate); possibly gaul_schema.py / a region-count config; tests/.