deped-geos

deped-geos combines accepted schools from Dataset with reviewed geography from Maps to publish area memberships, statistics, boundaries, and audit files. Installed applications use its supported read-only API instead of importing build internals.

deped-geos is a producer: a Polars-first build that turns school records and PSGC geography into school-derived area artifacts. It is layered on the shared deped-primitives library (reusable hierarchy, territory, boundary, PSGC, urbanicity, and legislative semantics) and keeps only what is bound to its own artifact contract. Canonical boundary artifacts remain exact; the producer also publishes a versioned current-year display tier for high-complexity derived division and legislative-district maps.

It emits three artifact hierarchies plus national and region-group reporting rows:

• Normalized PSGC stats by region, province, city, municipality, locality, and barangay.
• DepEd governance stats starting at region, with compatibility boundary GeoParquet for division and school district.
• Legislative stats starting at region, with compatibility boundary GeoParquet for legislative district.

(Strict psgc is an internal denominator/ordering construct only; it is never emitted as an artifact hierarchy.)

The producer consumes two sibling build outputs: ../deped-dataset/artifacts/db.sqlite3 for source-school tables and ../deped-maps/artifacts-current for PSGC CSV, area-boundary component, and GeoParquet artifacts. It writes Parquet tables, partitioned membership data, compatibility GeoParquet boundaries, and a manifest with source fingerprints, source-contract status, timings, and geometry policy. Boundary GeoParquet is written with PyArrow and Shapely; the geometry column is WKB bytes for consumers to decode at their own display/API boundary.

In plain terms: run just build after deped-dataset and deped-maps are fresh. The build reads those sibling outputs once and creates this repo's local artifacts-current/ folder. The notebook workbench then reads that folder read-only; it does not go back to the dataset DB or rerun the build. The only extra lookup is PSGC map geometry, which the notebook uses from bundled or recorded deped-maps artifacts when a PSGC map needs to be drawn.

The installed distribution is deped-geos; the Python package is deped_geos. Runtime code is grouped by ownership: deped_geos.artifacts, deped_geos.sources, deped_geos.products, deped_geos.producer, and the stable read-only deped_geos.consumer API used by installed applications and notebooks. See docs/architecture.md for the full role, ownership boundaries, and engine policy.

Installed applications import manifest validation, artifact resolution, and read-only downstream helpers from deped_geos.consumer. Version 0.10 removes the former deped_geos.workbench namespace completely; local and sibling notebooks use the corresponding deped_geos.consumer modules directly.

Quick start

uv sync --all-groups
cd ../deped-dataset && just build
cd ../deped-maps && just build
cd ../deped-geos
just check
just preflight
just build --output-dir artifacts-current
just status --json
just audit

The common just status command verifies publication identity and hashes. deped-geos artifact-status is the separate Geos-specific contract check for required area products and denominator warnings.

Version 0.8 is a clean command break: the former packaged deped-geos status command was removed. Use just status for the shared operating contract or deped-geos artifact-status for the Geos-only contract audit.

just build remains the single deped-geos build command. It validates sibling inputs but does not rebuild them. If the dataset DB or PSGC bundle is stale or incomplete, the build fails with instructions to refresh upstream:

cd ../deped-dataset && just build
cd ../deped-maps && just build

Compatibility boundary GeoParquet files are a required producer output when source geometry is present. The producer writes them with PyArrow and Shapely in the normal uv run deped-geos build / just build path. The default test environment installs the dev and workbench dependency groups so just check runs the full collected test suite.

Outputs

artifacts-current/ contains:

• manifest.json
• school_spine.parquet
• school_spine_current.parquet
• school_area_source_universe.parquet
• school_area_stats.parquet
• school_area_assignment_audit.parquet
• school_area_memberships/
• school_area_region_group_components.parquet
• school_area_division_coverage.parquet
• school_area_division_coverage_audit.parquet
• school_area_denominator_coverage.parquet
• school_area_insight_context.parquet
• school_area_display_subdivisions.parquet
• school_id_crosswalk.parquet
• school_area_boundaries_governance_division.parquet
• school_area_boundaries_governance_school_district.parquet
• school_area_boundaries_legislative_legis_district.parquet
• school_area_governance_context.parquet
• school_area_boundaries_*_current.parquet
• school_area_boundaries_governance_division_display_current.parquet
• school_area_boundaries_legislative_legis_district_display_current.parquet

Area-grain stats, assignment audit, denominator and division coverage, insight context, display subdivisions, and derived boundaries are keyed by school_type_scope and school_year_id. The manifest marks the latest year as current and lists school-type scopes beginning with __all_active__ (All active schools) and Public • DepEd. School spine and memberships stay single-copy school-grain artifacts with the existing school_type column. Coordinate correctness review belongs exclusively to the deped-maps review workbench; geos keeps latitude and longitude but does not mirror coordinate QA.

CSV exports are optional: run just build --export-csv when a downstream handoff needs CSV copies.

school_area_stats.parquet owns the complete group universe for every (school_type_scope, school_year_id, hierarchy, level, group_id): current-year group counts, denominators, labels, and parent ids belong there. Boundary GeoParquet files own renderable geometry only. Governance and legislative geometry comes from deped-maps area_boundary_components.csv source components, repeated only for matching scoped stats rows. Derived boundary GeoParquet is an exact dissolve. Contract 30 additionally publishes current-only display geometry for divisions and legislative districts. It is simplified as one coverage per parent region under derived_boundary_display_v1, with component, ring, bounds, and area guards; a failing partition keeps exact WKB. The manifest records the display path, fingerprint, QA receipt, and policy while the canonical path and current_path remain exact. A current stats row without a matching school_area_boundaries_*_current.parquet row is valid and means the group is known, but has no display geometry for consumers to draw. Downstream apps should select display_current_path for current maps, treat geometry as WKB, and perform GeoJSON serialization outside the producer.

school_area_governance_context.parquet is a separate ADM4 collection of canonical barangays classified upstream as no_observed_district. It is map context only: consumers render it neutrally behind observed governance areas and exclude it from metric joins, aggregation, tables, charts, and numeric legend bins.

school_area_stats.parquet also includes shared planning ratios such as students_per_school, students_per_capita, population_per_school, school_density, area_per_school, plus urban_barangay_share and barangay_urbanicity_band. These are producer-emitted artifact fields backed by deped-primitives metric and urbanicity semantics; notebooks and downstream consumers should read them rather than recomputing them at runtime.

school_area_source_universe.parquet preserves PSGC labels, parent ancestry, boundary mapping, and denominator seeds for no-school groups so boundary-first maps can render known geography without downstream inference.

Docs and notebooks

uv run zensical build
just notebooks
just check

The local Marimo notebooks inspect artifacts-current/ read-only and render maps in-cell. They do not expose a server/API contract.