nordisk/facett

Population Graph — the Sverige · Population showcase (design)

The Swedish 11-million-person population graph (the canonical knut-popsim model) rendered in facett-demo as three beautiful, linked, interactive graphs. Native AND the wasm web demo. THE showstopper graph tab.

Where it lives

• facett-demo/src/pop_model.rs — the query / LOD / colour layer over knut-popsim.
• facett-demo/src/pop_tab.rs — the Sverige Facet (the showcase tab).
• facett-demo/src/lib.rs — registers the Sverige tab in the deck + LEGEND, and drives the graph3d tab off the same knut-popsim population sample.
• Engine extension: facett-graphview NodeDecoration { scale } + l0::lower (domain-agnostic node sizing) — see "Engine extensions" below.

The data source — knut-popsim (NOT re-invented)

knut-popsim (knut/crates/knut-popsim, lib knut_popsim) is the canonical, pure-Rust generator of the synthetic Swedish population graph: 12 labels (Person + 11 register entities), deterministic LCG, no falkordb / arrow / network / fs deps (optional serde only) → it compiles for wasm32-unknown-unknown. The demo depends on it by path (../../knut/crates/knut-popsim) and calls:

• simulate(SimParams { n_persons, n_kommun, seed }) -> PopGraph — a sampled instance graph.
• schema() -> PopGraph — the 12-label meta-graph (the data model).
• LABELS, PERSON_TIE_PROB, PERSON_REL, geo::CITIES (14 kommuner w/ lon/lat).

PopGraph is the "graph-mem database that works in wasm". pop_model builds a thin query/filter/LOD layer over it — it never invents a second model.

Honest about scale (the LAW)

You cannot draw 100M nodes (≈400M with entities — OOMs; wasm has a hard memory ceiling). So:

• STATS are computed analytically at the requested scale (1M → 100M) from the per-label tie rates (PopStats::at_scale) — we report the honest 11M / 100M numbers without materialising the graph.
• Generation is clamped to GEN_CAP = 6_000 people (memory-safe, native + wasm).
• Rendering is clamped to RENDER_CAP = 1_400 nodes per instance view (the LOD budget), and at full scale the kommun aggregation (14 nodes) is shown instead of dots.
• The visible-vs-total split is surfaced on the tab AND in state_json ("rendering N of 11,000,000 — aggregated by kommun").

Scale field: a logarithmic slider 1M→100M + an 11M button. Default = the canonical 11,000,000.

The three linked views (one population, three lenses)

A 3D view (graph3d tab) renders the same knut-popsim sample as a spun 3D force cloud (depth-shaded), via facett-graph3d.

Toggles (drive the render, observable in state_json)

• 12 label toggles (Person, Address, Company, Debt, Loan, TaxRecord, KfmCase, PoliceRecord, FkBenefit, AfRecord, Vehicle, Property) — a disabled label's nodes + edges drop from every view.
• 12 relation toggles (the 11 Person→entity rels + ESCALATES_TO).
• colour-by: by-label (12-colour register palette) / by-kommun (hue wheel) / debt-heat (cool teal → hot red by per-person debt degree — the "follow the money" lens).

Beauty (the whole point)

Each view renders through the shared facett_core::render L0 SDF kernel (the same kernel the maps + graphview draw through), lowered with a glow lane: a soft RingInstance halo behind every node marker, sized to the marker. Pure CPU SDF → it lights up identically native AND in wasm with no GPU dependency and no l1-vello feature. Nodes read as luminous orbs over a dark ground. Colours from the 12-label palette / kommun wheel / debt-heat scale. Kommun orbs are population-sized (engine NodeDecoration.scale).

Engine extensions (domain-agnostic, kept in the engine)

• facett-graphview::NodeDecoration gained an optional scale: Option<f32> — a per-node marker-radius multiplier the caller drives by any importance metric (population, degree). l0::lower reads it (default 1.0). Tested: facett-graphview node_scale_decoration_enlarges_the_marker. The population specifics (what the size means) stay in facett-demo.

Observability (the LAW)

PopGraphTab::state_json() carries: n_persons, scale_min/max, generated_persons, gen_cap, render_cap, the analytic stats (total_nodes/edges + per_label), color_mode, active_labels, active_relations, and per-view {visible_nodes, total_nodes, edges, lit_px, zoom}. Plus a #[cfg(testmatrix)] functional_status emit from ui. Every view renders to a straight-RGBA Frame a test reads back (pop_graph_blob.rs) → PNG blobs at $TMPDIR/facett-demo/pop_*.png.

Tests (inject-assert, never no-crash)

• pop_model unit: stats scale to 11M without materialising; generation capped but scale honoured; full-model LOD cap + determinism; label/relation toggles drop nodes/edges; debt model is the OWES/ESCALATES story; kommun aggregates to 14; colour modes recolour; schema is the 12-label model.
• pop_tab unit: three linked non-blank views through L0+glow; scaling drives stats but caps generation; label toggle changes render + state; colour mode recolours the frame.
• tests/pop_graph_blob.rs: each view emits a non-blank, contained PNG (read back + asserted); debt-heat blob differs from by-label blob.
• facett-graphview node_scale_decoration_enlarges_the_marker (the engine ext).
• tests/mega_matrix.rs: a Sverige cell (+ a Velo L1 cell) in the coverage matrix.
• lib.rs demo_state_has_every_component_populated: the Sverige tab in the full deck.

wasm

cargo build -p facett-demo --target wasm32-unknown-unknown --no-default-features builds green — knut-popsim is wasm-clean and the L0+glow lane is CPU-only. The same GEN_CAP/RENDER_CAP caps apply in the browser. Deploys via cargo xtask deploy-pages (unchanged).

Future (→ korp)

This lives in facett-demo now; it will move to korp later (the surveillance / back- office tool). pop_model + pop_tab are self-contained over knut-popsim + facett-graphview, so the move is a copy of two modules + the dep.