Monograph V · Local architecture
Metro Hubs, Suburb Spokes and Four-Fact Grounding: The Link-Graph Anatomy of a 15,000-Page Local Leader
Eleven national service brands, 160,000 URLs, one recurring shape. The winners are not the sites with the most location pages. They are the sites whose location pages form a graph the engine can reason about.
| Corpus | 11 national home-service and legal brands, 2,100–15,400 indexed URLs each, crawled quarterly through 2025–2026 |
|---|---|
| Content Warehouse attributes | homepagePagerankNs, siteRadius, localityScore, geoInfo, navDemotion |
| Entity sources | Google Business Profile, state licensing registries, Secretary of State filings, E.164 phone records |
The national local-service site is the hardest architecture in search. It must rank for a service term in hundreds of metros and thousands of suburbs, it must do so from a single domain with a single root reservoir of PageRank, and it must do so without widening its topical radius into incoherence. Most attempts fail in a predictable way: a flat directory of 15,000 near-identical pages, each one click from a state index, each sharing the root’s equity fifteen-thousand ways, each saying nothing the engine cannot already infer from the template. The sites that lead their categories look different, and they look different in the same way.1Corpus: eleven brands in pest control, restoration, personal injury, HVAC and locksmithing. Link graphs were crawled quarterly and compared against map-pack and organic visibility by metro. Brand names are withheld under the advisory agreements described on the About page.
The hub tier
Every leader in the corpus concentrates root equity into a metro hub tier of between 60 and 220 pages. A hub is a substantial document, typically 1,400 to 2,600 words, about the service in a metropolitan area: regulations, climate or demographic factors that shape demand, pricing bands, named crews or partners, and a dense block of internal links to its suburbs. Hubs are linked directly from the homepage or from a single state-level index with low outdegree, so they sit one or two hops from homepagePagerankNs and inherit a meaningful fraction of it.
PR(hub) ≈ 0.85 · PR(home) / OBL(home) one hop, direct
PR(hub) ≈ 0.85² · PR(home) / (OBL(home) · OBL(state)) two hops, via state indexAt OBL(home) = 40 and OBL(state) = 25, the two-hop hub receives 0.072 percent of root. A flat directory page at the same depth but behind 300-link indexes receives 0.0008 percent, a 90-fold difference.
The spoke tier
Spokes are the suburb pages, and the leaders treat them as a budget rather than an inventory. A typical hub carries 25 to 60 spokes, each linked from the hub body and each linking back to the hub, to two or three neighbouring spokes, and to the relevant service page. Outdegree on a spoke stays under twenty. The pattern is a wheel: equity flows from hub to spokes and recirculates among adjacent spokes rather than leaking back to the root through a global navigation. In graph terms the metro is a dense sub-community with a single high-degree entry point, which is precisely the structure a PageRank-style walk concentrates on.2Sites that linked every spoke from every other spoke in the metro, a complete graph, underperformed the wheel. Outdegree climbed past 50, per-link transfer collapsed, and the spokes drifted toward the navDemotion profile of boilerplate-heavy pages.
Four-fact grounding
What separates a spoke that earns a map-pack slot from one that is merely indexed is entity grounding. The corpus shows a sharp threshold: spokes that carry four independently verifiable local facts outperform those with three or fewer by a margin that no amount of additional prose closes. The four facts that recur are:
- A legal operating entity for the area, with the licence or registration number as issued by the state body, stated in text rather than an image.
- A physical anchor: a street address, a named service-area polygon with its boundary roads, or a dispatch point that can be reconciled with a Business Profile.
- A local telephone number in canonical E.164 form, consistent across the spoke, the hub and the Business Profile for that metro.
- A third-party local citation: a municipal permit record, a chamber listing, a named local supplier or a local review source that mentions the entity in the same area.
These facts do two jobs. They give the local entity-resolution systems, exposed in the schema as geoInfo and the locality scores, enough anchors to reconcile the page with a real business. And they are topic-predictive phrases in the Patterson sense: licence numbers, boundary roads and supplier names co-occur with the service term in the way that genuine local documents do, which keeps the spoke’s embedding near the site centroid instead of out on the radius with the templated pages.
A spoke is not a page about a suburb. It is a page about a business operating in a suburb, with the evidence attached.
What the losers share
- Flat depth. Every location page one click from a state index with 400 outbound links. Root equity arrives as dust.
- Global navigation bloat. A footer listing every metro on every page, which turns the whole site into one high-outdegree node and erases the hub-and-spoke community structure.
- Template-only differentiation. City name, zip list, and a stock photo. The embedding of such a page is the embedding of the template, offset by a proper noun.
- Phone-number drift. Tracking numbers that differ between page, Profile and citations, defeating the one entity fact the engine can verify cheaply.
At 15,000 pages the arithmetic is unforgiving, but it is also legible. Concentrate root equity into hubs, discipline the outdegree of spokes, ground every spoke in four facts, and the site becomes a set of tight local communities on a tight topical centroid. The engine rewards the shape because the shape is what a real national operator with real local presence would naturally produce.